Method and apparatus for tuning a neural network

Abstract

This application provides a method and apparatus for adjusting a neural network. The method involves a first neural network and a second neural network, and includes: firstly determining whether the currently used neural network needs updating; if an update is needed, the first device can perform the update automatically or instruct the second device to perform the update. In other words, a more suitable neural network can be obtained by updating the neural network in the AI ​​model, thereby improving the accuracy of information transmission based on the AI ​​model.

Description

Technical Field

[0001] The present application relates to the field of communications, and more particularly, to a method and apparatus for adjusting a neural network. Background Art

[0002] In massive multiple-input multiple-output (Massive MIMO) technology, network equipment can use precoding to reduce interference between multiple terminal devices and between multiple signal streams from the same terminal device, thereby improving signal quality, enabling spatial division multiplexing, and increasing spectrum utilization. A terminal device can determine a precoding matrix based on downlink channel measurements, and through feedback, the network device can obtain a precoding matrix that is the same as or similar to the precoding matrix determined by the terminal device.

[0003] Currently, a method for information transmission based on artificial intelligence (AI) is known, in which a terminal device and a network device jointly train to obtain a first neural network and a second neural network, wherein the first neural network is set on the terminal device side and the second neural network is set on the network device side. The communication system including the first neural network and the second neural network is called an AI framework. Specifically, the AI-based information transmission method includes: the terminal device obtains information to be fed back, and the information to be fed back is processed by at least the first neural network to obtain information that needs to be fed back via the air interface. The terminal device then feeds back the information that needs to be fed back via the air interface to the network device via a feedback link. The network device receives the information fed back by the terminal device and processes it at least by the second neural network to recover the information to be fed back on the terminal device side. However, with this AI-based information transmission method, the first and second neural networks obtained through offline training are directly applied during the online information transmission process. If the terminal device is displaced, the matching degree between the communication environment and the first and second neural networks may be reduced, thereby affecting the performance of information transmission based on the trained first and second neural networks. Summary of the Invention

[0004] The present application provides a method and apparatus for adjusting a neural network in order to improve the performance of information transmission.

[0005] In a first aspect, a method for adjusting a neural network is provided. The method for adjusting a neural network can be executed by a first device, or can be executed by a chip or circuit provided in the first device. This application does not limit this. For ease of description, the following description is made using the execution by the first device as an example.

[0006] The method for adjusting a neural network involves a first neural network and a second neural network, wherein the first neural network is applied to a first device side and the second neural network is applied to a second device side. The method includes: the first device determines a third neural network and a fourth neural network, wherein the third neural network and the fourth neural network correspond to the first neural network and the second neural network respectively; the first device sends information of the fourth neural network to the second device, wherein the first neural network or the third neural network is used for the first device to perform first signal processing, and the second neural network or the fourth neural network is used for the second device to perform second signal processing, and the second signal processing corresponds to the first signal processing.

[0007] According to the method for adjusting a neural network provided in an embodiment of the present application, a first device can determine a new neural network and send information about the new neural network to a second device, so that the first device and the second device can transmit information through the new neural network, thereby improving the performance of information transmission.

[0008] In conjunction with the first aspect, in certain implementations of the first aspect, using the first neural network or the third neural network to process the first signal on the first device includes: using the first neural network to process the first signal on the first device to obtain a second signal; and using the third neural network to process the third signal on the first device to obtain a fourth signal.

[0009] Optionally, the second neural network or the fourth neural network is used for the second device to process the second signal, comprising: the second neural network is used for the second device to process the second signal to obtain a fifth signal; the fourth neural network is used for the second device to process the fourth signal to obtain a sixth signal; or

[0010] The second neural network or the fourth neural network is used by the second device to process the second signal, including: the second neural network is used by the second device to process the first signal to obtain the second signal; the fourth neural network is used by the second device to process the third signal to obtain the fourth signal;

[0011] Optionally, the first neural network or the third neural network is used by the first device to perform first signal processing, including: the first neural network is used by the first device to process the second signal to obtain the fifth signal; the third neural network is used by the first device to process the fourth signal to obtain the sixth signal.

[0012] In the method for adjusting a neural network provided in an embodiment of the present application, the first neural network and the second neural network can be regarded as a pair of neural networks, and the third neural network and the fourth neural network can be regarded as a pair of neural networks. The paired neural networks can perform corresponding signal processing.

[0013] In combination with the first aspect, in certain implementations of the first aspect, the difference between the first training signal related to the third signal and the second training signal related to the sixth signal satisfies the first condition; or, the similarity between the first training signal related to the third signal and the second training signal related to the sixth signal satisfies the second condition.

[0014] When the third neural network and the fourth neural network obtained by the above update are used for signal processing, the first training signal related to the third signal and the second training signal related to the sixth signal meet a preset condition.

[0015] In combination with the first aspect, in some implementations of the first aspect, the difference includes any one of the following: a difference value, a mean square error value, a normalized mean square error value, or a mean absolute error; and the similarity includes a correlation coefficient.

[0016] In the method for adjusting a neural network provided in an embodiment of the present application, the above-mentioned difference or similarity can be specifically expressed in a variety of forms, thereby increasing the flexibility of the solution.

[0017] In conjunction with the first aspect, in certain implementations of the first aspect, the first neural network or the second neural network is used to process the first signal to obtain the second signal, including: the first neural network or the second neural network and the first functional module are used to process the first signal to obtain the second signal; and / or the first neural network or the second neural network is used to process the second signal to obtain the fifth signal, including: the first neural network or the second neural network and the second functional module are used to process the second signal to obtain the fifth signal;

[0018] The third neural network or the fourth neural network is used to process the third signal to obtain the fourth signal, including: the third neural network or the fourth neural network and the third functional module are used to process the third signal to obtain the fourth signal; and / or, the third neural network or the fourth neural network is used to process the fourth signal to obtain the sixth signal, including: the third neural network or the fourth neural network and the fourth functional module are used to process the fourth signal to obtain the sixth signal.

[0019] The method for adjusting a neural network provided in an embodiment of the present application may further include other signal processing functional modules during the signal processing of the paired neural networks, without limiting the specific composition of the communication system, thereby improving the flexibility of the solution.

[0020] In combination with the first aspect, in some implementations of the first aspect, the first device determining the third neural network and the fourth neural network includes: the first device updating the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively.

[0021] According to the method for adjusting a neural network provided in an embodiment of the present application, a first device may update an original neural network in order to obtain a new four-neural network.

[0022] In combination with the first aspect, in certain implementations of the first aspect, before the first device updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively, the method further includes: the first device receives first indication information from the second device, the first indication information is used to instruct the first device to update the first neural network and the second neural network; or, the first device periodically or non-periodically determines whether the first neural network and the second neural network need to be updated, and the first device determines that the first neural network and the second neural network need to be updated; or, the first device receives second indication information from the second device, the second indication information is used to instruct the first device to determine whether the first neural network and the second neural network need to be updated, and the first device determines that the first neural network and the second neural network need to be updated; or, the first device determines whether the first neural network and the second neural network need to be updated based on a preset condition trigger, and the first device determines that the first neural network and the second neural network need to be updated.

[0023] The method for adjusting a neural network provided in an embodiment of the present application can have multiple conditions for triggering the first device to execute an update, thereby improving the flexibility of the solution.

[0024] In combination with the first aspect, in some implementations of the first aspect, the first device updating the first neural network and the second neural network includes: the first device periodically or aperiodically updating the first neural network and the second neural network.

[0025] In the method for adjusting a neural network provided in an embodiment of the present application, the first device may update the neural network periodically or non-periodically, and the present application does not impose any limitation thereto, thereby increasing the flexibility of the solution.

[0026] In combination with the first aspect, in certain implementations of the first aspect, the first device determines whether the first neural network and the second neural network need to be updated, including: the first device determines whether the first neural network and the second neural network need to be updated based on whether the difference or similarity between the first information and the second information satisfies a third condition, wherein the first information includes a first judgment signal related to the first signal, and the second information includes a second judgment signal related to the fifth signal.

[0027] In the method for adjusting a neural network provided in an embodiment of the present application, the first device determines whether to update the neural network based on the relationship between information processed by the neural network and a preset threshold, which can improve the accuracy of determining whether to update.

[0028] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the first device determines the second judgment signal based on the first judgment signal, the first neural network and the second neural network; or, the first device determines the second judgment signal based on the first judgment signal, the first neural network and the second neural network and the first functional module and / or the second functional module.

[0029] The method for adjusting a neural network provided in an embodiment of the present application can simulate the information of the first device side and the information of the other end, so that the first device side can determine whether to update the neural network based on the information processed by the neural network.

[0030] In combination with the first aspect, in certain implementations of the first aspect, the first judgment signal includes downlink channel information received by the first device from the second device, or the first judgment signal includes channel information measured by the first device based on the sounding reference signal SRS received from the second device; wherein, the downlink channel information is carried on a preset resource or the downlink channel information is indicated for determining the first judgment signal.

[0031] The method for adjusting a neural network provided in an embodiment of the present application can determine, by a first device side, the first signal for determining whether to update the neural network through information sent by a second device side, and provides different types of information, thereby improving the flexibility of the solution.

[0032] In combination with the first aspect, in certain implementations of the first aspect, the first device determines the third neural network and the fourth neural network, including: the first device selects the third neural network and the fourth neural network from a first set based on communication system data, where the first set includes multiple neural networks; or, the first device adjusts the network weights corresponding to the first neural network and the second neural network respectively based on the communication system data to obtain the third neural network and the fourth neural network; or, the first device selects the fifth neural network and the sixth neural network from a second set based on the communication system data, and then adjusts the network weights corresponding to the fifth neural network and the sixth neural network respectively based on the communication system data to obtain the third neural network and the fourth neural network.

[0033] In the method for adjusting a neural network provided in an embodiment of the present application, the first device can determine the third neural network and the fourth neural network in different ways, thereby improving the flexibility of the solution.

[0034] In combination with the first aspect, in some implementations of the first aspect, the communication system data includes at least one of channel characteristics, signal-to-noise ratio, downlink channel information, uplink channel information, and hybrid automatic repeat request HARQ information.

[0035] In the method for adjusting a neural network provided in an embodiment of the present application, the first device determines that the communication system data based on the third neural network and the fourth neural network may be in multiple forms, which can improve the flexibility of the solution.

[0036] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the first device sending information of the third neural network to the second device.

[0037] In the method for adjusting a neural network provided in an embodiment of the present application, in addition to sending information about a fourth neural network that can be applied to the second device side to the first device, the first device can also send information about a third neural network that can be applied to the first device side to the second device, so that the second device can subsequently determine whether to update the third neural network and the fourth neural network.

[0038] In conjunction with the first aspect, in certain implementations of the first aspect, the information of the third neural network includes at least one of the following information:

[0039] The identifier of the third neural network, information about the network structure corresponding to the third neural network and information about the network weights corresponding to the third neural network, the identifier of the network structure corresponding to the third neural network and information about the network weights corresponding to the third neural network, the identifier of the network structure corresponding to the third neural network and the identifier of the network weights corresponding to the third neural network, information about the network weights corresponding to the third neural network, a change between the network weights corresponding to the third neural network and the network corresponding to the first neural network, the identifier of the network weights corresponding to the third neural network, and third indication information, wherein the third indication information is used to indicate whether the network structure corresponding to the third neural network is the same as the network structure corresponding to the first neural network;

[0040] The information of the fourth neural network includes at least one of the following information:

[0041] The identifier of the fourth neural network, information on the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identifier of the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identifier of the network structure corresponding to the fourth neural network and the identifier of the network weights corresponding to the fourth neural network, information on the network weights corresponding to the fourth neural network, the amount of change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network, the identifier of the network weights corresponding to the fourth neural network, and fourth indication information, wherein the fourth indication information is used to indicate whether the network structure corresponding to the fourth neural network and the network structure corresponding to the second neural network are the same.

[0042] In the method for adjusting a neural network provided in the embodiments of the present application, the specific forms of the information of the neural network can be various, thereby increasing the flexibility of the solution.

[0043] In combination with the first aspect, in certain implementations of the first aspect, the change between the network weights corresponding to the third neural network and the network weights corresponding to the first neural network, and / or the change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network includes one or more of the following information: weight position information, weight interpretation rule indication information, weight value arrangement information, and weight value change arrangement information; the information on the network structure corresponding to the third neural network and / or the information on the network structure corresponding to the fourth neural network includes: indication information on the neural network type and indication information on the neural network structure; the information on the network weights corresponding to the third neural network and / or the information on the network weights corresponding to the fourth neural network includes: weight interpretation rule indication information and / or weight value arrangement information.

[0044] The above-mentioned changes between the network weights, the information of the network structure corresponding to the neural network, and the information of the network weights corresponding to the neural network can be indicated by different information.

[0045] In combination with the first aspect, in certain implementations of the first aspect, before the first device sends the information of the fourth neural network to the second device, or sends the information of the fourth neural network and the information of the third neural network, the method further includes: the first device sends first information to the second device, the first information being used to request the second device to receive the information of the fourth neural network, or being used to request the second device to receive the information of the fourth neural network and the information of the third neural network; the first device receives a first response message from the second device, the first response message being used to instruct the first device to send the information of the fourth neural network, or being used to instruct the first device to send the information of the fourth neural network and the information of the third neural network;

[0046] and / or,

[0047] The first device receives second information from the second device, where the second information is used to indicate that the second device can receive information about the fourth neural network, or to indicate that the second device can receive information about the fourth neural network and information about the third neural network;

[0048] The first device sending the information of the fourth neural network to the second device, or sending the information of the fourth neural network and the information of the third neural network includes:

[0049] After receiving the second information, the first device sends the information of the fourth neural network to the second device, or sends the information of the fourth neural network and the information of the third neural network;

[0050] Furthermore, after the first device receives the second information, it starts to send the information of the fourth neural network to the second device within a first preset time period after the first starting moment or after a second preset time period, or sends the information of the fourth neural network and the information of the third neural network.

[0051] and / or,

[0052] After the first device updates the first neural network and the second neural network to obtain a third neural network and a fourth neural network respectively, the first device sends information about the fourth neural network to the second device, or sends information about the fourth neural network and information about the third neural network;

[0053] Furthermore, after the first device updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively, it starts sending information about the fourth neural network to the second device within a second preset time period after the second starting moment or after the second preset time period, or sends information about the fourth neural network and information about the third neural network;

[0054] and / or,

[0055] The first device sends first information to the second device, where the first information is used to request the second device to receive information about the fourth neural network, or to request the second device to receive information about the fourth neural network and information about the third neural network;

[0056] The first device receives a first response message from the second device, where the first response message is used to instruct the first device to send information about the fourth neural network, or to instruct the first device to send information about the fourth neural network and information about the third neural network.

[0057] In the method for adjusting a neural network provided in an embodiment of the present application, a first device sends information of the fourth neural network mentioned above to a second device, or sends information of the fourth neural network and information of the third neural network. The sending may be done after negotiation between the first device and the second device, or the first device decides to send it on its own, or the second device instructs the first device to send it, etc. In the embodiment of the present application, multiple possibilities are provided for the premise that the first device sends information of the fourth neural network mentioned above to the second device, or sends information of the fourth neural network and information of the third neural network, thereby improving the flexibility of the solution.

[0058] In combination with the first aspect, in certain implementations of the first aspect, after the first device sends the information of the fourth neural network to the second device, or sends the information of the fourth neural network and the information of the third neural network, the method also includes: the first device receives a first notification message from the second device, and the first notification message is used to indicate whether the second device adopts the fourth neural network, or whether to adopt the fourth neural network and the third neural network for information transmission.

[0059] In the method for adjusting a neural network provided in an embodiment of the present application, the second device can decide whether to use the received new neural network, thereby avoiding limiting the use of the new neural network.

[0060] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the first device sends a second notification message to the second device, or the first device receives a second notification message from the second device, wherein the second notification message is used to notify that the information transmission mode is changed from the first mode to the second mode, the first mode includes information transmission based on an AI model, the AI ​​model includes the first neural network and the second neural network, and the second mode includes information transmission based on a non-AI model.

[0061] The method for adjusting a neural network provided in an embodiment of the present application can switch different transmission modes through a second notification message, thereby avoiding information transmission based on an inappropriate neural network.

[0062] In combination with the first aspect, in certain implementations of the first aspect, the AI ​​model also includes a seventh neural network and an eighth neural network.

[0063] The method for adjusting the neural network provided in the embodiment of the present application can be understood as that when the local device and the opposite device in the communication system decide to fall back to the non-AI mode for information transmission, the different information processing processes included in the communication system all fall back to the non-AI mode.

[0064] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the first device switching the information transmission mode from the first mode to the second mode;

[0065] Furthermore, the first device starts switching the information transmission mode from the first mode to the second mode within a fourth preset time period after the fourth starting moment or after the fourth preset time period;

[0066] The fourth starting time includes the time when the first device sends the second notification message to the second device, or the time when the first device receives the second notification message from the second device.

[0067] In the method for adjusting a neural network provided in an embodiment of the present application, the time for executing the switch can be determined based on the time for sending and receiving the second notification message.

[0068] In combination with the first aspect, in certain implementations of the first aspect, the first starting moment and the second starting moment include: the moment when the first device receives the first indication information from the second device; or, the moment when the first device sends the second notification message to the second device; or, the moment when the first device receives the second notification message from the second device; or, the moment when the first device sends the first signaling to the second device; or, the moment when the first device receives the second signaling from the second device, wherein the first signaling is used to indicate the start of timing and the second signaling is used to indicate the start of updating; wherein the second notification message is used to notify that the information transmission mode is changed from the first mode to the second mode.

[0069] In the method for adjusting a neural network provided in an embodiment of the present application, the above-defined starting time can be determined based on the time when different messages are received or sent, thereby improving the flexibility of the solution.

[0070] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the first device receives a second response message from the second device, the second response message being used to indicate a refusal to receive information about the fourth neural network, or being used to indicate a refusal to receive information about the fourth neural network and information about the third neural network.

[0071] In the method for adjusting a neural network provided in an embodiment of the present application, the second device can refuse to receive information about the new neural network, thereby providing initiative to the receiving device of the neural network.

[0072] In combination with the first aspect, in certain implementations of the first aspect, the first device includes a terminal device or a network device, and the second device includes the terminal device or the network device. When the first device is the terminal device, the second device is the network device; when the first device is the network device, the second device is the terminal device.

[0073] The first device and the second device mentioned above may be a terminal device and a network device respectively.

[0074] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the first device receiving fifth indication information from the second device, the fifth indication information being used to indicate the first device's ability to send the first device; the first device sending first capability information to the second device, the first capability information including at least one of the following information: computing capability information of the first device, storage capability information of the first device, or information on whether the first device stores a first AI model, a neural network structure supported by the first device, a neural network scale supported by the first device, and an AI communication function supported by the first device, wherein the first AI model is any one AI model; or, the method further includes: the first device sending sixth indication information to the second device, the sixth indication information being used to indicate the second device's ability to send the second device; the first device receiving second capability information from the second device, the second capability information including at least one of the following information: computing capability information of the second device, storage capability information of the second device, or information on whether the second device stores a second AI model, a neural network structure supported by the second device, a neural network scale supported by the second device, and an AI communication function supported by the second device, wherein the second AI model is any one AI model.

[0075] In the method for adjusting a neural network provided in an embodiment of the present application, a first device and a second device may learn about each other's capabilities through signaling interaction.

[0076] In combination with the first aspect, in some implementations of the first aspect, the first signal includes channel information; the first neural network includes a first compressor, the second neural network includes a first decompressor, the third neural network includes a second compressor, and the fourth neural network includes a second decompressor; or, the first neural network includes a first decompressor, the second neural network includes a first compressor, the third neural network includes a second decompressor, and the fourth neural network includes a second compressor.

[0077] The method for adjusting a neural network provided in an embodiment of the present application, in the scenario of channel information transmission, the above-mentioned neural network can be a compressor or a decompressor.

[0078] In combination with the first aspect, in some implementations of the first aspect, the channel information includes channel information received by multiple receiving antennas or channel information of multiple layers.

[0079] The method for adjusting a neural network provided in an embodiment of the present application, in the scenario of channel information transmission, jointly encodes the channel information of multiple receiving antennas / multiple layers into one codeword, thereby realizing joint compression of the channel information, improving the channel information compression rate, and reducing the feedback overhead of the channel information.

[0080] In combination with the first aspect, in certain implementations of the first aspect, one or more of the above-mentioned first indication information, second indication information, first information, first response message, second information, first notification message, second notification message, and second response message are carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink authorization UL grant or the random access channel RACH.

[0081] The messages involved in the embodiments of the present application can be transmitted in different ways to improve the flexibility of the solution.

[0082] In a second aspect, a method for adjusting a neural network is provided. The method for adjusting a neural network can be executed by a second device, or can be executed by a chip or circuit provided in the second device. This application does not limit this. For the sake of ease of description, it can be illustrated by taking the execution by the second device as an example.

[0083] The method for adjusting a neural network involves a first neural network and a second neural network, wherein the first neural network is applied to a first device side and the second neural network is applied to a second device side. The method includes: the second device receiving information from a fourth neural network of the first device, wherein the fourth neural network is a neural network corresponding to the second neural network;

[0084] The second device determines the fourth neural network based on the information of the fourth neural network.

[0085] According to the method for adjusting a neural network provided in an embodiment of the present application, a first device can determine a new neural network and send information about the new neural network to a second device, so that the first device and the second device can transmit information through the new neural network, thereby improving the performance of information transmission.

[0086] In conjunction with the second aspect, in certain implementations of the second aspect, the method further includes:

[0087] The second device receives information from a third neural network of the first device, where the third neural network is a neural network corresponding to the first neural network; wherein the first neural network or the third neural network is used for the first device to perform first signal processing, and the second neural network or the fourth neural network is used for the second device to perform second signal processing, where the second signal processing corresponds to the first signal processing.

[0088] In the method for adjusting a neural network provided in an embodiment of the present application, in addition to sending information about a fourth neural network that can be applied to the second device side to the first device, the first device can also send information about a third neural network that can be applied to the first device side to the second device, so that the second device can subsequently determine whether to update the third neural network and the fourth neural network.

[0089] In combination with the second aspect, in certain implementations of the second aspect, the first neural network or the third neural network is used for the first device to perform first signal processing, including: the first neural network is used for the first device to process the first signal to obtain a second signal; the third neural network is used for the first device to process the third signal to obtain a fourth signal; the second neural network or the fourth neural network is used for the second device to perform second signal processing, including: the second neural network is used for the second device to process the second signal to obtain a fifth signal; the fourth neural network is used for the second device to process the fourth signal to obtain a sixth signal; or, the second neural network or the fourth neural network is used for the second device to perform second signal processing, including: the second neural network is used for the second device to process the first signal to obtain the second signal; the fourth neural network is used for the second device to process the third signal to obtain the fourth signal; the first neural network or the third neural network is used for the first device to perform first signal processing, including: the first neural network is used for the first device to process the second signal to obtain the fifth signal; the third neural network is used for the first device to process the fourth signal to obtain the sixth signal.

[0090] In the method for adjusting a neural network provided in an embodiment of the present application, the first neural network and the second neural network can be regarded as a pair of neural networks, and the third neural network and the fourth neural network can be regarded as a pair of neural networks. The paired neural networks can perform corresponding signal processing.

[0091] In combination with the second aspect, in certain implementations of the second aspect, the difference between the first training signal related to the third signal and the second training signal related to the sixth signal satisfies the first condition; or, the similarity between the first training signal related to the third signal and the second training signal related to the sixth signal satisfies the second condition.

[0092] When the third neural network and the fourth neural network obtained by the above update are used for signal processing, the first training signal related to the third signal and the second training signal related to the sixth signal meet a preset condition.

[0093] In combination with the second aspect, in some implementations of the second aspect, the difference includes any one of the following: difference, mean square error, normalized mean square error, or mean absolute error; and the similarity includes a correlation coefficient.

[0094] In the method for adjusting a neural network provided in an embodiment of the present application, the above-mentioned difference or similarity can be specifically expressed in a variety of forms, thereby increasing the flexibility of the solution.

[0095] In conjunction with the second aspect, in certain implementations of the second aspect, the first neural network or the second neural network is used to process the first signal to obtain the second signal, including: the first neural network or the second neural network and the first functional module are used to process the first signal to obtain the second signal; and / or the first neural network or the second neural network is used to process the second signal to obtain the fifth signal, including: the first neural network or the second neural network and the second functional module are used to process the second signal to obtain the fifth signal;

[0096] The third neural network or the fourth neural network is used to process the third signal to obtain the fourth signal, including: the third neural network or the fourth neural network and the third functional module are used to process the third signal to obtain the fourth signal; and / or, the third neural network or the fourth neural network is used to process the fourth signal to obtain the sixth signal, including: the third neural network or the fourth neural network and the fourth functional module are used to process the fourth signal to obtain the sixth signal.

[0097] The method for adjusting a neural network provided in an embodiment of the present application may further include other signal processing functional modules during the signal processing of the paired neural networks, without limiting the specific composition of the communication system, thereby improving the flexibility of the solution.

[0098] In combination with the second aspect, in certain implementations of the second aspect, before the second device receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the third neural network from the first device, the method further includes: the second device determines that the first neural network and the second neural network need to be updated; the second device sends first indication information to the first device, and the first indication information is used to instruct the first device to update the first neural network and the second neural network; or, the method further includes: the second device sends second indication information to the first device, and the second indication information is used to instruct the first device to determine whether the first neural network and the second neural network need to be updated.

[0099] In the method for adjusting a neural network provided in an embodiment of the present application, the second device can instruct the first device to update or determine whether to update the neural network.

[0100] In combination with the second aspect, in certain implementations of the second aspect, the second device determines whether the first neural network and the second neural network need to be updated, including: the second device periodically or non-periodically determining whether the first neural network and the second neural network need to be updated; or, the second device receives seventh indication information from the first device, the seventh indication information being used to instruct the second device to determine whether the first neural network and the second neural network need to be updated; or, the second device determines whether the first neural network and the second neural network need to be updated based on a preset condition.

[0101] In the method for adjusting a neural network provided in an embodiment of the present application, the second device can determine and update the neural network in a variety of ways, which are not limited in this application, thereby improving the flexibility of the solution.

[0102] In conjunction with the second aspect, in certain implementations of the second aspect, the second device determining whether the first neural network and the second neural network need to be updated includes:

[0103] The second device determines whether the first neural network and the second neural network need to be updated based on whether the difference or similarity between the first information and the second information meets a third condition, wherein the first information includes a first judgment signal related to the first signal, and the second information includes a second judgment signal related to the fifth signal.

[0104] In the method for adjusting a neural network provided in an embodiment of the present application, the second device determines whether to update the neural network based on the relationship between information processed by the neural network and a preset threshold, which can improve the accuracy of determining whether to update.

[0105] In combination with the second aspect, in some implementations of the second aspect, the method further includes: the second device determines the second judgment signal based on the first judgment signal, the first neural network and the second neural network; or, the second device determines the second judgment signal based on the first judgment signal, the first neural network and the second neural network and the first functional module and / or the second functional module.

[0106] The method for adjusting a neural network provided in an embodiment of the present application can simulate the second device side to obtain information of the local end and the opposite end, so that the second device side can determine whether to update the neural network based on the information processed by the neural network.

[0107] In combination with the second aspect, in certain implementations of the second aspect, the first judgment signal includes downlink channel information sent by the first device to the second device, or the first judgment signal includes channel information measured based on the sounding reference signal SRS sent by the first device; wherein, the downlink channel information is carried on a preset resource or the downlink channel information is indicated for determining the first judgment signal.

[0108] The method for adjusting a neural network provided in an embodiment of the present application can determine, by the second device side, the first signal for determining whether to update the neural network through information sent by the first device side, and provides different types of information, thereby improving the flexibility of the solution.

[0109] In conjunction with the second aspect, in certain implementations of the second aspect, the information of the third neural network includes at least one of the following information:

[0110] The identifier of the third neural network, information about the network structure corresponding to the third neural network and information about the network weights corresponding to the third neural network, the identifier of the network structure corresponding to the third neural network and information about the network weights corresponding to the third neural network, the identifier of the network structure corresponding to the third neural network and the identifier of the network weights corresponding to the third neural network, information about the network weights corresponding to the third neural network, a change between the network weights corresponding to the third neural network and the network corresponding to the first neural network, the identifier of the network weights corresponding to the third neural network, and third indication information, wherein the third indication information is used to indicate whether the network structure corresponding to the third neural network is the same as the network structure corresponding to the first neural network;

[0111] The information of the fourth neural network includes at least one of the following information:

[0112] The identifier of the fourth neural network, information on the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identifier of the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identifier of the network structure corresponding to the fourth neural network and the identifier of the network weights corresponding to the fourth neural network, information on the network weights corresponding to the fourth neural network, the amount of change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network, the identifier of the network weights corresponding to the fourth neural network, and fourth indication information, wherein the fourth indication information is used to indicate whether the network structure corresponding to the fourth neural network and the network structure corresponding to the second neural network are the same.

[0113] In the method for adjusting a neural network provided in the embodiments of the present application, the specific forms of the information of the neural network can be various, thereby increasing the flexibility of the solution.

[0114] In combination with the second aspect, in certain implementations of the second aspect, the change between the network weights corresponding to the third neural network and the network weights corresponding to the first neural network, and / or the change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network includes one or more of the following information: weight position information, weight interpretation rule indication information, weight value arrangement information, and weight value change arrangement information; the information on the network structure corresponding to the third neural network and / or the information on the network structure corresponding to the fourth neural network includes: indication information on the neural network type and indication information on the neural network structure; the information on the network weights corresponding to the third neural network and / or the information on the network weights corresponding to the fourth neural network includes: weight interpretation rule indication information and / or weight value arrangement information.

[0115] The above-mentioned changes between the network weights, the information of the network structure corresponding to the neural network, and the information of the network weights corresponding to the neural network can be indicated by different information.

[0116] In conjunction with the second aspect, in certain implementations of the second aspect, before the second device receives the information of the fourth neural network from the first device, or the information of the fourth neural network and the information of the third neural network, the method further includes:

[0117] The second device receives first information from the first device, where the first information is used to notify the second device to receive information about the fourth neural network, or to notify the second device to receive information about the fourth neural network and information about the third neural network;

[0118] The second device sends a first response message to the first device, where the first response message is used to instruct the first device to send the information of the fourth neural network, or to instruct the first device to send the information of the fourth neural network and the information of the third neural network;

[0119] and / or,

[0120] The second device sends second information to the first device, where the second information is used to indicate that the second device can receive information about the fourth neural network, or information about the fourth neural network and information about the third neural network;

[0121] The second device receives information of the fourth neural network from the first device, or the information of the fourth neural network and the information of the third neural network include:

[0122] After sending the second information, the second device starts to receive information about the fourth neural network from the first device, or receives information about the fourth neural network and information about the third neural network from the first device;

[0123] Furthermore, after the second device sends the second information, it starts to receive information about the fourth neural network from the first device within a first preset time period after the first starting moment or after the first preset time period, or receives information about the fourth neural network and information about the third neural network from the first device;

[0124] and / or,

[0125] The second device receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network from the first device;

[0126] Furthermore, the second device begins to receive information about the fourth neural network from the first device within a second preset time period after the second starting moment or after the second preset time period, or receives information about the fourth neural network and information about the third neural network from the first device;

[0127] and / or,

[0128] The second device receives first information from the first device, where the first information is used to notify the second device to receive information about the fourth neural network, or information about the fourth neural network and information about the third neural network;

[0129] The second device receives information of the fourth neural network from the first device, or the information of the fourth neural network and the information of the third neural network include:

[0130] The second device starts receiving information about the fourth neural network from the first device within a third preset time period after receiving the first information from the first device or after the third preset time period, or receives information about the fourth neural network and the third neural network from the first device.

[0131] In the method for adjusting a neural network provided in an embodiment of the present application, a first device sends information of the fourth neural network mentioned above to a second device, or sends information of the fourth neural network and information of the third neural network. The sending may be done after negotiation between the first device and the second device, or the first device decides to send it on its own, or the second device instructs the first device to send it, etc. In the embodiment of the present application, multiple possibilities are provided for the premise that the first device sends information of the fourth neural network mentioned above to the second device, or sends information of the fourth neural network and information of the third neural network, thereby improving the flexibility of the solution.

[0132] In combination with the second aspect, in certain implementations of the second aspect, after the second device receives information about the fourth neural network from the first device, or receives information about the fourth neural network and information about the third neural network, the method further includes: the second device sends a first notification message to the first device, and the first notification message is used to indicate whether the second device adopts the fourth neural network, or whether to adopt the fourth neural network and the third neural network for information transmission.

[0133] In the method for adjusting a neural network provided in an embodiment of the present application, the second device can decide whether to use the received new neural network, thereby avoiding limiting the use of the new neural network.

[0134] In combination with the second aspect, in certain implementations of the second aspect, the method also includes: the second device sends a second notification message to the first device, or the second device receives a second notification message from the first device, wherein the second notification message is used to notify that the information transmission mode is changed from the first mode to the second mode, the first mode includes information transmission based on an AI model, the AI ​​model includes the first neural network and the second neural network, and the second mode includes information transmission based on a non-AI model.

[0135] The method for adjusting a neural network provided in an embodiment of the present application can switch different transmission modes through a second notification message, thereby avoiding information transmission based on an inappropriate neural network.

[0136] In combination with the second aspect, in certain implementations of the second aspect, the AI ​​model also includes a seventh neural network and an eighth neural network.

[0137] The method for adjusting the neural network provided in the embodiment of the present application can be understood as that when the local device and the opposite device in the communication system decide to fall back to the non-AI mode for information transmission, the different information processing processes included in the communication system all fall back to the non-AI mode.

[0138] In conjunction with the second aspect, in certain implementations of the second aspect, the method further includes:

[0139] The second device switches the information transmission mode from the first mode to the second mode,

[0140] Furthermore, the second device starts switching the information transmission mode from the first mode to the second mode within a fourth preset time period after the fourth starting moment or after the fourth preset time period.

[0141] The fourth starting moment includes the moment when the second device sends the second notification message to the first device, or the moment when the second device receives the second notification message from the first device.

[0142] In the method for adjusting a neural network provided in an embodiment of the present application, the time for executing the switch can be determined based on the time for sending and receiving the second notification message.

[0143] In combination with the second aspect, in some implementations of the second aspect, the first starting moment and the second starting moment include: the moment when the second device sends the first indication information to the first device; or, the moment when the second device receives the second notification message sent from the first device; or, the moment when the second device sends the second notification message to the first device; or, the moment when the second device receives the first signaling from the first device; or, the moment when the second device sends the second signaling to the first device, wherein the first signaling is used to indicate the start of timing and the second signaling is used to indicate the start of updating; wherein the second notification message is used to notify that the information transmission mode is changed from the first mode to the second mode.

[0144] In the method for adjusting a neural network provided in an embodiment of the present application, the above-defined starting time can be determined based on the time when different messages are received or sent, thereby improving the flexibility of the solution.

[0145] In conjunction with the second aspect, in certain implementations of the second aspect, the method further includes:

[0146] The second device sends a second response message to the first device, where the second response message is used to indicate a refusal to receive the information of the fourth neural network, or to indicate a refusal to receive the information of the fourth neural network and the information of the third neural network.

[0147] In the method for adjusting a neural network provided in an embodiment of the present application, the second device can refuse to receive information about the new neural network, thereby providing initiative to the receiving device of the neural network.

[0148] In combination with the second aspect, in certain implementations of the second aspect, the first device includes a terminal device or a network device, and the second device includes the terminal device or the network device. When the first device is the terminal device, the second device is the network device; when the first device is the network device, the second device is the terminal device.

[0149] The first device and the second device mentioned above may be a terminal device and a network device respectively.

[0150] In combination with the second aspect, in certain implementations of the second aspect, the method further includes: the second device sends fifth indication information to the first device, the fifth indication information is used to indicate the first device's ability to send the first device; the second device receives first capability information from the first device, the first capability information includes at least one of the following information: computing capability information of the first device, storage capability information of the first device or information on whether the first device stores a first AI model, a neural network structure supported by the first device, a neural network scale supported by the first device, and an AI communication function supported by the first device, wherein the first AI model is any one AI model; or,

[0151] The method further includes:

[0152] The second device receives sixth indication information from the first device, where the sixth indication information is used to indicate the second device's ability to send the second device; the second device sends second capability information to the first device, where the second capability information includes at least one of the following information: computing capability information of the second device, storage capability information of the second device or information on whether the second device stores a second AI model, a neural network structure supported by the second device, a neural network scale supported by the second device, and an AI communication function supported by the second device, wherein the second AI model is any one AI model.

[0153] In the method for adjusting a neural network provided in an embodiment of the present application, a first device and a second device may learn about each other's capabilities through signaling interaction.

[0154] In combination with the second aspect, in some implementations of the second aspect, the first signal includes channel information; the first neural network includes a first compressor, the second neural network includes a first decompressor, the third neural network includes a second compressor, and the fourth neural network includes a second decompressor; or, the first neural network includes a first decompressor, the second neural network includes a first compressor, the third neural network includes a second decompressor, and the fourth neural network includes a second compressor.

[0155] The method for adjusting a neural network provided in an embodiment of the present application, in the scenario of channel information transmission, the above-mentioned neural network can be a compressor or a decompressor.

[0156] In combination with the second aspect, in some implementations of the second aspect, the channel information includes channel information received by multiple receiving antennas or channel information of multiple layers.

[0157] The method for adjusting a neural network provided in an embodiment of the present application, in the scenario of channel information transmission, jointly encodes the channel information of multiple receiving antennas / multiple layers into one codeword, thereby realizing joint compression of the channel information, improving the channel information compression rate, and reducing the feedback overhead of the channel information.

[0158] In combination with the second aspect, in certain implementations of the second aspect, one or more of the above-mentioned first indication information, second indication information, first information, first response message, second information, first notification message, second notification message, and second response message are carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink authorization UL grant or the random access channel RACH.

[0159] The messages involved in the embodiments of the present application can be transmitted in different ways to improve the flexibility of the solution.

[0160] On the third aspect, a communication method is provided. The communication method can be executed by a first device, or can also be executed by a chip or circuit provided in the first device. This application does not limit this. For the sake of convenience of description, the following is explained using the execution by the first device as an example.

[0161] The communication method includes:

[0162] The first device receives tenth indication information from the second device, where the tenth indication information is used to indicate the AI ​​capability of the second device.

[0163] Optionally, the first device sends information of the AI ​​model to the second device based on the AI ​​capability of the second device;

[0164] Optionally, the first device sends AI model information to the second device based on the tenth indication information.

[0165] The communication method provided in the embodiment of the present application can enable the first device to determine the information of the AI ​​model to be sent based on the capabilities or indication information reported by the second device, thereby enabling the activation of the AI ​​function.

[0166] In combination with the third aspect, in certain implementations of the third aspect, the tenth indication information includes at least one of the following information: computing capability information of the second device, storage capability information of the second device or information on whether the second device stores the first AI model, the neural network structure supported by the second device, the neural network scale supported by the second device, and the AI ​​communication function supported by the second device, wherein the first AI model is any one AI model.

[0167] The above capability information may be expressed in various forms to increase the flexibility of the solution.

[0168] In combination with the third aspect, in certain implementations of the third aspect, before the first device receives the tenth indication information from the second device, the method further includes: the first device sending a first message to the second device, where the first message is used to instruct the second device to report the AI ​​capabilities of the second device.

[0169] In the communication method provided in the embodiment of the present application, the first device obtains the AI ​​capability of the second device by sending a first message.

[0170] In combination with the third aspect, in certain implementations of the third aspect, the first message is included in an eighth response message.

[0171] The eighth response message is used to confirm the conflict resolution in the random access process, and may be MSG4 in the random access process or a message similar to MSG4.

[0172] In combination with the third aspect, in certain implementations of the third aspect, the first message is included in a seventh response message.

[0173] The seventh response message is used to respond to the random access of the terminal device, and can be MSG2 in the random access process or a message similar to MSG2. The seventh response message includes the index number of the random preamble sequence corresponding to the terminal device determined by the base station.

[0174] In conjunction with the third aspect, in certain implementations of the third aspect, the method further includes:

[0175] The first device sends first resource indication information to the second device, where the first resource indication information indicates a resource for the second device to receive a first message.

[0176] Optionally, the first resource indication information is included in the seventh response message;

[0177] Optionally, the first resource indication information is included in the above-mentioned eighth response message.

[0178] In the communication method provided in the embodiment of the present application, there are multiple ways for the first device to send the first message, which improves the flexibility of the solution.

[0179] In conjunction with the third aspect, in certain implementations of the third aspect, before sending the seventh response message, the method further includes:

[0180] An initial access request is received from the second device, and the seventh response message responds to the initial access request.

[0181] In combination with the third aspect, in certain implementations of the third aspect, before the first device receives the tenth indication information from the second device, the method further includes: the first device sending the above-mentioned eighth response message to the second device.

[0182] In combination with the third aspect, in certain implementations of the third aspect, before the first device receives the tenth indication information from the second device, the method further includes: the first device sending the above-mentioned seventh response message to the second device.

[0183] In combination with the third aspect, in certain implementations of the third aspect, the tenth indication information is included in the eighth message.

[0184] The eighth message is used to request to establish a connection during the random access process, and can be MSG3 in the random access process or a message similar to MSG3. The eighth message includes the identifier of the terminal device.

[0185] In combination with the third aspect, in certain implementations of the third aspect, the tenth indication information is included in the seventh message.

[0186] The seventh message is used to initiate a random access procedure in the random access procedure, and may be MSG1 in the random access procedure or a message similar to MSG1.

[0187] In conjunction with the third aspect, in certain implementations of the third aspect, the method further includes:

[0188] The first device receives second resource indication information from the second device, where the second resource indication information indicates a resource from which the first device receives the tenth indication information.

[0189] Optionally, the second resource indication information is included in the seventh message;

[0190] Optionally, the second resource indication information is included in the above-mentioned eighth message.

[0191] In combination with the third aspect, in certain implementations of the third aspect, the information of the AI ​​model is included in the above-mentioned eighth response message.

[0192] In conjunction with the third aspect, in certain implementations of the third aspect, the method further includes:

[0193] The first device sends third resource indication information to the second device, where the third resource indication information indicates a resource from which the second device receives information about the AI ​​model.

[0194] Optionally, the third resource indication information is included in the above-mentioned eighth response message.

[0195] In the communication method provided in the embodiment of the present application, the above-mentioned signaling interaction between the first device and the second device can be completed at different stages during the initial access process, thereby being combined with the current initial access process to improve the integration of the solution with current technologies.

[0196] In combination with the third aspect, in certain implementations of the third aspect, the first device locally stores N neural networks, and the method further includes: the first device sends at least one neural network to the second device, where N is a positive integer, and the at least one neural network is part or all of the N neural networks.

[0197] In combination with the third aspect, in certain implementations of the third aspect, the N neural networks are applied to the second device or the first device; the first device sending at least one neural network to the second device includes at least one of the following: the first device sending M neural networks, the network weights corresponding to the M neural networks, the network structures corresponding to the M neural networks, or the network structure of a neural network to the second device, where M is a positive integer less than or equal to N.

[0198] According to the communication method provided in the embodiment of the present application, the first device can send a candidate neural network to the second device.

[0199] In conjunction with the third aspect, in certain implementations of the third aspect, the method further includes:

[0200] The first device sends eighth indication information to the second device, where the eighth indication information is used to indicate a currently applicable neural network.

[0201] In the communication method provided by the embodiment of the present application, the first device can specify the currently applicable neural network from the candidate neural networks through indication information.

[0202] In combination with the third aspect, in certain implementations of the third aspect, the N neural networks include N / 2 neural network pairs, each pair of neural network pairs includes a first neural network and a second neural network, and N is a non-zero even number; the first device sends at least one neural network to the second device, including at least one of the following: the first device sends to the second device Q second neural networks, Q first neural networks, network weights corresponding to the Q first neural networks, network weights corresponding to the Q second neural networks, network structures corresponding to the Q first neural networks, network structures corresponding to the Q second neural networks, a network structure of a first neural network, or a network structure of a second neural network, wherein Q is a positive integer less than or equal to N / 2.

[0203] In the communication method provided by the embodiment of the present application, a first device can send a candidate neural network pair to a second device for use by the first device and the second device as a pair.

[0204] Exemplarily, the neural networks that form a neural network pair may have the same index.

[0205] In combination with the third aspect, in certain implementations of the third aspect, the first device sends ninth indication information to the second device, where the ninth indication information is used to indicate a currently applicable neural network pair.

[0206] In the communication method provided by the embodiment of the present application, the first device can specify the currently applicable neural network pair from the candidate neural network pairs through indication information.

[0207] In combination with the third aspect, in some implementations of the third aspect, the first device sending at least one neural network to the second device includes: the first device sending a preset first neural network and / or a preset second neural network to the second device.

[0208] According to the communication method provided in the embodiment of the present application, the first device can only send the currently applicable neural network to the second device.

[0209] In combination with the third aspect, in certain implementations of the third aspect, the first device includes a terminal device or a network device, and the second device includes the terminal device or the network device. When the first device is the terminal device, the second device is the network device; when the first device is the network device, the second device is the terminal device.

[0210] In the fourth aspect, a communication method is provided. The communication method can be executed by a second device, or can be executed by a chip or circuit provided in the second device. This application does not limit this. For the sake of convenience of description, it can be illustrated by taking the execution by the second device as an example.

[0211] The communication method includes: a second device sending tenth indication information to a first device, where the tenth indication information is used to indicate an AI capability of the second device; and the second device receiving information about an AI model from the first device.

[0212] In the communication method provided in the embodiment of the present application, the first device can determine the information of the AI ​​model to be sent based on the tenth indication information reported by the second device and / or the AI ​​capability of the second device, thereby enabling the activation of the AI ​​function.

[0213] In combination with the fourth aspect, in certain implementations of the fourth aspect, the tenth indication information includes at least one of the following information: computing capability information of the second device, storage capability information of the second device or information on whether the second device stores the first AI model, the neural network structure supported by the second device, the neural network scale supported by the second device, and the AI ​​communication function supported by the second device, wherein the first AI model is any one AI model.

[0214] The above capability information may be expressed in various forms to increase the flexibility of the solution.

[0215] In combination with the fourth aspect, in certain implementations of the fourth aspect, before the second device sends the tenth indication information to the first device, the method further includes: the second device receives a first message from the first device, where the first message is used to instruct the second device to report the AI ​​capabilities of the second device.

[0216] In the communication method provided in the embodiment of the present application, the first device obtains the capability of the second device by sending a first message.

[0217] In combination with the fourth aspect, in certain implementations of the fourth aspect, the first message is included in an eighth response message.

[0218] The eighth response message is used to confirm the conflict resolution in the random access process, and may be MSG4 in the random access process or a message similar to MSG4.

[0219] In combination with the fourth aspect, in certain implementations of the fourth aspect, the first message is included in a seventh response message.

[0220] The seventh response message is used to respond to the random access of the terminal device, and can be MSG2 in the random access process or a message similar to MSG2. The seventh response message includes the index number of the random preamble sequence corresponding to the terminal device determined by the base station.

[0221] In conjunction with the fourth aspect, in certain implementations of the fourth aspect,

[0222] The second device receives first resource indication information from the first device, where the first resource indication information indicates a resource for the second device to receive a first message.

[0223] Optionally, the first resource indication information is included in the seventh response message;

[0224] Optionally, the first resource indication information is included in the above-mentioned eighth response message.

[0225] In the communication method provided in the embodiment of the present application, there are multiple ways for the first device to send the first message, which improves the flexibility of the solution.

[0226] In combination with the fourth aspect, in certain implementations of the fourth aspect, before the second device sends the tenth indication information to the first device, the method further includes: the second device receiving the eighth response message mentioned above from the first device.

[0227] In combination with the fourth aspect, in certain implementations of the fourth aspect, before the second device sends the tenth indication information to the first device, the method further includes: the second device receiving the seventh response message mentioned above from the first device.

[0228] In combination with the fourth aspect, in certain implementations of the fourth aspect, the tenth indication information is included in the eighth message.

[0229] The eighth message is used to request to establish a connection during the random access process, and can be MSG3 in the random access process or a message similar to MSG3. The eighth message includes the identifier of the terminal device.

[0230] In combination with the fourth aspect, in certain implementations of the fourth aspect, the tenth indication information is included in the seventh message.

[0231] The seventh message is used to initiate a random access procedure in the random access procedure, and may be MSG1 in the random access procedure or a message similar to MSG1.

[0232] In conjunction with the fourth aspect, in certain implementations of the fourth aspect, the method further includes:

[0233] The second device sends second resource indication information to the first device, where the second resource indication information indicates a resource from which the first device receives the tenth indication information.

[0234] Optionally, the second resource indication information is included in the seventh message;

[0235] Optionally, the second resource indication information is included in the above-mentioned eighth message.

[0236] In combination with the fourth aspect, in certain implementations of the fourth aspect, the information of the AI ​​model is included in the above-mentioned eighth response message.

[0237] In conjunction with the fourth aspect, in certain implementations of the fourth aspect,

[0238] The second device receives third resource indication information from the first device, where the third resource indication information indicates a resource from which the second device receives information about the AI ​​model.

[0239] Optionally, the third resource indication information is included in the above-mentioned eighth response message.

[0240] In the communication method provided in the embodiment of the present application, the above-mentioned signaling interaction between the first device and the second device can be completed at different stages during the initial access process, thereby being combined with the current initial access process to improve the integration of the solution with current technologies.

[0241] In combination with the fourth aspect, in certain implementations of the fourth aspect, the first device locally stores N neural networks, and the method also includes: the second device receives at least one neural network from the first device, where N is a positive integer, and the at least one neural network is part or all of the N neural networks.

[0242] In combination with the fourth aspect, in certain implementations of the fourth aspect, the N neural networks are applied to the second device or the first device; the second device receives at least one neural network from the first device, including at least one of the following: the second device receives M neural networks from the first device, the network weights corresponding to the M neural networks, the network structures corresponding to the M neural networks, or the network structure of a neural network, where M is a positive integer less than or equal to N.

[0243] In combination with the fourth aspect, in certain implementations of the fourth aspect, the N neural networks include N / 2 neural network pairs, each pair of neural network pairs includes a first neural network and a second neural network, and N is a non-zero even number; the second device receives at least one neural network from the first device, including at least one of the following: the second device receives from the first device Q second neural networks, Q first neural networks, network weights corresponding to the Q first neural networks, network weights corresponding to the Q second neural networks, network structures corresponding to the Q first neural networks, network structures corresponding to the Q second neural networks, a network structure of a first neural network, or a network structure of a second neural network, wherein Q is a positive integer less than or equal to N / 2.

[0244] According to the communication method provided in the embodiment of the present application, the first device can send a candidate neural network to the second device.

[0245] In combination with the fourth aspect, in certain implementations of the fourth aspect, the method further includes: the second device receiving eighth indication information from the first device, where the eighth indication information is used to indicate a currently applicable neural network.

[0246] In the communication method provided by the embodiment of the present application, the first device can specify the currently applicable neural network from the candidate neural networks through indication information.

[0247] In combination with the fourth aspect, in certain implementations of the fourth aspect, the method further includes: the second device receiving ninth indication information from the first device, the ninth indication information being used to indicate a currently applicable neural network pair.

[0248] In the communication method provided by the embodiment of the present application, the first device can specify the currently applicable neural network pair from the candidate neural network pairs through indication information.

[0249] In combination with the fourth aspect, in certain implementations of the fourth aspect, the first device includes a terminal device or a network device, and the second device includes the terminal device or the network device. When the first device is the terminal device, the second device is the network device; when the first device is the network device, the second device is the terminal device.

[0250] In a fifth aspect, a device for adjusting a neural network is provided, wherein the device for adjusting a neural network includes a processor for implementing the function of the first device in the method described in the first aspect.

[0251] Optionally, the apparatus for adjusting a neural network may further include a memory coupled to the processor, and the processor is used to implement the function of the first device in the method described in the first aspect above.

[0252] In one possible implementation, the memory is used to store program instructions and data. The memory is coupled to the processor, and the processor can call and execute the program instructions stored in the memory to implement the functions of the first device in the method described in the first aspect above.

[0253] Optionally, the apparatus for adjusting a neural network may further include a communication interface for communicating between the apparatus for adjusting a neural network and other devices. When the apparatus for adjusting a neural network is a network device, the communication interface may be a transceiver, an input / output interface, or a circuit.

[0254] In one possible design, the apparatus for adjusting a neural network includes: a processor and a communication interface, configured to implement the functions of the first device in the method described in the first aspect above, specifically including:

[0255] The processor communicates with the outside world using the communication interface;

[0256] The processor is used to run a computer program so that the device implements any one of the methods described in the first aspect above.

[0257] It can be understood that the external part may be an object other than the processor, or an object other than the device.

[0258] In another possible design, the apparatus for adjusting a neural network is a chip or a chip system. The communication interface may be an input / output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip or chip system. The processor may also be embodied as a processing circuit or a logic circuit.

[0259] In a sixth aspect, a device for adjusting a neural network is provided, wherein the device for adjusting a neural network includes a processor for implementing the function of the second device in the method described in the second aspect above.

[0260] Optionally, the apparatus for adjusting a neural network may further include a memory coupled to the processor, and the processor is used to implement the function of the second device in the method described in the second aspect above.

[0261] In one possible implementation, the memory is used to store program instructions and data. The memory is coupled to the processor, and the processor can call and execute the program instructions stored in the memory to implement the functions of the second device in the method described in the second aspect above.

[0262] Optionally, the apparatus for adjusting a neural network may further include a communication interface, wherein the communication interface is used for the apparatus for adjusting a neural network to communicate with other devices. When the apparatus for adjusting a neural network is a terminal device, the transceiver may be a communication interface, or an input / output interface.

[0263] In one possible design, the apparatus for adjusting a neural network includes: a processor and a communication interface, configured to implement the functions of the second device in the method described in the second aspect above, specifically including:

[0264] The processor communicates with the outside world using the communication interface;

[0265] The processor is used to run a computer program so that the device implements any one of the methods described in the second aspect above.

[0266] It can be understood that the external part may be an object other than the processor, or an object other than the device.

[0267] In another implementation, when the apparatus for adjusting a neural network is a chip or a chip system, the communication interface may be an input / output interface, an interface circuit, an output circuit, an input circuit, a pin, or related circuits on the chip or chip system. The processor may also be embodied as a processing circuit or a logic circuit.

[0268] In a seventh aspect, a device for adjusting a neural network is provided, wherein the device for adjusting a neural network includes a processor for implementing the function of the first device in the method described in the third aspect.

[0269] Optionally, the apparatus for adjusting a neural network may further include a memory coupled to the processor, and the processor is used to implement the function of the first device in the method described in the third aspect above.

[0270] In one possible implementation, the memory is used to store program instructions and data. The memory is coupled to the processor, and the processor can call and execute the program instructions stored in the memory to implement the functions of the first device in the method described in the third aspect.

[0271] Optionally, the apparatus for adjusting a neural network may further include a communication interface for communicating between the apparatus for adjusting a neural network and other devices. When the apparatus for adjusting a neural network is a network device, the communication interface may be a transceiver, an input / output interface, or a circuit.

[0272] In one possible design, the apparatus for adjusting a neural network includes: a processor and a communication interface, configured to implement the functions of the first device in the method described in the third aspect above, specifically including:

[0273] The processor communicates with the outside world using the communication interface;

[0274] The processor is used to run a computer program so that the device implements any one of the methods described in the third aspect above.

[0275] It can be understood that the external part may be an object other than the processor, or an object other than the device.

[0276] In another possible design, the apparatus for adjusting a neural network is a chip or a chip system. The communication interface may be an input / output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip or chip system. The processor may also be embodied as a processing circuit or a logic circuit.

[0277] In an eighth aspect, a device for adjusting a neural network is provided, wherein the device for adjusting a neural network includes a processor for implementing the function of the second device in the method described in the fourth aspect.

[0278] Optionally, the apparatus for adjusting a neural network may further include a memory coupled to the processor, and the processor is used to implement the function of the second device in the method described in the fourth aspect above.

[0279] In one possible implementation, the memory is used to store program instructions and data. The memory is coupled to the processor, and the processor can call and execute the program instructions stored in the memory to implement the functions of the second device in the method described in the fourth aspect.

[0280] Optionally, the apparatus for adjusting a neural network may further include a communication interface, wherein the communication interface is used for the apparatus for adjusting a neural network to communicate with other devices. When the apparatus for adjusting a neural network is a terminal device, the transceiver may be a communication interface, or an input / output interface.

[0281] In one possible design, the apparatus for adjusting a neural network includes: a processor and a communication interface, configured to implement the functions of the second device in the method described in the fourth aspect above, specifically including:

[0282] The processor communicates with the outside world using the communication interface;

[0283] The processor is used to run a computer program so that the device implements any one of the methods described in the fourth aspect above.

[0284] It can be understood that the external part may be an object other than the processor, or an object other than the device.

[0285] In another implementation, when the apparatus for adjusting a neural network is a chip or a chip system, the communication interface may be an input / output interface, an interface circuit, an output circuit, an input circuit, a pin, or related circuits on the chip or chip system. The processor may also be embodied as a processing circuit or a logic circuit.

[0286] In a ninth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a communication device, the communication device implements the method in the first aspect and any possible implementation manner of the first aspect.

[0287] In a tenth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a communication device, the communication device implements the method in the second aspect and any possible implementation manner of the second aspect.

[0288] In the eleventh aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a communication device, the communication device implements the method in the third aspect and any possible implementation manner of the third aspect.

[0289] In the twelfth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a communication device, the communication device implements the method in the fourth aspect and any possible implementation manner of the fourth aspect.

[0290] In a thirteenth aspect, a computer program product comprising instructions is provided, which, when executed by a computer, enables a communication device to implement the method in the first aspect and any possible implementation manner of the first aspect.

[0291] In a fourteenth aspect, a computer program product comprising instructions is provided, which, when executed by a computer, enables a communication device to implement the method in the second aspect and any possible implementation manner of the second aspect.

[0292] In a fifteenth aspect, a computer program product comprising instructions is provided, which, when executed by a computer, enables a communication device to implement the method in the third aspect and any possible implementation manner of the third aspect.

[0293] In the sixteenth aspect, a computer program product comprising instructions is provided, which, when executed by a computer, enables a communication device to implement the method in the fourth aspect and any possible implementation manner of the fourth aspect.

[0294] In the seventeenth aspect, a communication system is provided, comprising the apparatus for adjusting a neural network shown in the fifth aspect and the apparatus for adjusting a neural network shown in the sixth aspect.

[0295] In the eighteenth aspect, a communication system is provided, comprising the communication device shown in the seventh aspect and the communication device shown in the eighth aspect.

[0296] In the nineteenth aspect, a chip device is provided, comprising a processing circuit, which is used to call and run a program from a memory, so that a communication device equipped with the chip device executes a method in any possible implementation of the first to fourth aspects above. BRIEF DESCRIPTION OF THE DRAWINGS

[0297] Figure 1 1 is a schematic diagram of a system 100 to which the method for adjusting a neural network according to an embodiment of the present application can be applied.

[0298] Figure 2 It is a schematic flowchart of CSI feedback based on the existing CSI feedback process.

[0299] Figure 3 This is a schematic diagram of an AI-based CSI feedback process provided in an embodiment of the present application.

[0300] Figure 4 This is a stage diagram of AI-based CSI compression feedback provided by this application.

[0301] Figure 5 (a)-(e) are schematic diagrams of a method for adjusting a neural network provided in an embodiment of the present application.

[0302] Figure 6 (a) and (b) are schematic diagrams of determining and estimating downlink channel information provided in an embodiment of the present application.

[0303] Figure 7 (a)-(d) are schematic diagrams of another method for determining and estimating downlink channel information provided in an embodiment of the present application.

[0304] Figure 8 (a)-(f) are schematic flowcharts of sending information of an updated encoder and / or decoder provided in an embodiment of the present application.

[0305] Figure 9 (a)-(f) are another schematic flowchart of sending updated encoder and / or decoder information provided by an embodiment of the present application.

[0306] Figure 10 This is a schematic flow chart of a communication method provided in an embodiment of the present application.

[0307] Figure 11 This is a schematic flow chart of initial access.

[0308] Figure 12 1 is a schematic structural diagram of an apparatus 1200 for adjusting a neural network applicable to an embodiment of the present application.

[0309] Figure 13 is a schematic diagram of the communication device 1300 proposed in this application.

[0310] Figure 14 It is a schematic diagram of the first device 1400 provided in this application.

[0311] Figure 15 is a schematic diagram of an apparatus 1500 for adjusting a neural network proposed in this application.

[0312] Figure 16 is a schematic diagram of the communication device 1600 proposed in this application.

[0313] Figure 17 It is a schematic diagram of the second device 1700 provided in this application. DETAILED DESCRIPTION

[0314] The technical solution in this application will be described below with reference to the accompanying drawings.

[0315] The technical solutions of the embodiments of the present application can be applied to various communication systems, such as: long term evolution (LTE) system, LTE frequency division duplex (FDD) system, LTE time division duplex (TDD) system, world-wide interoperability for microwave access (WiMAX) communication system, fifth generation (5G) system, new radio (NR) or future network, etc. The 5G mobile communication system described in this application includes a non-standalone (NSA) 5G mobile communication system or a standalone (SA) 5G mobile communication system. The technical solutions provided in this application can also be applied to future communication systems, such as the sixth generation mobile communication system. The communication system can also be a public land mobile network (PLMN) network, a device-to-device (D2D) communication system, a machine-to-machine (M2M) communication system, an Internet of Things (IoT) communication system or other communication systems.

[0316] The terminal equipment in the embodiments of the present application may refer to an access terminal, a user unit, a user station, a mobile station, a mobile station, a relay station, a remote station, a remote terminal, a mobile device, a user terminal, a user equipment (UE), a terminal, a wireless communication device, a user agent, or a user device. The terminal equipment may also be a cellular phone, a cordless phone, a session initiation protocol (SIP) phone, a wireless local loop (WLL) station, a personal digital assistant (PDA), a handheld device with wireless communication capabilities, a computing device or other processing device connected to a wireless modem, an in-vehicle device, a wearable device, a terminal device in a 5G network, a terminal device in a future evolved public land mobile network (PLMN), or a terminal device in a future Internet of Vehicles, etc., and the embodiments of the present application are not limited thereto.

[0317] As an example and not a limitation, in the embodiments of the present application, wearable devices may also be referred to as wearable smart devices, which are a general term for wearable devices that are intelligently designed and developed using wearable technology for daily wear, such as glasses, gloves, watches, clothing, and shoes. A wearable device is a portable device that is worn directly on the body or integrated into the user's clothes or accessories. Wearable devices are not just hardware devices, but also achieve powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable smart devices include those that are fully functional, large in size, and can achieve complete or partial functions without relying on smartphones, such as smart watches or smart glasses, as well as those that only focus on a certain type of application function and need to be used in conjunction with other devices such as smartphones, such as various smart bracelets and smart jewelry for vital sign monitoring.

[0318] Furthermore, in the embodiments of the present application, the terminal device may also be a terminal device in an IoT system. IoT is an important component of the future development of information technology. Its main technical feature is to connect objects to the network through communication technology, thereby realizing an intelligent network that interconnects humans and machines and things. In the embodiments of the present application, IoT technology can achieve massive connections, deep coverage, and terminal power saving through, for example, narrowband (NB) technology.

[0319] In addition, in an embodiment of the present application, the terminal device may also include sensors such as smart printers, train detectors, and gas stations. Its main functions include collecting data (part of the terminal device), receiving control information and downlink data from the network device, and sending electromagnetic waves to transmit uplink data to the network device.

[0320] The network device in the embodiment of the present application can be any communication device with wireless transceiver functions for communicating with a terminal device. The device includes but is not limited to: an evolved Node B (eNB), a radio network controller (RNC), a Node B (NB), a home evolved Node B (HeNB, or home Node B, HNB), a baseband unit (BBU), an access point (AP) in a wireless fidelity (WIFI) system, a wireless relay node, a wireless backhaul node, a transmission point (TP) or a transmission and reception point (TRP), etc., and can also be a 5G system, such as a gNB in ​​an NR system, or a transmission point (TRP or TP), one or a group of antenna panels (including multiple antenna panels) of a base station in a 5G system, or a network node constituting a gNB or a transmission point, such as a baseband unit (BBU) or a distributed unit (DU), etc.

[0321] In some deployments, the network device in the embodiments of the present application may refer to a centralized unit (CU) or a distributed unit (DU), or the network device may include both a CU and a DU. The gNB may also include an active antenna unit (AAU). The CU implements some of the gNB's functions, while the DU implements some of the gNB's functions. For example, the CU is responsible for processing non-real-time protocols and services, implementing the functions of the radio resource control (RRC) and packet data convergence protocol (PDCP) layers. The DU is responsible for processing physical layer protocols and real-time services, implementing the functions of the radio link control (RLC), media access control (MAC), and physical (PHY) layers. The AAU implements some physical layer processing functions, RF processing, and active antenna-related functions. Because RRC layer information ultimately becomes PHY layer information, or is converted from PHY layer information, in this architecture, higher-layer signaling, such as RRC layer signaling, can also be considered to be sent by the DU, or by both the DU and the AAU. It is understood that the network device may include one or more of a CU node, a DU node, and an AAU node. In addition, the CU may be classified as a network device in an access network (RAN) or a network device in a core network (CN), which is not limited in this application.

[0322] Furthermore, the CU can be divided into a control plane central unit (CU-CP) and a user plane central unit (CU-UP). The CU-CP and CU-UP can also be deployed on different physical devices. The CU-CP is responsible for control plane functions, primarily including the RRC layer and the PDCP-C layer. The PDCP-C layer is primarily responsible for encryption and decryption, integrity protection, and data transmission of control plane data. The CU-UP is responsible for user plane functions, primarily including the SDAP layer and the PDCP-U layer. The SDAP layer is primarily responsible for processing core network data and mapping flows to bearers. The PDCP-U layer is primarily responsible for at least one function in the data plane, including encryption and decryption, integrity protection, header compression, sequence number maintenance, and data transmission. Specifically, the CU-CP and CU-UP are connected via a communication interface (e.g., the E1 interface). The CU-CP, representing a network device, connects to the core network device via a communication interface (e.g., the Ng interface) and connects to the DU via a communication interface (e.g., the F1-C (control plane) interface). The CU-UP connects to the DU via a communication interface (e.g., the F1-U (user plane) interface).

[0323] There is another possible implementation in which the PDCP-C layer is also included in the CU-UP.

[0324] It can be understood that the above protocol layer divisions of CU and DU, as well as CU-CP and CU-UP are only examples, and there may be other division methods, which are not limited in the embodiments of the present application.

[0325] The network device mentioned in the embodiments of the present application may be a device including a CU, or a DU, or a device including a CU and a DU, or a device including a control plane CU node (CU-CP node) and a user plane CU node (CU-UP node) and a DU node.

[0326] The network equipment and terminal devices can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on the water; they can also be deployed in the air on aircraft, balloons, and satellites. The embodiments of this application do not limit the scenarios in which the network equipment and terminal devices are located.

[0327] In an embodiment of the present application, a terminal device or network device includes a hardware layer, an operating system layer running on the hardware layer, and an application layer running on the operating system layer. The hardware layer includes hardware such as a central processing unit (CPU), a memory management unit (MMU), and memory (also known as main memory). The operating system can be any one or more computer operating systems that implement business processing through processes, such as a Linux operating system, a Unix operating system, an Android operating system, an iOS operating system, or a Windows operating system. The application layer includes applications such as browsers, address books, word processing software, and instant messaging software.

[0328] In addition, various aspects or features of the present application can be implemented as methods, apparatus, or articles of manufacture using standard programming and / or engineering techniques. The term "article of manufacture" as used in this application covers computer programs that can be accessed from any computer-readable device, carrier, or medium. For example, computer-readable media may include, but are not limited to: magnetic storage devices (e.g., hard disks, floppy disks, or magnetic tapes, etc.), optical disks (e.g., compact discs (CDs), digital versatile discs (DVDs), etc.), smart cards, and flash memory devices (e.g., erasable programmable read-only memories (EPROMs), cards, sticks, or key drives, etc.). In addition, the various storage media described herein may represent one or more devices and / or other machine-readable media for storing information. The term "machine-readable storage medium" may include, but is not limited to, wireless channels and various other media capable of storing, containing, and / or carrying instructions and / or data.

[0329] To facilitate understanding of the embodiments of the present application, first Figure 1 The communication system shown in FIG. 1 is used as an example to describe in detail a communication system applicable to an embodiment of the present application. Figure 1 1 is a schematic diagram of a communication system 100 applicable to a method for adjusting a neural network according to an embodiment of the present application. Figure 1 As shown, the communication system 100 may include at least one network device, such as Figure 1 The network device 110 shown; the communication system 100 may also include at least one terminal device, such as Figure 1The terminal device 120 shown. The network device 110 and the terminal device 120 can communicate via a wireless link. Each communication device, such as the network device 110 or the terminal device 120, can be configured with multiple antennas. For each communication device in the communication system 100, the configured multiple antennas may include at least one transmit antenna for sending signals and at least one receive antenna for receiving signals. Therefore, the communication devices in the communication system 100, such as the network device 110 and the terminal device 120, can communicate using multi-antenna technology.

[0330] It should be understood that Figure 1 This is a simplified schematic diagram for ease of understanding only. The communication system 100 may also include other network devices or other terminal devices. Figure 1 Not drawn in.

[0331] To facilitate understanding of the embodiments of the present application, several basic concepts involved in the embodiments of the present application are briefly explained. It should be understood that the basic concepts introduced below are briefly explained using the basic concepts specified in the NR protocol as an example, but the embodiments of the present application are not limited to being applicable only to NR systems. Therefore, the standard names that appear when describing the NR system as an example are all functional descriptions, and the specific names are not limited. They only indicate the functions of the device and can be extended to other systems accordingly, such as 2G, 3G, 4G or future communication systems.

[0332] 1. Precoding technology.

[0333] When the channel state is known, the network device can use a precoding matrix that matches the channel state to process the signal to be transmitted, so that the precoded signal to be transmitted is adapted to the channel, so that the receiving device can suppress the signal influence between multiple receiving devices and maximize the signal-to-interference-plus-noise ratio of the received signal. Therefore, by precoding the signal to be transmitted, the quality of the received signal (for example, signal to interference plus noise ratio (SINR)) is improved. Therefore, the use of precoding technology can enable the transmitting device and multiple receiving devices to transmit on the same time-frequency resources, that is, to achieve MU-MIMO.

[0334] It should be understood that the description of precoding technology in this application is for illustrative purposes only and is not intended to limit the scope of protection of the embodiments of this application. During implementation, the transmitting device may also perform precoding in other ways. For example, when channel information (such as, but not limited to, a channel matrix) is unavailable, a precoding matrix or weighted processing method may be used for precoding. For the sake of brevity, the specific details are not repeated in this application.

[0335] 2. Precoding matrix indicator (PMI).

[0336] It can be used to indicate a precoding matrix. The precoding matrix can, for example, be a precoding matrix determined by the terminal device based on the channel matrix of each frequency domain unit. The precoding matrix can be determined by the terminal device through channel estimation or other methods, or based on channel reciprocity. However, it should be understood that the specific method for determining the precoding matrix by the terminal device is not limited to that described above. Specific implementation methods can be referenced in the prior art and are not listed here for the sake of brevity.

[0337] For example, the precoding matrix can be obtained by performing singular value decomposition (SVD) on the channel matrix or the covariance matrix of the channel matrix, or by performing eigenvalue decomposition (EVD) on the covariance matrix of the channel matrix. It should be understood that the methods for determining the precoding matrix listed above are only examples and should not constitute any limitation to this application. The methods for determining the precoding matrix can refer to the prior art and are not listed here for the sake of brevity.

[0338] The precoding matrix determined by the terminal device can be referred to as the precoding matrix to be fed back, or in other words, the precoding matrix to be reported. The terminal device can indicate the precoding matrix to be fed back through the PMI, so that the network device can recover the precoding matrix based on the PMI. The precoding matrix recovered by the network device based on the PMI can be the same as or similar to the precoding matrix to be fed back.

[0339] In downlink channel measurement, the higher the degree of similarity between the precoding matrix determined by the network device based on the PMI and the precoding matrix determined by the terminal device, the more adaptable the precoding matrix determined by the network device for data transmission will be to the downlink channel, thereby improving the signal transmission quality.

[0340] It should be understood that PMI is only a name and should not constitute any limitation to this application. This application does not exclude the possibility of defining signaling with other names in future protocols for the same or similar functions.

[0341] It should be noted that, according to the method provided by the embodiment of the present application, the network device can determine the precoding matrix corresponding to one or more frequency domain units based on the feedback of the terminal device. The precoding matrix thus determined by the network device can be used directly for downlink data transmission; it can also be subjected to some beamforming methods, such as zero forcing (ZF), regularized zero-forcing (RZF), minimum mean-squared error (MMSE), maximizing the signal-to-leakage-and-noise ratio (SLNR), etc., to obtain the precoding matrix ultimately used for downlink data transmission. This application is not limited to this. Unless otherwise specified, the precoding matrices referred to below may all refer to the precoding matrices determined based on the method provided in this application.

[0342] In the embodiment of the present application, the method for implementing channel information feedback based on the above-mentioned precoding technology and precoding matrix indication is referred to as a traditional channel information feedback method. It should be noted that this is only a name and does not constitute any limitation on the scope of protection of this application. It is to distinguish the channel information feedback method in the AI ​​mode from the method for implementing channel information feedback based on the precoding matrix indication. In addition, in this application, the method for feeding back channel information in the non-AI mode may also be referred to as a traditional channel information feedback method or a conventional channel information feedback method.

[0343] 3. Channel reciprocity.

[0344] In time division duplexing (TDD) mode, uplink and downlink channels transmit signals on the same frequency domain resources but different time domain resources. Within a relatively short period of time (e.g., the coherence time of channel propagation), it can be assumed that the signals on the uplink and downlink channels traverse the same channel, and the uplink and downlink channels can be acquired equivalently. This is the reciprocity of the uplink and downlink channels. Based on the reciprocity of the uplink and downlink channels, network equipment can measure the uplink channel based on uplink reference signals, such as the sounding reference signal (SRS). The downlink channel can also be estimated based on the uplink channel, thereby determining the precoding matrix used for downlink transmission.

[0345] However, in frequency division duplexing (FDD) mode, since the frequency band spacing between the uplink and downlink channels is much larger than the coherence bandwidth, the uplink and downlink channels do not have complete reciprocity. The precoding matrix used to determine the downlink transmission using the uplink channel may not be compatible with the downlink channel. In an FDD system, network devices can rely on the channel state information (CSI) fed back by the terminal device to the network device for downlink precoding. Specifically, the basic flow chart of CSI feedback between the network device and the terminal device is as follows: Figure 2 As shown, Figure 2 It is a schematic flowchart of CSI feedback based on the existing CSI feedback process.

[0346] like Figure 2 As shown, the CSI feedback process includes: S210, the network device sends channel measurement configuration information to the terminal device. The channel measurement configuration information notifies the terminal device to perform channel measurement (and may also include the time for performing channel measurement). After the network device sends the channel measurement configuration information, it also needs to send reference information for channel measurement to the terminal device, that is, Figure 2 The method flow shown also includes S220: the network device sends a reference signal to the terminal device. The terminal device receives the reference signal and can perform channel measurement based on the reference signal to calculate the final CSI feedback amount, that is, Figure 2 The method flow shown further includes S230: the terminal device obtains CSI. Specifically, the terminal device needs to feed back the obtained CSI to the network device. Figure 2 The method flow shown further includes S240: the terminal device feeds back CSI to the network device.

[0347] Furthermore, after the network device learns the CSI, it can send data to the terminal device based on the CSI. Figure 2 The method flow shown further includes S250: the network device sends data to the terminal device.

[0348] Specifically, the network device determines the number of data streams transmitted to the terminal device based on the channel rank indicator (RI) fed back by the terminal device; the network device determines the modulation order and channel coding code rate of the data transmitted to the terminal device based on the channel quality indicator (CQI) fed back by the terminal device; the network device determines the precoding of the data transmitted to the terminal device based on the precoding matrix indicator (PMI) fed back by the terminal device.

[0349] 4. Reference signal (RS)

[0350] RS may also be referred to as a pilot, a reference sequence, etc. In an embodiment of the present application, a reference signal may be a reference signal for channel measurement. For example, the reference signal may be a channel state information reference signal (CSI-RS) for downlink channel measurement, or a sounding reference signal (SRS) for uplink channel measurement. It should be understood that the reference signals listed above are only examples and should not constitute any limitation to the present application. The present application does not exclude the possibility of defining other reference signals in future protocols to achieve the same or similar functions.

[0351] 5. Compression channel.

[0352] The compressed channel involved in the embodiments of the present application refers to the information obtained after the channel information or auxiliary CSI information is compressed by a compressor, which is called a compressed channel and can also be called a compressed codeword.

[0353] In addition, the information input to the decompressor in the embodiment of the present application can also be referred to as a compression channel. The compression channel in the present application will be described in detail below with reference to the accompanying drawings, and will not be described here in detail.

[0354] It should be noted that in this application, the information output by the compressor and / or the input information of the decompressor is referred to as a compressed channel for example only and does not constitute any limitation on the scope of protection of this application. For example, it can also be referred to as a codeword, compressed information or complex coefficient, etc., and examples will not be given one by one here.

[0355] 6. Channel sparsity.

[0356] In massive MIMO systems, the limited number of angles, multipath delays, and Dopplers in the signal propagation space, along with relatively concentrated angle spread, delay spread, and Doppler spread, means that the channel energy is concentrated along a few key paths. The energy on other paths is negligible, making the channel sparse. In the spatial domain, converting the spatial channel to the angle domain reveals that the channel energy is concentrated at certain angles. In the frequency domain, converting the frequency channel to the delay domain reveals that the channel energy is concentrated at certain multipath delays. In the time domain, converting the time channel to the Doppler domain reveals that the channel energy is concentrated at certain Dopplers.

[0357] 7. Auxiliary CSI information.

[0358] As can be seen from the above description of reciprocity, the uplink and downlink channels in FDD mode do not have complete reciprocity. However, the uplink and downlink channels in FDD mode still have partial reciprocity, such as angle reciprocity and delay reciprocity. Therefore, angle and delay can also be called reciprocity parameters.

[0359] As signals travel through wireless channels, they can take multiple paths from the transmitting antenna to the receiving antenna. Multipath delay causes frequency-selective fading, which is essentially a variation in the frequency domain channel. Latency is the transmission time of a wireless signal along different transmission paths. It is determined by distance and speed and has nothing to do with the frequency domain of the wireless signal. When signals travel along different transmission paths, different transmission delays occur due to varying distances. Therefore, the latency in the uplink and downlink channels in FDD mode can be considered the same, or reciprocal.

[0360] In addition, the angle may refer to the angle of arrival (AOA) of the signal arriving at the receiving antenna via the wireless channel, or the angle of departure (AOD) of the signal transmitted by the transmitting antenna. In an embodiment of the present application, the angle may refer to the angle of arrival of the uplink signal arriving at the network device, or the angle of departure of the downlink signal transmitted by the network device. The angle of arrival of the uplink reference signal and the angle of departure of the downlink reference signal may be considered to be the same, or reciprocal.

[0361] Since the uplink and downlink channels in FDD mode have partial reciprocity, that is, the network equipment in FDD mode can use uplink channel information to obtain reciprocal channel characteristics. In this application, the uplink channel information in FDD mode can be referred to as auxiliary CSI information.

[0362] 8. Artificial intelligence (AI).

[0363] The wave of artificial intelligence is sweeping the world, and many words are always lingering in our ears, such as artificial intelligence, machine learning, deep learning, etc. This application relates to deep learning in AI. Deep learning is a learning process that uses deep neural networks to solve feature expression. Neural networks generally include two major stages: training and testing. Training is the process of extracting model parameters from training data using a neural network model. Testing is to run the test data with a trained model (neural network model + model parameters) and then check the results. Among them, the data of the links involved in the training process are unified and abstracted to form a usable framework called a training framework. This application mainly relates to the adjustment of the compressor and decompressor in the AI-based CSI feedback process, wherein the compressor can also be called a CSI compressor, and the decompressor can also be called a CSI decompressor.

[0364] 9. Compressor and decompressor.

[0365] Current research on CSI feedback in massive MIMO systems leverages the channel sparsity in time, frequency, and space, as described above, to compress the CSI to be fed back. This compressed CSI is then transmitted to the network device via a feedback link, where it reconstructs the original sparse channel matrix from the feedback. This CSI feedback method, based on an automated compressor and decompressor architecture, involves encoding on the terminal device and decoding and reconstruction on the network device.

[0366] It should be noted that in this application, the unit that compresses the CSI to be fed back is referred to as a compressor for example only, and does not constitute any limitation on the scope of protection of this application. For example, it can also be referred to as an encoder, a compression unit, etc.; similarly, in this application, the unit that decompresses the compressed CSI is referred to as a decompressor for example only, and does not constitute any limitation on the scope of protection of this application. For example, it can also be referred to as a decoder, a decompression unit, etc.

[0367] It should be understood that the compressor and decompressor are a type of neural network.

[0368] 10. AI-based CSI feedback.

[0369] Specifically, if Figure 3 As shown, Figure 3 This is a schematic diagram of an AI-based CSI feedback process provided in an embodiment of the present application. In the embodiment of the present application, the framework including the compressor and decompressor may be referred to as an AI model or AI framework.

[0370] The network device and terminal device jointly train the resulting compressor and decompressor. The compressor is deployed in the terminal device, and the decompressor is deployed in the network device. The terminal device measures downlink channel information based on the downlink reference signal. This downlink channel information is encoded using the compressor to obtain compressed channel information that needs to be fed back to the air interface. This compressed channel information is then fed back to the network device via a feedback link. After receiving this compressed channel information, the network device decodes it using a locally deployed decompressor to obtain reconstructed information about the channel information fed back by the terminal device.

[0371] It should be noted that the training process of the compressor and decompressor involved in the AI-based CSI feedback process is relatively complex. Generally, the training of the compressor and decompressor is completed offline, and then the trained compressor and decompressor are applied to the compression and decompression of the online channel.

[0372] like Figure 4 As shown, Figure 4This is a phase diagram of AI-based CSI compression feedback provided by this application. Specifically, the AI-based CSI feedback process includes Figure 4 (a) shows the AI-based training process and Figure 4 (b) shows the AI-based feedback CSI process.

[0373] in, Figure 4 The AI-based training process shown in (a) is highly complex. Existing related research basically assumes that the AI ​​model is trained offline and the training results are then applied to the online CSI feedback process. In order to ensure the performance of channel compression, Figure 4 The training channel shown in (a) and Figure 4 The estimated channel (also called measurement channel, channel to be fed back, etc.) in (b) should have the same statistical performance (for example, similarity of angle and delay domain characteristics).

[0374] From the above, we can see that Figure 4 In the AI-based CSI compression process shown, it is generally necessary to ensure that the training channel and the estimated channel have the same statistical performance. Therefore, if the terminal device moves, the characteristics of the estimated channel may change, thereby affecting the performance of channel compression based on the trained compressor.

[0375] For example, based on the compressor #1 and decompressor #1 obtained by offline training of training channel #1, in principle, when the statistical performance of estimated channel #1 and training channel #1 are similar, the difference between the compressed channel #1 obtained by compressing estimated channel #1 by compressor #1 and then decompressing it by decompressor #1 to obtain the recovered channel #1 should meet the difference requirement (such as being less than a preset threshold value). If the terminal device is displaced, the channel characteristics of estimated channel #2 measured by the terminal device may be inconsistent with the above-mentioned estimated channel #1, and thus the difference between the recovered channel #2 obtained by compressing and decompressing estimated channel #2 by compressor #1 and decompressing estimated channel #2 by compressor #1 and decompressor #1 obtained by offline training of training channel #1 and estimated channel #2 may not meet the difference requirement. In this case, the compressor #1 and decompressor #1 need to be updated.

[0376] To solve Figure 4 The problem of deterioration of channel compression performance due to changes in channel characteristics that may occur in the AI ​​feedback CSI process shown in the figure, the present application provides a method for adjusting the neural network, which is applied in the scenario of channel information feedback. By updating the above-mentioned compressor and / or decompressor in real time during the channel measurement process, channel feedback is performed based on the updated compressor and / or decompressor, which can improve the performance of channel compression.

[0377] It should be noted that the scenario in which the method for adjusting the neural network provided in the embodiment of the present application can be applied is not limited to the scenario of channel information compression feedback, but can also be used in other scenarios, for example, the scenario in which the terminal device side and the network device side are respectively configured with an encoder and a decoder, or the scenario in which the terminal device side and the network device side are respectively configured with a modulator and a demodulator, etc.

[0378] That is to say, in the embodiments of the present application, there is no limitation on the specific forms of the neural networks configured on the terminal device side and the network device side respectively. It can be the above-mentioned compressor and decompressor, or other neural networks, and no further examples will be given here.

[0379] In order to facilitate understanding of the embodiments of the present application, the following explanations are made.

[0380] First, in this application, "used to indicate" can include being used for direct indication and being used for indirect indication. When describing that a certain indication information is used to indicate A, it can include that the indication information directly indicates A or indirectly indicates A, but it does not mean that the indication information must include A.

[0381] The information indicated by the indication information is called the information to be indicated. In the specific implementation process, there are many ways to indicate the information to be indicated. For example, but not limited to, the information to be indicated can be directly indicated, such as the information to be indicated itself or the index of the information to be indicated. The information to be indicated can also be indirectly indicated by indicating other information, wherein there is an association between the other information and the information to be indicated. It is also possible to indicate only a part of the information to be indicated, while the other parts of the information to be indicated are known or agreed in advance. For example, it is also possible to use the arrangement order of each piece of information that is pre-agreed (for example, specified by the protocol) to implement the indication of specific information, thereby reducing the indication overhead to a certain extent. At the same time, it is also possible to identify the common parts of each piece of information and indicate them uniformly to reduce the indication overhead caused by indicating the same information separately. For example, those skilled in the art should understand that the precoding matrix is ​​composed of precoding vectors, and the precoding vectors in the precoding matrix may have the same parts in terms of composition or other properties.

[0382] In addition, the specific indication method can also be various existing indication methods, such as but not limited to the above-mentioned indication methods and various combinations thereof. The specific details of the various indication methods can be referred to the prior art and will not be repeated herein. As can be seen from the above, for example, when it is necessary to indicate multiple information of the same type, there may be a situation where the indication methods for different information are different. In the specific implementation process, the required indication method can be selected according to specific needs. The embodiment of the present application does not limit the selected indication method. In this way, the indication method involved in the embodiment of the present application should be understood to cover various methods that can enable the party to be indicated to obtain the information to be indicated.

[0383] The information to be indicated may exist in other equivalent forms. For example, a row vector can be expressed as a column vector, a matrix can be represented by its transposed matrix, a matrix can be expressed as a vector or array, and the vector or array can be formed by connecting the row vectors or column vectors of the matrix, etc. The technical solutions provided in the embodiments of this application should be understood to cover various forms. For example, some or all of the features involved in the embodiments of this application should be understood to cover various forms of such features.

[0384] The information to be indicated can be sent as a whole, or divided into multiple sub-information and sent separately, and the sending period and / or sending timing of these sub-information can be the same or different. The specific sending method is not limited in this application. The sending period and / or sending timing of these sub-information can be predefined, for example, predefined according to a protocol, or configured by the transmitting device by sending configuration information to the receiving device. The configuration information can include, for example, but not limited to, one or a combination of at least two of radio resource control signaling, media access control (MAC) layer signaling, and physical layer signaling. The radio resource control signaling includes, for example, radio resource control (RRC) signaling; the MAC layer signaling includes, for example, a MAC control element (CE); and the physical layer signaling includes, for example, downlink control information (DCI).

[0385] Second, the first, second, and various numerical numbers (e.g., "#1," "#2") in this application are merely distinctions for ease of description and are not intended to limit the scope of the embodiments of this application. For example, to distinguish different information.

[0386] Third, in this application, "pre-set" may include indications by network device signaling or pre-defined, such as protocol definitions. "Pre-defined" may be implemented by pre-storing corresponding codes, tables, or other methods that can be used to indicate relevant information in devices (e.g., including terminal devices and network devices). This application does not limit the specific implementation method.

[0387] Fourth, the term "storage" used in the embodiments of this application may refer to storage in one or more memories. The one or more memories may be independently configured or integrated into a compressor or decompressor, a processor, or a communication device. The one or more memories may also be partially independently configured and partially integrated into a decompressor, a processor, or a communication device. The memory may be any type of storage medium, and this application is not limited thereto.

[0388] Fifth, the "protocol" involved in the embodiments of the present application may refer to a standard protocol in the communication field, for example, it may include an LTE protocol, an NR protocol, and related protocols used in future communication systems. This application does not limit this.

[0389] Sixth, the names of the signaling involved in the embodiments of the present application are all functional descriptions, and the specific names are not limited. That is, the messages, information, etc. that appear in the embodiments of the present application can be called other names, and for the sake of brevity, they will not be repeated in this application. Figure 1 The following briefly introduces the scenarios in which the method for adjusting a neural network provided in the embodiments of the present application can be applied, and introduces the basic concepts that may be involved in the embodiments of the present application. The following describes in detail the method for adjusting a neural network provided in the embodiments of the present application in conjunction with the accompanying drawings.

[0390] It should be understood that the method for adjusting the neural network provided in the embodiments of the present application can be applied to systems communicating via multi-antenna technology, for example, Figure 1 The communication system 100 shown in FIG. The communication system may include at least one network device and at least one terminal device. The network device and the terminal device may communicate with each other via a multi-antenna technology.

[0391] It should also be understood that the embodiments shown below do not specifically limit the specific structure of the execution subject of the method provided in the embodiments of the present application. As long as it is possible to communicate according to the method provided in the embodiments of the present application by running a program that records the code of the method provided in the embodiments of the present application, for example, the execution subject of the method provided in the embodiments of the present application can be a terminal device or a network device, or a functional module in the terminal device or the network device that can call and execute the program.

[0392] Below, without loss of generality, the method for adjusting a neural network provided in an embodiment of the present application is described in detail by taking the interaction between a network device and a terminal device as an example.

[0393] Figure 5 This is a schematic flow chart of a method for adjusting a neural network provided in an embodiment of the present application. Figure 5 include Figure 5 (a)- Figure 5 (e), where Figure 5 (a) is a general process of the method for adjusting the neural network provided in this application. The information to be fed back can be any information that needs to be fed back via the communication system. Figure 5 (b) is a schematic flow chart of the method for adjusting the neural network provided in the present application applied to a scenario where channel information feedback is implemented based on an AI model. Figure 5 In the scenario shown in (b), the information to be fed back is channel information (e.g., CSI), Figure 5 (c) Figure 5 (a) Communication system diagram corresponding to the application scenario, Figure 5 (d) Figure 5 (b) AI framework diagram corresponding to the application scenario.

[0394] Adjusting the neural network involved in the embodiments of the present application includes changing the network architecture corresponding to a certain neural network, and / or changing the network weights corresponding to a certain neural network.

[0395] In addition, the neural network involved in the embodiments of the present application can be a trained neural network, a neural network defined by a network architecture and network weights, or a function related to an input and an output, wherein the input and output are related include the output being related to the current input, or the output being related to both current and historical inputs. The specific definition of the neural network in the embodiments of the present application is not limited, and reference can be made to the relevant definition of neural networks in the current or future communications field.

[0396] The following are Figure 5 (a) and Figure 5 (b) is used to illustrate the method flow.

[0397] Figure 5 The method flow shown in (a) can be applied to Figure 5 (c) shows the transmission framework. Figure 5 (c) shows that the first signal is processed at least by the first neural network to obtain the second signal, and the second signal is processed at least by the second neural network to obtain the fifth signal; or

[0398] The first signal is processed at least through the second neural network to obtain a second signal, and the second signal is processed at least through the first neural network to obtain a fifth signal. Figure 5 This situation is not shown in (c).

[0399] It should be understood that Figure 5 (c) is only an example and does not constitute any limitation on the protection of this application. The above process of processing the first signal to obtain the second signal may also be:

[0400] The first signal is processed by the first neural network and the first functional module to obtain the second signal, or the first signal is processed by the second neural network and the first functional module to obtain the second signal;

[0401] Similarly, the process of processing the second signal to obtain the fifth signal may also be:

[0402] The second signal is processed by the first neural network and the second functional module to obtain the fifth signal, or the second signal is processed by the second neural network and the second functional module to obtain the fifth signal.

[0403] In the embodiment of the present application, there is no limitation on the number of the first functional modules and the second functional modules included in the communication system.

[0404] It should be noted that the first functional module, second functional module, third functional module and fourth functional module involved in this application can be understood as AI units or non-AI units for processing signals, used to implement certain functions, and can also be called functional units or other names.

[0405] In theory, through Figure 5 The communication system shown in (c) can ensure that the difference or similarity between the fifth signal and the first signal is within an acceptable range, but the first neural network and the second neural network are obtained through offline training. If the first signal changes during the online information feedback process, it may cause the difference or similarity between the fifth signal and the first signal to be outside the acceptable range after the communication system based on at least the first neural network and the second neural network feeds back the first signal. In this case, it may be necessary to dynamically adjust the first neural network and the second neural network online, or to adjust other functional modules in the communication system. The present application mainly involves adjusting the first neural network and the second neural network so that when the communication system based on at least the updated first neural network and the second neural network feeds back information, the difference or similarity between the fifth signal and the first signal is within an acceptable range.

[0406] In addition, in a scenario where one of the first neural network and the second neural network needs to be updated or other functional modules in the system need to be updated, the signaling interaction between the first device and the second device and the steps performed by the first device and the second device are similar to updating the first neural network and the second neural network, and will not be repeated in this application.

[0407] For example, in a scenario where only the first neural network or the second neural network needs to be updated, it is sufficient to update the first neural network or the second neural network, and the information of the updated neural network can also be sent to the opposite device. The opposite device and / or the local device can choose to transmit information based on the updated neural network, or can also choose not to transmit information based on the updated neural network.

[0408] The following combination Figure 5 (a) Detailed description of how to update the first neural network and the second neural network online.

[0409] Figure 5 The method shown in (a) includes some or all of the following steps.

[0410] S510a: The first device or the second device determines a third neural network and a fourth neural network.

[0411] From the above, it can be seen that when only the first neural network or the second neural network needs to be updated, it is sufficient to update the first neural network or the second neural network, that is, the third neural network and / or the fourth neural network can be determined.

[0412] It should be noted that the embodiments of the present application are mainly described by taking the first device as one device and the second device as another device as an example. The first device and the second device can be any two devices in a communication system including multiple devices (for example, the first device and the second device can be a terminal device and a terminal device, a network device and a network device, or a network device and a terminal device). The signaling interaction between the other two devices included in the communication system and the steps performed by each device can refer to the signaling interaction between the first device and the second device and the steps performed by the first device and the second device in this application.

[0413] However, the present application does not limit the number of first devices or second devices that communicate. For example, the first device is a network device, the second device is a terminal device, the network device communicates with multiple terminal devices, and the embodiments of the present application are also applicable to the scenario where the network device and multiple terminal devices communicate and collaborate to process feedback information. The scenarios of collaborative processing of feedback information include:

[0414] The network device receives information fed back by multiple terminal devices and jointly processes the information fed back by the multiple terminal devices; or, at least one terminal device among the multiple terminal devices can obtain (e.g., through a D2D manner) information related to the information that needs to be fed back by at least one terminal device among the multiple terminal devices other than the at least one terminal device (e.g., information that needs to be fed back, or information processed according to the information that needs to be fed back), and jointly processes the obtained information (e.g., information that needs to be fed back locally and information related to the information that needs to be fed back by at least one terminal device) and then feeds it back to the network device; or, both the network device side and the terminal device side perform joint processing.

[0415] Optionally, when a network device communicates with multiple terminal devices and collaboratively processes feedback information, one or more first neural networks are set on the network device side, and multiple terminal devices are respectively set on multiple second neural networks (e.g., each terminal device is set with a second neural network). This example illustrates the first device or the second device determining the third neural network and the fourth neural network. The one or more first neural networks set on the network device side correspond to the multiple second neural networks, and it can be understood that the input information of the one or more first neural networks is related to the output information of the multiple second neural networks. It should be noted that when the network device communicates with other terminal devices (e.g., terminal devices other than the multiple terminal devices mentioned above) and collaboratively processes feedback information, other neural networks other than the one or more first neural networks mentioned above may be set in the network device.

[0416] For example, the network device may determine that one or more second neural networks among multiple second neural networks need to be updated, and / or the network device may determine that one or more first neural networks among one or more first neural networks need to be updated; for example, a terminal device among multiple terminal devices may also determine whether an update is required. In this application, when the network device and multiple terminal devices communicate and collaboratively process feedback information, the example of the network device determining whether an update is required is mainly used to illustrate.

[0417] In addition, when the network device communicates with multiple terminal devices and collaboratively processes feedback information, the network device may update a particular second neural network among the multiple second neural networks, or may update multiple second neural networks. When the network device updates multiple second neural networks, the network device may separately instruct the multiple terminal devices configured with the multiple second neural networks to receive the updated neural networks, or may uniformly instruct (e.g., by broadcasting) the multiple terminal devices configured with the multiple second neural networks to receive the updated neural networks.

[0418] Optionally, the one or more first neural networks provided on the network device side may be considered as a whole, referred to as an overall first neural network. The one or more first neural networks may be associated through functional modules. The one or more first neural networks may be associated through functional modules, where the output of at least one of the one or more first neural networks is related to the input of other first neural networks, such as Figure 5 As shown in (e), the first neural network #1 is associated with the first neural network #2;

[0419] Optionally, the one or more first neural networks provided on the network device side may also be a first neural network provided on the network device side.

[0420] Determining the third neural network may include determining a network structure corresponding to the third neural network, determining a network weight corresponding to the third neural network, determining a network structure corresponding to the third neural network and a network weight corresponding to the third neural network, determining a portion of the network weights corresponding to the third neural network (e.g., a portion of the network weights corresponding to the third neural network is the same as a portion of the network weights corresponding to the first neural network, i.e., determining the portion of the network weights corresponding to the third neural network that is different from the network weights corresponding to the first neural network), or determining a portion of the network structure corresponding to the third neural network.

[0421] Similarly, determining the fourth neural network is similar to determining the third neural network described above, and will not be repeated here.

[0422] Specifically, in the embodiment of the present application, the first device or the second device determining the third neural network and the fourth neural network can be understood as the first device or the second device updating the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively.

[0423] The first neural network corresponds to the third neural network and is used for the first device or the second device to perform first signal processing; the second neural network corresponds to the fourth neural network and is used for the first device or the second device to perform second signal processing. The first signal processing and the second signal processing are corresponding signal processing, for example, the first signal processing is encoding processing and the second signal processing is decoding processing, or the second signal processing is encoding processing and the first signal processing is decoding processing; for example, the first signal processing is modulation processing and the second signal processing is demodulation processing, or the second signal processing is modulation processing and the first signal processing is demodulation processing; for example, the first signal processing is compression processing and the second signal processing is decompression processing, or the second signal processing is compression processing and the first signal processing is decompression processing.

[0424] Exemplarily, the first signal processing may also be a signal processing process performed by a neural network determined by the first device or the second device based on information of the third neural network and which is not completely identical to the third neural network;

[0425] Exemplarily, the above-mentioned second signal processing may also be a signal processing process performed by a neural network that is not completely identical to the fourth neural network and is determined by the first device or the second device based on information of the fourth neural network.

[0426] It should be noted that the embodiments of the present application do not limit the application of the third neural network and the fourth neural network determined by the first device or the second device. For example, the third neural network and the fourth neural network may only be determined, but the first device and / or the second device do not adopt the determined third neural network and the fourth neural network.

[0427] Furthermore, the first neural network and the second neural network are understood as a pair of neural networks, performing first signal processing and second signal processing; the third neural network and the fourth neural network are understood as a pair of neural networks, performing first signal processing and second signal processing.

[0428] Optionally, the first neural network performing the first signal processing includes: processing the first signal to obtain a second signal;

[0429] Optionally, the second neural network performing the second signal processing includes: processing the second signal to obtain a fifth signal;

[0430] or,

[0431] Optionally, the second neural network performing the second signal processing includes: processing the first signal to obtain the second signal;

[0432] Optionally, the first neural network performing the first signal processing includes: processing the second signal to obtain a fifth signal.

[0433] Optionally, the third neural network performing the first signal processing includes: processing the third signal to obtain a fourth signal;

[0434] Optionally, the fourth neural network performing the second signal processing includes: processing the fourth signal to obtain a sixth signal;

[0435] or,

[0436] Optionally, the fourth neural network performing the second signal processing includes: processing the third signal to obtain a fourth signal;

[0437] Optionally, the third neural network performing the first signal processing includes: processing the fourth signal to obtain a sixth signal.

[0438] The above-mentioned third signal and first signal may be the same or different signals, for example, channel information at different times (e.g., CSI at different times); for example, it may be channel information at different frequencies (e.g., CSI at different frequencies); for example, it may be channel information at different times and different frequencies (e.g., CSI at different times and different frequencies).

[0439] The first neural network or the second neural network is used to process the first signal to obtain the second signal, including:

[0440] The first neural network or the second neural network processes the first signal to obtain the second signal; or

[0441] The first neural network or the second neural network and the first functional module process the first signal to obtain a second signal.

[0442] The first neural network or the second neural network is used to process the second signal to obtain the fifth signal, including:

[0443] The first neural network or the second neural network processes the second signal to obtain a fifth signal; or

[0444] The first neural network or the second neural network and the second functional module process the second signal to obtain a fifth signal.

[0445] The third neural network or the fourth neural network is used to process the third signal to obtain the fourth signal, including:

[0446] The third neural network or the fourth neural network processes the third signal to obtain a fourth signal; or

[0447] The third neural network or the fourth neural network and the third functional module process the third signal to obtain a fourth signal.

[0448] The third neural network or the fourth neural network is used to process the fourth signal to obtain the sixth signal, including:

[0449] The third neural network or the fourth neural network processes the fourth signal to obtain a sixth signal; or

[0450] The third neural network or the fourth neural network and the fourth functional module process the fourth signal to obtain a sixth signal.

[0451] The first functional module and the third functional module may be the same functional module, or the third functional module may be an updated module of the first functional module; the second functional module and the fourth functional module may be the same module, or the fourth functional module may be an updated module of the second functional module. When the first device is a terminal device, the second device may be a network device; when the first device is a network device, the second device may be a terminal device.

[0452] In one possible implementation, the difference between the first training signal related to the third signal and the second training signal related to the sixth signal satisfies the first condition. The first training signal related to the third signal indicates that, in the process of determining the third neural network and the fourth neural network, the first device or the second device that determines the third neural network and the fourth neural network simulates a signal similar to the third signal (e.g., the same size as the third signal and / or the same type as the third signal); and the second training signal related to the sixth signal indicates that, in the process of determining the third neural network and the fourth neural network, the first device or the second device that determines the third neural network and the fourth neural network simulates a signal similar to the sixth signal (e.g., the same size as the sixth signal and / or the same type as the sixth signal).

[0453] The difference involved in this application may be any one of a difference value, a mean square error value, a normalized mean square error value, or a mean absolute error, or the difference may also be other values ​​that can represent the degree of difference between the first training signal related to the third signal and the second training signal related to the sixth signal. The embodiment of this application does not limit the specific form of the difference;

[0454] Exemplarily, the above-mentioned first condition can be that the difference is less than or equal to the first preset value, or the absolute value of the difference is less than or equal to the first preset value, or the opposite of the difference is less than or equal to the first preset value, or a certain value determined based on the difference (such as, difference + Δ) is less than or equal to the first preset value. For the sake of convenience of description, the following description takes the difference being less than or equal to the first preset value as an example, but it should be understood that the present application is not limited to this.

[0455] In another possible implementation, the similarity between the first training signal related to the third signal and the second training signal related to the sixth signal satisfies the second condition.

[0456] The similarity involved in this application may be a correlation coefficient, or the similarity may also be other values ​​that can represent the degree of similarity between the first training signal related to the third signal and the second training signal related to the sixth signal. The embodiment of this application does not limit the specific form of the similarity.

[0457] Exemplarily, the above-mentioned second condition can be that the similarity is greater than or equal to the second preset value, or the opposite of the similarity is greater than or equal to the second preset value, or a certain value determined based on the similarity is greater than or equal to the second preset value. For the sake of convenience of description, the following description takes the similarity being greater than or equal to the second preset value as an example, but it should be understood that the present application is not limited to this.

[0458] The following describes several ways in which the first device determines to update the first neural network and the second neural network:

[0459] Method 1:

[0460] The first device periodically updates the first neural network and the second neural network. That is, in the following manner, after the time for periodically updating the first neural network and the second neural network arrives, the first device determines that the first neural network and the second neural network need to be updated.

[0461] The "period" involved in the embodiments of the present application can be understood as a time interval in units of time slots, subframes or frames, for example, a period can be 10 time slots or 10 subframes, etc.

[0462] In addition, "periodicity" refers to a time interval. When the time interval represented by a cycle is determined, the first neural network and the second neural network can be updated once every half a cycle, or once every quarter of a cycle. In the embodiments of the present application, "periodicity" is understood as updating at fixed time intervals, but the length of the time interval is not limited.

[0463] For example, the first device periodically updates the first and second neural networks based on a first preset period T1. If the first device updates the first and second neural networks at time t, then the first device updates the first and second neural networks at time t+T1. T1 may be predefined by the protocol, determined by negotiation between the first and second devices, or a period randomly selected by the first device. This application does not limit how the first device learns the first preset period.

[0464] Method 2:

[0465] The first device updates the first neural network and the second neural network aperiodically. That is, in the second mode, the first device can randomly determine the time to update the first neural network and the second neural network.

[0466] For example, the first device may update the first neural network and the second neural network at any time, and the time when the first device updates the first neural network and the second neural network currently in use does not have regularity.

[0467] Method 3:

[0468] The first device receives first instruction information sent by the second device, and the first instruction information indicates that the first neural network and the second neural network should be updated. In other words, in the third mode, after receiving the first instruction information, the first device determines the time to update the first neural network and the second neural network.

[0469] That is, in method three Figure 5 The method flow shown in (a) also includes S511a, the second device sends first indication information to the first device.

[0470] Exemplarily, the transmitting device sends the first indication information to the receiving device, including: the local device sends the first indication information to the receiving device on the uplink physical control channel (physical uplink control channel, PUCCH), the downlink physical control channel (physical downlink control channel, PDCCH), the physical uplink shared channel (physical uplink shared channel, PUSCH), the physical downlink shared channel (physical downlink shared channel, PDSCH), the uplink authorization UL grant or the random access channel (random access channel, RACH).

[0471] It should be understood that, since PUCCH and PDCCH resources are limited, sending the first indication information to the receiving end device on PUSCH or PDSCH can increase the flexibility of the solution.

[0472] Method 4:

[0473] When the first device determines that the first neural network and the second neural network need to be updated, the first device updates the first neural network and the second neural network.

[0474] For example, the first device may determine whether the first neural network and the second neural network need to be updated based on whether the difference or similarity between the first information and the second information satisfies a third condition. When the difference or similarity between the first information and the second information satisfies the third condition, the first device determines that the first neural network and the second neural network do not need to be updated; when the difference or similarity between the first information and the second information does not satisfy the third condition, the first device determines that the first neural network and the second neural network need to be updated.

[0475] Exemplarily, when a network device communicates with multiple terminal devices and collaboratively processes feedback information, the network device may determine whether one or more of the multiple second neural networks need to be updated based on the input of a simulated second neural network and the output of the first neural network. Alternatively, the network device may determine whether one or more of the multiple second neural networks need to be updated based on the combined input of multiple second neural networks and the output of the first neural network. If a single first neural network is configured on the network device, the output of the aforementioned first neural network is the output of the first neural network. If multiple first neural networks are configured on the network device, the output of the aforementioned first neural network is the output of the first neural network corresponding to the second neural network in the multiple first neural networks.

[0476] Among them, the first information can be understood as information related to the input information of the first neural network simulated by the first device or the second device (such as the first judgment signal related to the above-mentioned first signal), and the second information can be understood as information related to the output information of the second neural network simulated by the first device or the second device (such as the second judgment signal related to the above-mentioned fifth signal).

[0477] The first judgment signal related to the first signal indicates that in the process of judging whether the first neural network and the second neural network need to be updated, the first device or the second device executing the judgment simulates a signal similar to the first signal (for example, the same size as the first signal, and / or the same type as the first signal); the second judgment signal related to the fifth signal indicates that in the process of judging whether the first neural network and the second neural network need to be updated, the first device or the second device executing the judgment simulates a signal similar to the fifth signal (for example, the same size as the fifth signal, and / or the same type as the fifth signal).

[0478] Exemplarily, the above-mentioned third condition can be that the difference is less than or equal to the third preset value, or the absolute value of the difference is less than or equal to the third preset value, or a certain value determined based on the difference (such as, difference + Δ) is less than or equal to the third preset value. For the sake of convenience of description, the following example will be used to illustrate that the third condition is satisfied when the difference is less than or equal to the third preset value, and the third condition is not satisfied when the difference is greater than the third preset value, but it should be understood that the present application is not limited to this.

[0479] Exemplarily, the above-mentioned third condition can be that the similarity is greater than or equal to the fourth preset value, or it can be that a certain value determined based on the similarity is greater than or equal to the fourth preset value. For the convenience of description, the following example uses the similarity being greater than or equal to the fourth preset value as satisfying the third condition and the similarity being less than the fourth preset value as not satisfying the third condition, but it should be understood that the present application is not limited to this.

[0480] For example, for Figure 5 For the communication system shown in (c), the first signal is processed by the first neural network and then sent to the receiving device. After the receiving device receives the information fed back over the air interface, the signal is processed by the second neural network to obtain the fifth signal. In this case, the first information is the input information (first signal) of the first neural network simulated by the first device or the second device, and the second information is the output information (fifth signal) of the second neural network simulated by the first device or the second device. The output information of the first neural network is the input information of the second neural network.

[0481] For example, for Figure 5 For the communication system shown in (c), the first signal is processed by the first functional module and then processed by the first neural network and sent to the receiving device side. After the receiving device receives the information fed back by the air interface, it is processed by the second functional module and then processed by the second neural network to obtain the fifth signal. In this case: the first information is the input information (first signal) of the first functional module simulated by the first device or the second device, and the second information is the output information (fifth signal) of the second neural network simulated by the first device or the second device. The output information of the first neural network, after being processed by the second functional module, is the input information of the second neural network. Among them, the number of first functional modules and second functional modules can be at least one.

[0482] For example, for Figure 5 For the communication system shown in (c), the first signal is processed by the first functional module and then by the first neural network and sent to the receiving device side. After the receiving device receives the information fed back by the air interface, it is processed by the second neural network and then by the second functional module to obtain the fifth signal. In this case: the first information is the input information (first signal) of the first functional module simulated by the first device or the second device, and the second information is the output information (fifth signal) of the second functional module simulated by the first device or the second device. The output information of the first neural network and the information after being processed by the second neural network are the input information of the second functional module, wherein the number of the first functional module and the second functional module can be at least one.

[0483] For example, for Figure 5For the communication system shown in (c), the first signal is processed by the first neural network and then processed by the first functional module and sent to the receiving device side. After the receiving device receives the information fed back by the air interface, it is processed by the second functional module and then processed by the second neural network to obtain the fifth signal. In this case: the first information is the input information (first signal) of the first neural network simulated by the first device or the second device, and the second information is the output information (fifth signal) of the second neural network simulated by the first device or the second device. The output information of the first neural network, after being processed by the first functional module and the second functional module, is the input information of the second neural network, wherein the number of the first functional module and the second functional module can be at least one.

[0484] For example, for Figure 5 For the communication system shown in (c), the first signal is processed by the first neural network and then processed by the first functional module and sent to the receiving device side. After the receiving device receives the information fed back by the air interface, it is processed by the second neural network and then processed by the second functional module to obtain the fifth signal. In this case: the first information is the input information (first signal) of the first neural network simulated by the first device or the second device, and the second information is the output information (fifth signal) of the second functional module simulated by the first device or the second device. The output information of the first neural network, after being processed by the first functional module, is the input information of the second neural network, wherein the number of the first functional module and the second functional module can be at least one.

[0485] For example, for Figure 5 For the communication system shown in (c), the first signal is processed by the first neural network and then sent to the receiving device side. After the receiving device receives the air interface feedback information, it is processed by the second neural network based on the auxiliary information and the received feedback information to obtain the fifth signal. In this case: the first information is the input information of the first neural network simulated by the first device or the second device (the first signal), and the second information is the output information of the second neural network simulated by the first device or the second device (the fifth signal).

[0486] The above describes possible forms of the first information and the second information. For ease of explanation in the embodiments of the present application, the first information is used as the input information of the first neural network simulated by the first device or the second device, which is also the first signal, and the second information is used as the output information of the second neural network simulated by the first device or the second device, which is also the fifth signal. The case where the first information and the second information are the output / input information of other functional modules is similar to the case where the first information is the input information of the first neural network and the second information is the output information of the second neural network, and will not be described one by one.

[0487] When the difference between the first information and the second information is greater than or equal to the third preset value, or when the similarity between the first information and the second information is less than or equal to the fourth preset value, the first device determines that the first neural network and the second neural network need to be updated; when the difference between the first information and the second information is less than the third preset value, or when the similarity between the first information and the second information is greater than the fourth preset value, the first device determines that the first neural network and the second neural network do not need to be updated.

[0488] Alternatively, when the difference between the first information and the second information is greater than a third preset value, or when the similarity between the first information and the second information is less than a fourth preset value, the first device determines that the first neural network and the second neural network need to be updated; when the difference between the first information and the second information is less than or equal to the third preset value, or when the similarity between the first information and the second information is greater than or equal to a fourth preset value, the first device determines that the first neural network and the second neural network do not need to be updated.

[0489] It should be noted that the above-described determination of whether the first and second neural networks need to be updated based on the magnitude relationship between the difference or similarity between the first and second information and the corresponding preset value is merely an example and does not limit the scope of protection of this application. For example, the first device may further determine whether the first and second neural networks need to be updated based on the further difference or similarity between the difference or similarity between the first and second information and the corresponding preset value.

[0490] It can be understood that the difference or similarity between the first information and the second information can reflect the degree of matching between the first neural network and the second neural network and the current communication environment. For example, the difference between the first information and the second information can reflect the degree of difference between the first information and the second information obtained after the first information is processed by the first neural network and the second neural network, thereby deducing the recovery performance of the first neural network and the second neural network; for example, the similarity between the first information and the second information can reflect the degree of similarity between the first information and the second information obtained after the first information is processed by the first neural network and the second neural network, thereby deducing the recovery performance of the first neural network and the second neural network.

[0491] If the difference or similarity between the first information and the second information is within an acceptable range, it is considered that the recovery performance of the first neural network and the second neural network is good, and information transmission can continue based on the communication system composed of the first neural network and the second neural network; if the difference or similarity between the first information and the second information is not within an acceptable range, it is considered that the recovery performance of the first neural network and the second neural network is poor, and the first neural network and the second neural network need to be updated.

[0492] That is, in the fourth way Figure 5 The method flow shown in (a) further includes S512a, where the first device determines that the first neural network and the second neural network need to be updated. This embodiment of the present application primarily involves online updates of the first neural network and the second neural network, and will not be further described regarding the case where the first device determines that the first neural network and the second neural network do not need to be updated.

[0493] Furthermore, in the fourth mode, the first device may determine whether the first neural network and the second neural network need to be updated based on one of the following possible triggering modes:

[0494] Possibility 1: The first device performs the above determination periodically.

[0495] For example, the first device periodically determines whether the first neural network and the second neural network need to be updated based on the second preset period T2, where T2 can be predefined by the protocol, or determined by the first device and the second device through negotiation, or can be a period randomly selected by the first device. This application does not limit how the first device obtains the second preset period.

[0496] Possibility two: the first device performs the above determination non-periodically.

[0497] For example, the first device can make the above judgment at any time.

[0498] Possibility three: the first device may make the above judgment based on the second indication information sent by the second device. The second indication information may be called trigger signaling, which is used to trigger the first device to make the above judgment. Figure 5 The method flow shown in (a) may further include S513a: the second device sends second indication information to the first device.

[0499] For example, the first device receives second indication information sent by the second device, where the second indication information is used to instruct the first device to determine whether the first neural network and the second neural network need to be updated.

[0500] Exemplarily, the transmitting end device sending the second indication information to the receiving end device includes: the local device sending the second indication information to the receiving end device on PUCCH, PDCCH, PUSCH, PDSCH, UL grant or RACH.

[0501] It should be understood that, since PUCCH and PDCCH resources are limited, sending the second indication information to the receiving end device on the PUSCH or PDSCH can increase the flexibility of the solution.

[0502] Possibility four: the first device may make the above judgment based on its own trigger information.

[0503] For example, when the first device senses that its own transmission performance has decreased (eg, throughput performance has decreased), the first device may trigger itself to perform the above-mentioned judgment.

[0504] The way in which the second device determines to update the first neural network and the second neural network is similar to that of the first device, except that when the second device is triggered by the first device to update, the first indication information is sent by the first device to the second device; similarly, when the second device is triggered by the first device to make a judgment, the second indication information is sent by the first device to the second device. Other ways of determining the update and making the judgment are the same as those of the first device, except that the execution subject is changed from the first device to the second device, which will not be repeated here.

[0505] The above describes how the first device or the second device determines that the first neural network and the second neural network need to be updated.

[0506] Specifically, under the premise that the first device or the second device determines that the first neural network and the second neural network need to be updated, or when the first device or the second device directly updates the first neural network and the second neural network periodically or aperiodically and needs to be updated, the first device or the second device updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network.

[0507] Here are some ways to update the first neural network and the second neural network to obtain the third neural network and the fourth neural network:

[0508] Method 1:

[0509] Based on the communication system data, a suitable third neural network and a fourth neural network are selected from the first set, wherein the third neural network is the updated first neural network and the fourth neural network is the updated second neural network, and the communication system data includes system data collected by the first device or the second device during a recent communication process, for example, at least one of channel characteristics, signal-to-noise ratio, downlink channel information, uplink channel information and hybrid automatic repeat request HARQ information.

[0510] For example, the first device or the second device locally stores three pairs of neural networks: neural network pair #1, neural network pair #2, and neural network pair #3. The neural network pairs shown in this application include a neural network set in the first device and a neural network in the second device.

[0511] The above-mentioned neural network pair #1 (e.g., the first neural network and the second neural network) can be applied to a communication system in which the communication system data is communication system data #1, the neural network pair #2 (e.g., the third neural network and the fourth neural network) can be applied to a communication system in which the communication system data is communication system data #2, and the neural network pair #3 (e.g., the ninth neural network and the tenth neural network) can be applied to a communication system in which the communication system data is communication system data #3. When the first device or the second device determines that the neural network pair #1 (e.g., the first neural network and the second neural network) needs to be updated, and the currently obtained communication system data is communication system data #2, the first device or the second device can update the neural network pair #1 in the communication system to the neural network pair #2, that is, the first neural network is updated to the third neural network and the second neural network is updated to the fourth neural network.

[0512] It should be noted that, in the first embodiment, the first device or the second device stores multiple pairs of usable neural network pairs (referred to as the first set). When the neural network pair in the current communication system is inappropriate (for example, the difference or similarity between the first information and the second information is not within an acceptable range), a suitable neural network pair can be selected from the multiple pairs of usable neural network pairs.

[0513] Method 2:

[0514] The first neural network and the second neural network are adjusted based on the communication system data so that the adjusted first neural network and the second neural network can be used in the communication system. The adjusted first neural network can be referred to as a third neural network, and the adjusted second neural network can be referred to as a fourth neural network.

[0515] For example, the first neural network corresponds to network architecture #1 and network weight #1, and the second neural network corresponds to network architecture #2 and network weight #2. When the first device or the second device determines that the first neural network and the second neural network need to be updated, the first device or the second device determines based on current communication system data that when the network weight corresponding to the first neural network is updated to network weight #3 and the network weight corresponding to the second neural network is updated to network weight #4, the updated first neural network and the second neural network can be used for information feedback.

[0516] It should be noted that the adjustment of the neural network involved in the second method means that the network architecture corresponding to the neural network does not change, only the network weights in the network architecture change. It can be understood as changing the network weights corresponding to a certain neural network on the basis of the neural network, so as to achieve the purpose of updating the neural network.

[0517] Method 3:

[0518] A ninth neural network and a tenth neural network are selected from the second set based on the communication system data, and the ninth neural network and the tenth neural network are further adjusted to obtain a suitable third neural network and a fourth neural network.

[0519] For example, the first device or the second device locally stores three pairs of neural networks: neural network pair #1, neural network pair #2, and neural network pair #3. Among them, neural network pair #1 (e.g., the first neural network and the second neural network) can be applied to the communication system when the communication system data is communication system data #1, neural network pair #2 (e.g., the seventh neural network and the eighth neural network) can be applied to the communication system when the communication system data is communication system data #2, and neural network pair #3 (e.g., the ninth neural network and the tenth neural network) can be applied to the communication system when the communication system data is communication system data #3.

[0520] When the first device or the second device determines that the neural network pair #1 (eg, the first neural network and the second neural network) needs to be updated, and the currently obtained communication system data is communication system data #4.

[0521] The first device or the second device can first select a neural network pair #3 from the second set, that is, the first neural network is updated to the ninth neural network and the second neural network is updated to the tenth neural network; then, the first device or the second device determines based on the communication system data #4 that when the network weight corresponding to the ninth neural network is updated to the network weight #5 and the network weight corresponding to the tenth neural network is updated to the network weight #6, the ninth neural network and the tenth neural network after the network weights are updated can be applied to the communication system.

[0522] It should be understood that the above-mentioned methods 1 to 3 are only examples of how to update the third neural network and the fourth neural network, and do not constitute any limitation on the scope of protection of this application. The above-mentioned update process can also be completed by other methods. For example, the first neural network and the second neural network can be adjusted multiple times to obtain a suitable third neural network and a fourth neural network. It should be noted that after the first device or the second device executes the above-mentioned steps of updating the first neural network and the second neural network, if it is unable to select and / or adjust to a suitable third neural network and a fourth neural network, the first device or the second device can choose to fall back to the non-AI mode by means of information feedback; in other words, even if the first device or the second device executes the above-mentioned steps of updating the first neural network and the second neural network, and selects and / or adjusts to a suitable third neural network and a fourth neural network, the first device or the second device can also choose to fall back to the non-AI mode by means of information feedback.

[0523] Among them, the non-AI mode includes a mode that does not implement information transmission based on a neural network. For example, when the first neural network and the second neural network are compressors and decompressors, the non-AI mode can be understood as using traditional compression and decompression methods to implement information transmission between the first device and the second device; for another example, when the first neural network and the second neural network are modulators and demodulators, the non-AI mode can be understood as using traditional modulation and demodulation methods to implement information transmission between the first device and the second device; for another example, when the first neural network and the second neural network are encoders and decoders, the non-AI mode can be understood as using traditional encoding and decoding methods to implement information transmission between the first device and the second device.

[0524] Exemplarily, when the first device or the second device determines that the first neural network and the second neural network need to be updated, and before executing the above-mentioned steps of updating the first neural network and the second neural network, the method for adjusting the neural network provided in the embodiment of the present application, in order to ensure that information is no longer transmitted based on the communication system composed of the inappropriate first neural network and the second neural network, the first device and the second device can first fall back to the non-AI mode for information transmission, thereby avoiding affecting the timeliness of information transmission. In this application, the non-AI mode can be referred to as the second mode, and the AI ​​mode can be referred to as the first mode, that is, Figure 5 The method flow shown in (a) further includes S522a, falling back to the non-AI mode.

[0525] As an example and not a limitation, in the embodiments of the present application, the first device or the second device determines to fall back to the non-AI mode for information transmission, including the following two possibilities:

[0526] Possibility 1: Based on protocol pre-definition, that is, the protocol stipulates that when the first neural network and the second neural network need to be updated, the information feedback mode can fall back to the non-AI mode.

[0527] Possibility 2: Based on the instruction of the opposite device, the first device sends a second notification message to the second device, or the second device sends a second notification message to the first device. Figure 5 The method flow shown in (a) further includes S521a, sending a second notification message. The second notification message is used to instruct the opposite device to fall back to a non-AI mode for information feedback.

[0528] Exemplarily, the sending of the second notification message by the opposite device to the local device includes: the local device sending the second notification message to the receiving opposite device on PUCCH, PDCCH, PUSCH, PDSCH, UL grant or RACH.

[0529] It should be understood that, since PUCCH and PDCCH resources are limited, sending the second notification message to the local device on the PUSCH or PDSCH can increase the flexibility of the solution.

[0530] The opposite-end device involved in this application includes a first device or a second device. When the first device is a local device, the second device is a opposite-end device; when the second device is a local device, the first device is a opposite-end device.

[0531] For example, the first device determines that the first neural network and the second neural network need to be updated, and the first device sends the second notification message to the second device; for another example, the second device determines that the first neural network and the second neural network need to be updated, and the second device sends the second notification message to the first device; for another example, the first device determines that the first neural network and the second neural network need to be updated, and the second device sends the second notification message to the first device; for another example, the second device determines that the first neural network and the second neural network need to be updated, and the first device sends the second notification message to the second device.

[0532] As a possible implementation manner, the local device falling back to the non-AI mode for information transmission based on the instruction of the opposite device includes: the local device switching the information transmission mode from the AI ​​mode to the non-AI mode.

[0533] It should be noted that, in the embodiment of the present application, the timing for the local device to switch the information transmission mode is not limited, and can be any time after receiving the instruction from the opposite device.

[0534] For example, the local device receives an instruction from the opposite device and starts switching the information transmission mode from AI mode to non-AI mode within a fourth preset time period after the fourth starting moment or after the fourth preset time period, where the fourth starting moment includes the moment when the local device receives the second notification message from the opposite device.

[0535] As another possible implementation manner, the above-mentioned step of the opposite-end device falling back to the non-AI mode for information transmission includes: the opposite-end device switching the information transmission mode from the AI ​​mode to the non-AI mode.

[0536] It should be noted that, in the embodiment of the present application, the timing for the above-mentioned opposite-end device to switch the information transmission mode is not limited and can be any time.

[0537] For example, the opposite device starts to switch the information transmission mode from AI mode to non-AI mode within the fourth preset time period after the fourth starting moment or after the fourth preset time period, where the fourth starting moment includes the moment when the opposite device sends the second notification message to the local device.

[0538] As a possible implementation method, under AI mode Figure 5 The system shown in (c) includes other neural networks (e.g., a seventh neural network and an eighth neural network) in addition to the first neural network and the second neural network.

[0539] For example, the first and second neural networks are used for channel coding and decoding, and the seventh and eighth neural networks are used for modulation and demodulation. When the local device and the remote device in the communication system decide to return to non-AI mode for information transmission, it can be understood that all different information processing processes included in the communication system return to non-AI mode, that is, the channel coding and decoding and modulation and demodulation described above are all performed in non-AI mode.

[0540] Furthermore, after the first device or the second device completes the above-mentioned update step, the first device or the second device needs to send the obtained information of the third neural network and / or the fourth neural network to the opposite device. Figure 5 The method flow shown in (a) also includes S520a, sending information of the third neural network and / or information of the fourth neural network.

[0541] Exemplarily, when the execution subject of the above-mentioned S510a is the first device, that is, the first device sends the information of the third neural network and / or the information of the fourth neural network to the second device;

[0542] Exemplarily, when the execution subject of the above-mentioned S510a is the second device, the second device sends the information of the third neural network and / or the information of the fourth neural network to the first device.

[0543] Specifically, the content sent by S520a includes the following two possibilities:

[0544] Possibility 1: The peer device needs to configure a third neural network, i.e., S520a, sends information of the third neural network, or sends information of the third neural network and information of the fourth neural network;

[0545] Possibility two: the opposite end device needs to configure the fourth neural network, that is, S520a, sends information of the fourth neural network, or sends information of the third neural network and information of the fourth neural network.

[0546] It should be noted that, in the embodiment of the present application, the first device or the second device may send information about the third neural network or the fourth neural network required by the opposite device to the opposite device, or,

[0547] The first device or the second device can send information about the third neural network and the fourth neural network to the opposite device, and then the opposite device can determine whether the third neural network and the fourth neural network need to be updated based on the third neural network and the fourth neural network.

[0548] Specifically, with respect to the three different methods of updating the first neural network and the second neural network to obtain the third neural network and the fourth neural network, the specific contents of the information of the third neural network and the information of the fourth neural network are respectively introduced. In order to correspond to the above-mentioned update methods, the following is also divided into three methods for explanation:

[0549] Method 1:

[0550] Corresponding to the above-mentioned update method 1, the third neural network and the fourth neural network are selected from the first set. In this method, the specific content of the information of the third neural network and the information of the fourth neural network includes the following possibilities:

[0551] Possibility one: when both the first device and the second device locally store a third neural network, and the index of the third neural network in the first device is consistent with the index of the third neural network in the second device, then the information of the third neural network may be the index of the third neural network; similarly, when both the first device and the second device locally store a fourth neural network, and the index of the fourth neural network in the first device is consistent with the index of the fourth neural network in the second device, then the information of the fourth neural network may be the index of the fourth neural network.

[0552] Optionally, the index of the third neural network and the index of the fourth neural network are the same. In this case, the information of the third neural network and the information of the fourth neural network can be one index, that is, when the information of the third neural network and the information of the fourth neural network need to be sent, it is sufficient to send this one index.

[0553] It should be noted that the index involved in the embodiments of the present application can be understood as an identifier (identify, ID), that is, it is used to identify a certain neural network.

[0554] Possibility 2: When the third neural network can be represented by the first network weights and the first network structure, the information of the third neural network can be the first network weights and the first network structure; similarly, when the fourth neural network can be represented by the second network weights and the second network structure, the information of the fourth neural network can be the second network weights and the second network structure.

[0555] Exemplarily, the neural network involved in the embodiments of the present application includes at least one of neural networks such as convolutional neural networks (CNN), deep neural networks (DNN) and recurrent neural networks (RNN), among which DNN can also be called fully connected neural networks (FC), and is described as DNN / FC below.

[0556] It is understood that the neural networks involved in the embodiments of this application may include neural networks such as CNN, DNN / FC, or RNN, and may also include neural networks composed of neural networks such as CNN, DNN / FC, and RNN. For example, the neural networks involved in the embodiments of this application are composed of DNN / FC and RNN.

[0557] As an example and not a limitation, the network structure indication information in the information of the neural network involved in the embodiments of the present application may include indication information of the neural network type and indication information of the neural network structure.

[0558] For example, when the neural network includes CNN, the indication information of the neural network type in the network structure indication information is used to indicate that the neural network is CNN; the indication information of the neural network structure in the network structure indication information includes at least one of the information such as the number of neural network layers, the number of convolution kernels in each layer, the scale of each convolution kernel, and the activation function indication of each layer.

[0559] Among them, when the scale of the convolution kernels of each layer of the neural network in the neural network is the same, the scale information of each convolution kernel included in the indication information of the above-mentioned neural network structure can be understood as including the scale information of a convolution kernel corresponding to each layer of the neural network or the scale information of a convolution kernel corresponding to the entire neural network; similarly, when the activation function indication information of each layer of the neural network in the neural network is the same, the activation function indication information of each layer included in the indication information of the above-mentioned neural network structure can be understood as including an activation function indication information corresponding to the entire neural network.

[0560] For example, when the neural network includes DNN / FC, the indication information of the neural network type in the network structure indication information is used to indicate that the neural network is DNN / FC; the indication information of the neural network structure in the network structure indication information includes at least one of the information such as the number of neural network layers, the number of nodes in each layer, and the activation function indication of each layer.

[0561] Among them, when the activation function indication information of each layer of the neural network in the neural network is the same, the activation function indication information of each layer included in the indication information of the above-mentioned neural network structure can be understood as including an activation function indication information corresponding to the entire neural network.

[0562] For another example, when the neural network includes an RNN, the indication information of the neural network type in the network structure indication information is used to indicate that the neural network is an RNN; the indication information of the neural network structure in the network structure indication information includes at least one of the information such as the position of the RNN loop, the activation function indication, etc.

[0563] It should be noted that the neural network involved in this application may include multiple CNNs, and the corresponding network structure indication information when the neural network includes CNNs may include multiple sets; similarly, the neural network involved in this application may include multiple DNNs / FCs, and the corresponding network structure indication information when the neural network includes DNNs / FCs may include multiple sets; the neural network involved in this application may include multiple RNNs, and the corresponding network structure indication information when the neural network includes RNNs may include multiple sets. In addition, the neural network involved in this application may also include any combination of the above-mentioned CNNs, DNNs / FCs, and RNNs, and in this case, the network structure indication information may be a combination of the above-mentioned network structure indication information.

[0564] As an example and not a limitation, the network weight indication information in the neural network information involved in the embodiments of the present application may include weight interpretation rule indication information and weight value arrangement information. The weight interpretation rule indication information is optional.

[0565] The weight interpretation rule indication information may include at least one of the following information: the number of weights, the number of quantization bits for each weight, an indication of a weight position sorting rule, and the like.

[0566] It should be understood that the above is only an example of the possible forms of the network structure indication information and the possible specific forms of the indication information of the network weights corresponding to the neural network when the neural network is of different types, and does not constitute any limitation on the scope of protection of this application. When the neural network is of other types, the network structure indication information and the network weight indication information can change accordingly. In the embodiment of this application, it is only limited that the network weights and network architecture corresponding to the neural network can determine the neural network, and there is no limitation on the specific forms of expression of the network weights and network architecture.

[0567] Possibility three: when the first device and the second device both locally store the first network structure corresponding to the third neural network, and the index of the first network structure in the first device is consistent with the index of the first network structure in the second device, the third neural network can be represented by the first network weight and the first network structure, then the information of the third neural network can be the index of the first network structure and the first network weight; similarly, when the first device and the second device both locally store the second network structure corresponding to the fourth neural network, and the index of the second network structure in the first device is consistent with the index of the second network structure in the second device, the fourth neural network can be represented by the second network weight and the second network structure, then the information of the fourth neural network can be the index of the second network structure and the second network weight.

[0568] Optionally, the index of the first network structure and the index of the second network structure are the same. In this case, the index of the first network structure and the index of the second network structure may be one index.

[0569] Possibility four: when the first device and the second device both locally store the first network structure corresponding to the third neural network, and the index of the first network structure in the first device is consistent with the index of the first network structure in the second device, the third neural network can be represented by the first network weight and the first network structure. In addition, when the first device and the second device both locally store the first network weight corresponding to the third neural network, and the index of the first network weight in the first device is consistent with the index of the first network weight in the second device, the information of the third neural network can be the index of the first network structure and the index of the first network weight; similarly, when the first device and the second device both locally store the second network structure corresponding to the fourth neural network, and the index of the second network structure in the first device is consistent with the index of the second network structure in the second device, the fourth neural network can be represented by the second network weight and the second network structure. In addition, when the first device and the second device both locally store the second network weight corresponding to the fourth neural network, and the index of the second network weight in the first device is consistent with the index of the second network weight in the second device, the information of the fourth neural network can be the index of the second network structure and the index of the second network weight.

[0570] Optionally, the index of the first network structure and the index of the second network structure are the same, in which case the index of the first network structure and the index of the second network structure can be one index; the index of the first network weight and the index of the second network weight are the same, in which case the index of the first network weight and the index of the second network weight can be one index.

[0571] Possibility five: when both the first device and the second device locally store only the first network structure corresponding to the third neural network, the third neural network can be represented by the first network weights and the first network structure, and the information of the third neural network can be the first network weights; similarly, when both the first device and the second device locally store only the second network structure corresponding to the fourth neural network, the fourth neural network can be represented by the second network weights and the second network structure, and the information of the third neural network can be the first network weights.

[0572] Possibility six: when both the first device and the second device locally store only the first network structure corresponding to the third neural network, the third neural network can be represented by the first network weight and the first network structure. In addition, the first device and the second device locally store the first network weight corresponding to the third neural network respectively, and the index of the first network weight in the first device is consistent with the index of the first network weight in the second device, then the information of the third neural network can be the index of the first network weight; similarly, when both the first device and the second device locally store only the second network structure corresponding to the fourth neural network, the fourth neural network can be represented by the second network weight and the second network structure. In addition, the first device and the second device locally store the second network weight corresponding to the fourth neural network respectively, and the index of the second network weight in the first device is consistent with the index of the second network weight in the second device, then the information of the fourth neural network can be the index of the second network weight.

[0573] Optionally, the index of the first network weight and the index of the second network weight are the same. In this case, the index of the first network weight and the index of the second network weight are just one index.

[0574] Possibility seven: when the network structures corresponding to the first neural network and the third neural network are consistent, and the change between the third network weight corresponding to the first neural network and the first network weight corresponding to the third neural network is the change of the first network weight, then the information of the third neural network may be the change of the first network weight; similarly, when the network structures corresponding to the second neural network and the fourth neural network are consistent, and the change between the fourth network weight corresponding to the second neural network and the second network weight corresponding to the fourth neural network is the change of the second network weight, then the information of the fourth neural network may be the change of the second network weight.

[0575] As an example and not a limitation, the network weight change indication information in the information of the neural network involved in the embodiments of the present application may include one or more of weight position information, weight interpretation rule indication information, weight value change arrangement information and weight value arrangement information.

[0576] For example, the weight change indication information includes weight interpretation rule indication information, weight position information, and weight numerical arrangement information; for another example, the weight change indication information includes weight interpretation rule indication information, weight position information, and weight numerical change arrangement information; for another example, the weight change indication information includes weight interpretation rule indication information and weight numerical change arrangement information; for another example, the weight change indication information includes weight position information and weight numerical arrangement information; for another example, the weight change indication information includes weight position information and weight numerical change arrangement information; for another example, the weight change indication information includes weight numerical change arrangement information.

[0577] Among them, the weight interpretation rule indication information is optional, and the order of the three types of information, namely the set of weight positions, the weight interpretation rule indication information and the arrangement information of weight values, can be combined arbitrarily, which will not be repeated here.

[0578] It should be understood that the above-mentioned set of weight positions includes at least one weight position. The weights involved in the embodiments of the present application include the following granularities:

[0579] The transfer weights connecting two nodes (e.g., the weights between different layers and different nodes in a DNN / FC neural network), the convolution kernel weights (e.g., the weights corresponding to different convolution kernels in a CNN neural network), and the transfer weights of RNN rings (e.g., the weights corresponding to different RNN rings in an RNN neural network).

[0580] Furthermore, different weight granularities correspond to different weight position data structures, which include the following possibilities:

[0581] Possible 1. Weight layer sequence number and weight sequence number within the layer (for example, the different weights included in the DNN / FC neural network can be determined by the weight layer sequence number and the weight sequence number within the layer). The difference between the weight layer and the network layer is that the network layer refers to a network layer composed of nodes, and the weight layer refers to a weight layer composed of transfer weights between different nodes between two network layers;

[0582] Possibly two, the weight layer number and the convolution kernel number within the layer (e.g., the different weights included in a CNN neural network can be determined by the weight layer number and the convolution kernel number within the layer);

[0583] Possibility three: RNN loop start node position and RNN loop end node position (eg, different weights included in the RNN neural network can be determined by the RNN loop start node position and the RNN loop end node position).

[0584] It can be seen from the above-mentioned weight position data structure that it involves the position of the node (for example, the weight position data structure in the RNN neural network involves the position of the starting node of the RNN loop and the position of the ending node of the RNN loop), so the data structure of the node position is also given in the embodiment of the present application: the node layer sequence number and the node sequence number within the node layer, that is, the position of a node can be determined by the sequence number of the node layer in which the node is located and the node sequence number of the node in the node layer.

[0585] The weight interpretation rule indication information may include at least one of the following information: the number of weights, the number of quantization bits for each weight, an indication of a weight position sorting rule, and the like.

[0586] The relative order between the multiple weight positions included in the above set of weight positions can be determined by the weight position sorting rule indication, or can also be predefined by the protocol. The following briefly describes how to determine the relative order between the multiple weight positions based on the weight position sorting rule indication:

[0587] For DNN / FC neural networks, first traverse the weight sequence number within the layer (from small to large or from large to small), and then traverse the weight layer sequence number (from small to large or from large to small);

[0588] For the CNN neural network, first traverse the convolution kernel sequence number within the layer (from small to large or from large to small), and then traverse the weight layer sequence number (from small to large or from large to small);

[0589] For the RNN neural network, first traverse the end node position of the RNN loop (generally, the RNN loop is placed on the DNN / FC, first traverse the node sequence number in the layer from small to large or from large to small, then traverse the node layer sequence number from small to large or from large to small), then traverse the starting node position of the RNN loop (first traverse the node sequence number in the layer from small to large or from large to small, then traverse the node layer sequence number from small to large or from large to small); or,

[0590] For the RNN neural network, first traverse the starting node position of the RNN loop (generally, the RNN loop is placed on the DNN / FC, first traverse the node sequence numbers within the layer from small to large or from large to small, then traverse the node layer sequence numbers from small to large or from large to small), and then traverse the ending node position of the RNN loop (first traverse the node sequence numbers within the layer from small to large or from large to small, then traverse the node layer sequence numbers from small to large or from large to small).

[0591] As an example and not a limitation, the network weight change indication information in the neural network information involved in the embodiments of the present application may include the weight position and the weight change at the weight position.

[0592] As an example and not a limitation, the network weight change indication information in the neural network information involved in the embodiments of the present application may include position information where no weight exists and the weight change at all positions.

[0593] It should be understood that the above is only an example to illustrate the possible specific forms of the network weight change indication information corresponding to the neural network, and does not constitute any limitation on the scope of protection of this application. In the embodiments of this application, it is only limited that the weight change indication information can determine the neural network, and the specific form of the network weight change indication information is not limited.

[0594] It should be noted that the seven possible specific contents of the information of the third neural network and the information of the fourth neural network listed in the above method one are only examples and do not constitute any limitation to the scope of protection of this application. Simple deformations of the above possible one to possible seven and other possible contents of the information of the third neural network and the information of the fourth neural network are also within the scope of protection of this application. For example, the information of the third neural network can be the network structure and network weight index corresponding to the third neural network, etc., and no further examples will be given here.

[0595] In addition, method one mainly describes the sending of information of the fourth neural network corresponding to the above-mentioned update method one, or sending information of the third neural network and information of the fourth neural network. However, this application does not limit the sending of information of the fourth neural network shown in method one, or the sending of information of the third neural network and information of the fourth neural network cannot be applied to the scenarios shown in the above-mentioned update methods two and / or three. That is to say, when the update method is as shown in the above-mentioned update methods two and / or three, the method of sending information of the fourth neural network, or sending information of the third neural network and information of the fourth neural network can also choose the method described in method one.

[0596] Method 2:

[0597] Corresponding to the second updating method described above, the first and second neural networks are adjusted to obtain suitable third and fourth neural networks. To distinguish them from the third and fourth neural networks described in the first updating method, the suitable third neural network obtained by adjustment in the second updating method may also be referred to as the adjusted first neural network, and the suitable fourth neural network obtained by adjustment may also be referred to as the adjusted second neural network.

[0598] In this manner, the specific contents of the information of the third neural network and the information of the fourth neural network may include the following possibilities:

[0599] Possibility one: when both the first device and the second device locally store only the third network structure corresponding to the first neural network, the adjusted first neural network can be represented by the third network structure and the third network weight, and the information of the third neural network can be the third network weight; similarly, when both the first device and the second device locally store only the fourth network structure corresponding to the second neural network, the adjusted second neural network can be represented by the fourth network structure and the fourth network weight, and the information of the fourth neural network can be the fourth network weight.

[0600] Possibility two: when both the first device and the second device locally store only the third network structure corresponding to the first neural network, the difference between the adjusted first neural network and the first neural network can be represented by the change in the first network weight, then the information of the third neural network can be the change in the third network weight; similarly, when both the first device and the second device locally store only the fourth network structure corresponding to the second neural network, the difference between the adjusted second neural network and the second neural network can be represented by the change in the fourth network weight, then the information of the fourth neural network can be the change in the fourth network weight.

[0601] It should be noted that the seven possible specific contents of the information of the third neural network and the information of the fourth neural network listed in the above-mentioned method 2 are only examples and do not constitute any limitation to the scope of protection of this application. Simple variations of the above-mentioned possibility 1 and possibility 2 and other possible contents of the information of the third neural network and the information of the fourth neural network are also within the scope of protection of this application. For example, the information of the third neural network may be the index of the network weight corresponding to the third neural network, etc., and no further examples will be given here.

[0602] In addition, method 2 mainly describes the sending of information of the fourth neural network corresponding to the above-mentioned update method 2, or the sending of information of the third neural network and the fourth neural network. However, this application does not limit the sending of information of the fourth neural network shown in method 2, or the sending of information of the third neural network and the fourth neural network cannot be applied to the scenarios shown in the above-mentioned update methods 1 and / or 3. That is to say, when the update method is as shown in the above-mentioned update methods 1 and / or 3, the method of sending information of the fourth neural network, or the sending of information of the third neural network and the fourth neural network can also choose the method described in method 2.

[0603] Method 3:

[0604] Corresponding to the above-mentioned updating method three, a suitable third neural network and fourth neural network are selected and adjusted.

[0605] In this third method, the specific contents of the information of the third neural network and the information of the fourth neural network include the following possibilities:

[0606] Possibility 1: When the first network structure corresponding to the third neural network and the third network structure corresponding to the first neural network are the same, the third neural network can be represented by the first network structure and the first network weight, then the information of the third neural network can be the first network weight and the third indication information, wherein the third indication is used to indicate that the first network structure corresponding to the third neural network and the third network structure corresponding to the first neural network are the same, that is, indicating that the network structure has not changed; similarly, when the second network structure corresponding to the fourth neural network and the fourth network structure corresponding to the second neural network are the same, the fourth neural network can be represented by the second network structure and the second network weight, then the information of the fourth neural network can be the second network weight and the fourth indication information, wherein the fourth indication is used to indicate that the second network structure corresponding to the fourth neural network and the fourth network structure corresponding to the second neural network are the same, that is, indicating that the network structure has not changed.

[0607] Possibility 2: When the first network structure corresponding to the third neural network is the same as the third network structure corresponding to the first neural network, the third neural network can be represented by the first network structure and the first network weight. In addition, the first device and the second standby device respectively store the first network weight corresponding to the third neural network, and the index of the first network weight in the first device is consistent with the index of the first network weight in the second device. Then, the information of the third neural network can be the index of the first network weight and the third indication information, wherein the third indication is used to indicate that the first network structure corresponding to the third neural network is the same as the third network structure corresponding to the first neural network, that is, indicating that the network structure has not changed. Similarly, when the second network structure corresponding to the fourth neural network is the same as the fourth network structure corresponding to the second neural network, the fourth neural network can be represented by the second network structure and the second network weight. In addition, the first device and the second standby device respectively store the second network weight corresponding to the fourth neural network, and the index of the second network weight in the first device is consistent with the index of the second network weight in the second device. Then, the information of the fourth neural network can be the index of the second network weight and the fourth indication information, wherein the fourth indication is used to indicate that the second network structure corresponding to the fourth neural network is the same as the fourth network structure corresponding to the second neural network, that is, indicating that the network structure has not changed.

[0608] Optionally, the index of the first network weight and the index of the second network weight are the same. In this case, the index of the first network weight and the index of the second network weight are just one index.

[0609] Possibility three: when the first network structure corresponding to the third neural network and the third network structure corresponding to the first neural network are the same, the third neural network can be represented by the first network structure and the first network weight, the first neural network can be represented by the third network structure and the third network weight, and the change between the third network weight and the first network weight is the fifth network weight change. Then, the information of the third neural network can be the fifth network weight change and the third indication information, wherein the third indication is used to indicate that the first network structure corresponding to the third neural network and the third network structure corresponding to the first neural network are the same, that is, indicating that the network structure has not changed. Similarly, when the second network structure corresponding to the fourth neural network and the fourth network structure corresponding to the second neural network are the same, the fourth neural network can be represented by the second network structure and the second network weight, the second neural network can be represented by the fourth network structure and the fourth network weight, and the change between the fourth network weight and the second network weight is the sixth network weight change. Then, the information of the fourth neural network can be the sixth network weight change and the fourth indication information, wherein the fourth indication is used to indicate that the second network structure corresponding to the fourth neural network and the fourth network structure corresponding to the second neural network are the same, that is, indicating that the network structure has not changed.

[0610] Possibility four: when the first network structure corresponding to the third neural network and the third network structure corresponding to the first neural network are different or the same, the first device and the second device both locally store the third neural network, and the index of the third neural network in the first device is consistent with the index of the third neural network in the second device, then the information of the third neural network may be the index of the third neural network, and the information of the third neural network may also include indication information for indicating whether the network structure has changed; similarly, when the second network structure corresponding to the fourth neural network and the fourth network structure corresponding to the second neural network are different or the same, the first device and the second device both locally store the fourth neural network, and the index of the fourth neural network in the first device is consistent with the index of the fourth neural network in the second device, then the information of the fourth neural network may be the index of the fourth neural network, and the information of the fourth neural network may also include indication information for indicating whether the network structure has changed.

[0611] Optionally, the index of the third neural network and the index of the fourth neural network are the same. In this case, the information of the third neural network and the information of the fourth neural network are the same index and include indication information of whether the network structure has changed.

[0612] Possibility five: when the first network structure corresponding to the third neural network and the third network structure corresponding to the first neural network are different, the third neural network can be represented by the first network structure and the first network weight, then the information of the third neural network can be the first network structure, the first network weight and the fifth indication information, wherein the fifth indication is used to indicate that the first network structure corresponding to the third neural network and the third network structure corresponding to the first neural network are different, that is, to indicate that the network structure has changed; similarly, when the second network structure corresponding to the fourth neural network and the fourth network structure corresponding to the second neural network are different, the fourth neural network can be represented by the second network structure and the second network weight, then the information of the fourth neural network can be the second network structure, the second network weight and the sixth indication information, wherein the sixth indication is used to indicate that the second network structure corresponding to the fourth neural network and the fourth network structure corresponding to the second neural network are different, that is, to indicate that the network structure has changed.

[0613] Possibility six: when the first network structure corresponding to the third neural network and the third network structure corresponding to the first neural network are different, the third neural network can be represented by the first network structure and the first network weight, the first device and the second device both locally store the first network structure, and the index of the first network structure in the first device is consistent with the index of the first network structure in the second device, then the information of the third neural network can be the first network weight and the index of the first network structure and the fifth indication information, wherein the fifth indication is used to indicate that the first network structure corresponding to the third neural network and the third network structure corresponding to the first neural network are different, that is, indicating that the network structure has changed; similarly, when the second network structure corresponding to the fourth neural network and the fourth network structure corresponding to the second neural network are different, the fourth neural network can be represented by the second network structure and the second network weight, the first device and the second device both locally store the second network structure, and the index of the second network structure in the first device is consistent with the index of the second network structure in the second device, then the information of the fourth neural network can be the second network weight, the index of the second network structure and the sixth indication information, wherein the sixth indication is used to indicate that the second network structure corresponding to the fourth neural network and the fourth network structure corresponding to the second neural network are different, that is, indicating that the network structure has changed.

[0614] Optionally, the index of the first network structure and the index of the second network structure are the same. In this case, the index of the first network structure and the index of the second network structure may be one index.

[0615] Possibility seven: When the first network structure corresponding to the third neural network is different from the third network structure corresponding to the first neural network, the third neural network can be represented by the first network structure and the first network weight, the first device and the second device both locally store the first network structure and the first network weight, and the index of the first network structure in the first device is consistent with the index of the first network structure in the second device, and the index of the first network weight in the first device is consistent with the index of the first network weight in the second device, then the information of the third neural network can be the index of the first network weight and the index of the first network structure and the fifth indication information, wherein the fifth indication is used to indicate that the first network structure corresponding to the third neural network is different from the third network structure corresponding to the first neural network, that is, to indicate that the network structure has changed. ; Similarly, when the second network structure corresponding to the fourth neural network and the fourth network structure corresponding to the second neural network are different, the fourth neural network can be represented by the second network structure and the second network weights, the first device and the second device both locally store the second network structure and the second network weights, and the index of the second network structure in the first device is consistent with the index of the second network structure in the second device, and the index of the second network weight in the first device is consistent with the index of the second network weight in the second device, then the information of the fourth neural network can be the index of the second network weight, the index of the second network structure and the sixth indication information, wherein the sixth indication is used to indicate that the second network structure corresponding to the fourth neural network is different from the fourth network structure corresponding to the second neural network, that is, to indicate that the network structure has changed.

[0616] Optionally, the index of the first network structure and the index of the second network structure are the same, in which case the index of the first network structure and the index of the second network structure can be one index; the index of the first network weight and the index of the second network weight are the same, in which case the index of the first network weight and the index of the second network weight can be one index.

[0617] In addition, method three mainly describes the sending of information of the fourth neural network corresponding to the above-mentioned update method three, or the sending of information of the third neural network and the fourth neural network. However, this application does not limit the sending of information of the fourth neural network shown in method three, or the sending of information of the third neural network and the fourth neural network cannot be applied to the scenarios shown in the above-mentioned update methods one and / or two. That is to say, when the update method is as shown in the above-mentioned update methods one and / or two, the method of sending information of the fourth neural network, or the sending of information of the third neural network and the fourth neural network can also choose the method described in method three.

[0618] By way of example and not limitation, in the embodiments of the present application, before executing the sending of the third neural network information and / or the fourth neural network information, the first device or the second device may determine to send the third neural network information and / or the fourth neural network information to the opposite device in the following manner:

[0619] Method 1:

[0620] After the first device or the second device executes the above S510a to obtain the third neural network and the fourth neural network, it sends first information to the opposite device, where the first information is used to notify the opposite device to receive the information of the third neural network and / or the information of the fourth neural network. Figure 5 The method flow shown in (a) further includes S530a, sending first information.

[0621] Exemplarily, the transmitting end device sending the first information to the receiving end device includes: the local device sending the first information to the receiving end device on a PUCCH, a PDCCH, a PUSCH, a PDSCH, a UL grant, or a RACH.

[0622] It should be understood that, since PUCCH and PDCCH resources are limited, sending the first information to the receiving end device on PUSCH or PDSCH can increase the flexibility of the solution.

[0623] Furthermore, after receiving the first information, the peer device determines that it can receive the information of the third neural network and / or the information of the fourth neural network from the first device or the second device, that is, Figure 5 The method flow shown in (a) also includes S531a, sending a first response message, which is used to indicate that the first device or the second device can send the above-mentioned third neural network information and / or fourth neural network information.

[0624] Exemplarily, the transmitting end device sending the first response message to the receiving end device includes: the transmitting end device sending the first response message to the receiving end device on PUCCH, PDCCH, PUSCH, PDSCH, UL grant or RACH.

[0625] Exemplarily, the first response message is carried on an uplink grant (UL Grant), and the UL Grant includes a preset field indicating whether the first device or the second device sends the above-mentioned third neural network information and / or fourth neural network information.

[0626] For example, the preset field includes 2 bits. When the preset field bit value is 00, it indicates that the first device or the second device does not send the above-mentioned information of the third neural network and the information of the fourth neural network; when the preset field bit value is 01, it indicates that the first device or the second device does not send the above-mentioned information of the third neural network, but sends the information of the fourth neural network; when the preset field bit value is 10, it indicates that the first device or the second device does not send the above-mentioned information of the fourth neural network, but sends the information of the third neural network; when the preset field bit value is 11, it indicates that the first device or the second device sends the above-mentioned information of the third neural network and the information of the fourth neural network.

[0627] It should be noted that the number of bits and the meanings of the bit values ​​of the preset fields in the above UL Grant are only examples and do not constitute any limitation on the scope of protection of the present application.

[0628] It should be understood that, since PUCCH and PDCCH resources are limited, sending the first response message to the receiving end device on PUSCH or PDSCH can increase the flexibility of the solution.

[0629] Depending on the execution subject of the above S510a and the different neural networks that need to be configured on the peer device, the following S530a and S531a may include the following possibilities:

[0630] Possibility 1: When the first device updates the first neural network and the second neural network, and the second device needs to configure the third neural network. S530a is the first device sending first information to the second device, where the first information is used to notify the second device to receive information about the third neural network, or the first information is used to notify the second device to receive information about the third neural network and information about the fourth neural network. S531a is the second device sending a first response message to the first device, where the first response message is used to instruct the first device to send information about the third neural network, or the first response message is used to instruct the first device to send information about the third neural network and information about the fourth neural network.

[0631] Possibility 2: When the first device updates the first neural network and the second neural network, and the second device needs to configure the fourth neural network. S530a is the first device sending a first message to the second device, where the first message is used to notify the second device to receive information about the fourth neural network, or the first message is used to notify the second device to receive information about the third neural network and information about the fourth neural network. S531a is the second device sending a first response message to the first device, where the first response message is used to instruct the first device to send information about the fourth neural network, or the first response message is used to instruct the first device to send information about the third neural network and information about the fourth neural network.

[0632] Possibility three: When the second device updates the first neural network and the second neural network, and the first device needs to configure the third neural network. S530a is the second device sending first information to the first device, where the first information is used to notify the first device to receive information about the third neural network, or the first information is used to notify the first device to receive information about the third neural network and information about the fourth neural network. S531a is the first device sending a first response message to the second device, where the first response message is used to instruct the second device to send information about the third neural network, or the first response message is used to instruct the second device to send information about the third neural network and information about the fourth neural network.

[0633] Possibility 4: When the second device updates the first neural network and the second neural network, and the first device needs to configure the fourth neural network. S530a is the second device sending a first message to the first device, where the first message is used to notify the first device to receive information about the fourth neural network, or the first message is used to notify the first device to receive information about the third neural network and information about the fourth neural network. S531a is the first device sending a first response message to the second device, where the first response message is used to instruct the second device to send information about the fourth neural network, or the first response message is used to instruct the second device to send information about the third neural network and information about the fourth neural network.

[0634] Method 2:

[0635] After the first device or the second device executes the above S510a to obtain the third neural network and the fourth neural network, it sends first information to the opposite device, where the first information is used to notify the opposite device to receive the information of the third neural network and / or the information of the fourth neural network. Figure 5 The method flow shown in (a) further includes S530a, sending first information.

[0636] Furthermore, after receiving the above-mentioned first information, the opposite device may not agree to transmit information based on the AI ​​mode. For example, the opposite device sends a second response message to the first device or the second device, and the second response message is used to instruct the first device or the second device to transmit information based on the non-AI mode. For another example, after receiving the first information, the opposite device does not send any response message to the first device or the second device within a period of time. If the first device or the second device does not receive any response message within a fourth preset time period after sending the first information, the first device or the second device may determine to transmit information based on the traditional information transmission mode.

[0637] In this second method, Figure 5The method flow shown in (a) further includes S532a, where the opposite device sends a second response message to the first device or the second device, or the opposite device does not send a response message to the first device or the second device within a period of time.

[0638] Depending on the execution subject of the above S510a and the different neural networks that need to be configured on the peer device, S530a and S532a in the second mode include the following possibilities:

[0639] Possibility 1: When the first device updates the first neural network and the second neural network, and the second device needs to configure the third neural network. S530a is the first device sending a first message to the second device, where the first message is used to notify the second device to receive information from the third neural network, or the first message is used to notify the second device to receive information from the third neural network and information from the fourth neural network; S532a is the second device sending a second response message to the first device, where the second response message is used to instruct the first device to transmit information based on a non-AI mode, or the second device not to send a response message to the first device within a period of time.

[0640] Exemplarily, the transmitting end device sending the second response message to the receiving end device includes: the transmitting end device sending the second response message to the receiving end device on PUCCH, PDCCH, PUSCH, PDSCH, UL grant or RACH.

[0641] It should be understood that, since PUCCH and PDCCH resources are limited, sending the second response message to the receiving end device on PUSCH or PDSCH can increase the flexibility of the solution.

[0642] Possibility 2: When the first device updates the first neural network and the second neural network, and the second device needs to configure the fourth neural network. S530a is the first device sending a first message to the second device, where the first message is used to notify the second device to receive information from the fourth neural network, or the first message is used to notify the second device to receive information from the third neural network and information from the fourth neural network. S532a is the second device sending a second response message to the first device, where the second response message is used to instruct the first device to transmit information based on a non-AI mode, or the second device not to send a response message to the first device within a period of time.

[0643] Possibility three: When the second device updates the first neural network and the second neural network, and the first device needs to configure the third neural network. S530a is the second device sending a first message to the first device, where the first message is used to notify the first device to receive information from the third neural network, or where the first message is used to notify the first device to receive information from the third neural network and information from the fourth neural network. S532a is the first device sending a second response message to the second device, where the second response message is used to instruct the second device to transmit information based on a non-AI mode, or where the first device does not send a response message to the second device within a period of time.

[0644] Possibility 4: When the second device updates the first neural network and the second neural network, and the first device needs to configure the fourth neural network. S530a is the second device sending a first message to the first device, where the first message is used to notify the first device to receive information from the fourth neural network, or where the first message is used to notify the first device to receive information from the third neural network and information from the fourth neural network. S532a is the first device sending a second response message to the second device, where the second response message is used to instruct the second device to transmit information based on a non-AI mode, or where the first device does not send a response message to the second device within a period of time.

[0645] Method 3:

[0646] The first device or the second device executes the above S510a to obtain the third neural network and the fourth neural network. After receiving the second information, the first device or the second device sends the information of the fourth neural network and / or the information of the third neural network to the second device.

[0647] It should be noted that, in the embodiment of the present application, after the above-mentioned first device or second device receives the second information sent by the opposite device, the timing of sending the information of the fourth neural network and / or the information of the third neural network to the opposite device is not limited, and can be any time after receiving the second information.

[0648] For example, after the first device or the second device receives the second information sent by the opposite device, it starts to send the information of the third neural network and / or the information of the fourth neural network to the opposite device within the first preset time period after the first starting moment or after the first preset time period. The second information is used to indicate that the opposite device can receive the information of the third neural network and / or the information of the fourth neural network, that is, Figure 5 The method flow shown in (a) further includes S540a, sending second information.

[0649] The first starting time includes any one of the following times:

[0650] The moment when the receiving device receives the first indication information from the sending device, the moment when the sending device sends the second notification message to the receiving device, the moment when the receiving device receives the second notification message from the sending device, the moment when the sending device sends the first signaling to the receiving device, and the moment when the receiving device receives the second signaling from the sending device, wherein the first signaling is used to indicate the start of timing and the second signaling is used to indicate the start of updating.

[0651] Exemplarily, the transmitting end device sending the second information to the receiving end device includes: the transmitting end device sending the second information to the receiving end device on a PUCCH, a PDCCH, a PUSCH, a PDSCH, a UL grant, or a RACH.

[0652] It should be understood that, since PUCCH and PDCCH resources are limited, sending the second information to the receiving end device on PUSCH or PDSCH can increase the flexibility of the solution.

[0653] Depending on the execution subject of the above S510a and the different neural networks that need to be configured on the peer device, S540a in the third mode includes the following possibilities:

[0654] Possibility 1: When the first device updates the first neural network and the second neural network, and the second device needs to configure the third neural network, S540a is the second device sending second information to the first device, where the second information is used to indicate that the second device can receive information about the third neural network, or the second information is used to indicate that the second device can receive information about the third neural network and information about the fourth neural network.

[0655] Possibility 2: When the first device updates the first neural network and the second neural network, and the second device needs to configure the fourth neural network, S540a is the second device sending second information to the first device, where the second information is used to indicate that the second device can receive information about the fourth neural network, or the second information is used to indicate that the second device can receive information about the third neural network and information about the fourth neural network.

[0656] Possibility three: When the second device updates the first neural network and the second neural network, and the first device needs to configure the third neural network, S540a is the first device sending second information to the second device, where the second information is used to indicate that the first device can receive information about the third neural network, or the second information is used to indicate that the first device can receive information about the third neural network and information about the fourth neural network.

[0657] Possibility 4: When the second device updates the first neural network and the second neural network, and the first device needs to configure the fourth neural network. S540a is the first device sending second information to the second device, where the second information is used to indicate that the first device can receive information about the fourth neural network, or the second information is used to indicate that the first device can receive information about the third neural network and information about the fourth neural network.

[0658] Method 4:

[0659] The first device or the second device executes the above S510a to obtain the third neural network and the fourth neural network, and the first device or the second device sends information about the third neural network and / or information about the fourth neural network to the opposite device.

[0660] It should be noted that, in the embodiment of the present application, the timing at which the first device or the second device sends the information of the third neural network and / or the information of the fourth neural network to the opposite device is not limited.

[0661] For example, the first device or the second device starts to send information about the third neural network and / or information about the fourth neural network to the opposite device within the second preset time period after the second starting moment or after the second preset time period, that is, the above S520a is directly executed after the preset time period under method four to send information about the third neural network and / or information about the fourth neural network.

[0662] The second starting time includes any of the following times:

[0663] The moment when the receiving device receives the first indication information from the sending device, the moment when the sending device sends the second notification message to the receiving device, the moment when the receiving device receives the second notification message from the sending device, the moment when the sending device sends the first signaling to the receiving device, and the moment when the receiving device receives the second signaling from the sending device, wherein the first signaling is used to indicate the start of timing and the second signaling is used to indicate the start of updating.

[0664] Depending on the execution subject of the above S510a and the different neural networks that need to be configured on the peer device, the fourth method S520a includes the following possibilities:

[0665] Possibility 1: When the first device updates the first neural network and the second neural network, and the second device needs to configure the third neural network, S520a is the first device sending information about the third neural network to the second device, or the first device sending information about the third neural network and information about the fourth neural network to the second device.

[0666] Possibility 2: When the first device updates the first neural network and the second neural network, and the second device needs to configure the fourth neural network, S520a is the first device sending information about the fourth neural network to the second device, or the first device sending information about the third neural network and the fourth neural network to the second device.

[0667] Possibility three: When the second device updates the first neural network and the second neural network, and the first device needs to configure the third neural network, S520a is the second device sending information about the third neural network to the first device, or the second device sending information about the third neural network and information about the fourth neural network to the first device.

[0668] Possibility 4: When the second device updates the first neural network and the second neural network, and the first device needs to configure the fourth neural network, S520a is the second device sending information about the fourth neural network to the first device, or the second device sending information about the third neural network and the fourth neural network to the first device.

[0669] It should be noted that in both Mode 3 and Mode 4, both the first device and the second device can be informed of the moment when the update of the first neural network and the second neural network begins (for example, the local device performing the update can be informed, and the other device can perceive the occurrence of the update based on the interruption of information transmission or the change in mode). After a preset period of time, the first device and the second device are informed that the update is complete and can agree to transmit the updated information of the third neural network and / or the fourth neural network, without the need for an instruction (such as the first information) to notify the other device that the update is complete.

[0670] Method 5:

[0671] After the first device or the second device executes the above S510a to obtain the third neural network and the fourth neural network, it sends the first information to the opposite device, and the first information is used to notify the opposite device to receive the information of the third neural network and / or the information of the fourth neural network, that is, the following method 5 Figure 5 The method flow shown in (a) also includes the above-mentioned S530a, sending the first information.

[0672] Furthermore, after the first device or the second device sends the above-mentioned first information, it starts to send the third neural network information and / or the fourth neural network information to the opposite device within the third preset time period or after the third preset time period, that is, the fifth method is as follows: Figure 5 The method flow shown in (a) also includes the above-mentioned S520a, sending information of the third neural network and / or information of the fourth neural network.

[0673] Different from the above-mentioned method 1, under method 5, after the first device or the second device sends the above-mentioned first information, it does not need to wait for the response of the opposite device and directly sends the updated third neural network information and / or fourth neural network information to the opposite device after a period of time.

[0674] Depending on the execution subject of the above S510a and the different neural networks that need to be configured on the peer device, S530a and S520a in the fifth mode include the following possibilities:

[0675] Possibility 1: When the first device updates the first neural network and the second neural network, and the second device needs to configure the third neural network. S530a is the first device sending first information to the second device, where the first information is used to notify the second device to receive information about the third neural network, or the first information is used to notify the second device to receive information about the third neural network and information about the fourth neural network. S520a is the first device sending information about the third neural network to the second device, or the first device sending information about the third neural network and information about the fourth neural network to the second device.

[0676] Possibility 2: When the first device updates the first neural network and the second neural network, and the second device needs to configure the fourth neural network. S530a is the first device sending first information to the second device, where the first information is used to notify the second device to receive information about the fourth neural network, or the first information is used to notify the second device to receive information about the third neural network and information about the fourth neural network. S520a is the first device sending information about the fourth neural network to the second device, or the first device sending information about the third neural network and information about the fourth neural network to the second device.

[0677] Possibility three: When the second device updates the first neural network and the second neural network, and the first device needs to configure the third neural network. S530a is the second device sending first information to the first device, where the first information is used to notify the first device to receive information about the third neural network, or the first information is used to notify the first device to receive information about the third neural network and information about the fourth neural network. S520a is the second device sending information about the third neural network to the first device, or the second device sending information about the third neural network and information about the fourth neural network to the first device.

[0678] Possibility 4: When the second device updates the first neural network and the second neural network, and the first device needs to configure the fourth neural network. S530a is the second device sending first information to the first device, where the first information is used to notify the first device to receive information about the fourth neural network, or the first information is used to notify the first device to receive information about the third neural network and information about the fourth neural network. S520a is the second device sending information about the fourth neural network to the first device, or the second device sending information about the third neural network and information about the fourth neural network to the first device.

[0679] It should be understood that the above-mentioned methods 1 to 5 are only examples. The first device and the second device can determine whether to transmit the above-mentioned third neural network information and / or fourth neural network information through negotiation, or the first device and the second device can directly transmit the above-mentioned third neural network information and / or fourth neural network information without negotiation, which does not constitute any limitation on the scope of protection of the present application. It can also be determined whether to transmit the above-mentioned third neural network information and / or fourth neural network information through other methods, for example, through protocol predefinition: after the local device updates the third neural network and the fourth neural network, the protocol stipulates that the opposite device needs to accept the updated third neural network information and / or the fourth neural network information, and transmit information based on the updated communication system including the third neural network and the fourth neural network. Examples will not be given one by one here.

[0680] It should also be understood that after the receiving device receives the information of the third neural network and / or the fourth neural network, it can locally determine whether the performance of the third neural network and / or the fourth neural network meets the requirements, and decide whether to use the third neural network and the fourth neural network to transmit information with the sending device.

[0681] Exemplarily, the receiving device sends a first notification message to the sending device, where the first notification message is used to indicate whether the receiving device adopts the fourth neural network, or whether to adopt the fourth neural network and the third neural network for information transmission.

[0682] For example, when the receiving device and the sending device have negotiated to return to the non-AI mode for information transmission, the first notification message may indicate the use of the AI ​​mode for information transmission, that is, the use of the received third neural network and the fourth neural network to transmit information with the sending device, or the first notification message may indicate the use of the non-AI mode for information transmission, that is, the refusal to use the received third neural network and the fourth neural network to transmit information with the sending device; for another example, when the receiving device and the sending device have not negotiated to return to the non-AI mode, but use the first neural network and the second neural network for information transmission, the first notification message may indicate the use of the AI ​​mode for information transmission, that is, switching from using the first neural network and the second neural network for information transmission to using the received third neural network and the fourth neural network for information transmission with the sending device, or the first notification message may indicate the use of the non-AI mode for information transmission, that is, refusing to use the received third neural network and the fourth neural network for information transmission with the sending device to return to the non-AI mode.

[0683] Figure 5 The method flow shown in (b) can be applied to Figure 5 (d) shows the transmission framework. Figure 5 It can be seen from (d) that the terminal device side obtains the downlink channel information to be fed back, and processes the downlink channel information to be fed back through the first compressor to obtain the compressed channel information that needs to be fed back to the air interface. After the network device side obtains the compressed channel information, it decompresses it through the first decompressor to obtain the recovered downlink channel information.

[0684] Theoretically, the AI ​​model composed of the trained first compressor and the first decompressor can make the difference or similarity between the recovered downlink channel information and the downlink channel information to be fed back within an acceptable range. However, the first compressor and the first decompressor are obtained through offline training. In the process of online downlink channel information feedback, if the downlink channel to be fed back is different from the training channel characteristics used to train the first compressor and the first decompressor, it may cause the difference or similarity between the recovered downlink channel information and the downlink channel information to be fed back to be not within an acceptable range after the AI ​​model composed of the first compressor and the first decompressor feeds back the downlink channel information to be fed back. In this case, it is necessary to update the first compressor and the first decompressor online so that when the downlink channel information is fed back based on the updated first compressor and the first decompressor, the difference or similarity between the recovered downlink channel information and the downlink channel information to be fed back is within an acceptable range. The following is combined with Figure 5 (b) Describe in detail how to update the first compressor and the first decompressor online.

[0685] It should be noted that Figure 5 (b) The method flow shown is the same as the above Figure 5 The method shown in (a) is similar except that Figure 5 (b) Mainly introduces the online update of the first compressor and the first decompressor in the scenario of channel information feedback.

[0686] in addition, Figure 5 (b) The interactive signaling between the first device and the second device Figure 5 Similar to (a), it can be carried on PUCCH, PDCCH, PUSCH, PDSCH, UL grant or RACH, and will not be repeated below. Among them, due to the limited resources of PUCCH and PDCCH, signaling interaction on PUSCH or PDSCH can increase the flexibility of the solution.

[0687] Specifically, Figure 5 The first device or the second device in (b) can perform the steps of updating the first compressor and the first decompressor, and Figure 5 Similar to the above (a), the first device or the second device can periodically or aperiodically update the first compressor and the first decompressor by itself without triggering the update. In order to reflect the specific implementation in the scenario of channel information feedback, Figure 5 (b) mainly describes in detail the process of online updating of the first compressor and the first decompressor in the scenario of channel information feedback from the perspective of the first device determining that the first compressor and the first decompressor need to be updated. For the periodic or non-periodic update of the first compressor and the first decompressor by the first device or the second device, please refer to the above Figure 5 The description in (a), Figure 5 (b) No further details will be given.

[0688] but Figure 5 The method shown in (b) includes some or all of the following steps.

[0689] S510: The first device determines whether the first compressor and the first decompressor need to be updated.

[0690] The first device includes a terminal device or a network device.

[0691] The following describes two methods in which the first device can be used to determine whether the first compressor and the first decompressor need to be updated:

[0692] Method 1:

[0693] The first device determines estimated downlink channel information based on downlink channel information, a first compressor, and a first decompressor, and determines whether the first compressor and the first decompressor need to be updated based on a relationship between a difference or similarity between the downlink channel information and the estimated downlink channel information and a corresponding preset threshold. The difference between the downlink channel information and the estimated downlink channel information is compared with a first preset threshold, and the similarity between the downlink channel information and the estimated downlink channel information is compared with a second preset threshold.

[0694] In this application, the preset threshold may also be referred to as a threshold value, a threshold value, etc. The specific name does not constitute any limitation on the protection scope of this application.

[0695] Exemplarily, when the difference between the downlink channel information and the estimated downlink channel information is greater than or equal to a first preset threshold, or when the similarity between the downlink channel information and the estimated downlink channel information is less than or equal to a second preset threshold, the first device determines that the first compressor and the first decompressor need to be updated; when the difference between the downlink channel information and the estimated downlink channel information is less than the first preset threshold, or when the similarity between the downlink channel information and the estimated downlink channel information is greater than the second preset threshold, the first device determines that the first compressor and the first decompressor do not need to be updated.

[0696] Alternatively, when the difference between the downlink channel information and the estimated downlink channel information is greater than a first preset threshold, or when the similarity between the downlink channel information and the estimated downlink channel information is less than a second preset threshold, the first device determines that the first compressor and the first decompressor need to be updated; when the difference between the downlink channel information and the estimated downlink channel information is less than or equal to the first preset threshold, or when the similarity between the downlink channel information and the estimated downlink channel information is greater than or equal to the second preset threshold, the first device determines that the first compressor and the first decompressor do not need to be updated.

[0697] It should be noted that the above-mentioned determination of whether the first compressor and the first decompressor need to be updated based on the relationship between the difference or similarity between the downlink channel information and the estimated downlink channel information and the corresponding preset threshold is merely an example and does not constitute any limitation on the scope of protection of this application. For example, the first device may further determine whether the first compressor and the first decompressor need to be updated based on the further difference or similarity between the difference or similarity between the downlink channel information and the estimated downlink channel information and the corresponding preset threshold.

[0698] The specific value of the above-mentioned preset threshold can be predefined by the protocol, or determined by negotiation between the network device and the terminal device, or determined by the network device, or determined by the terminal device, or notified to the network device and the terminal device after being determined by a third-party device. In the embodiment of the present application, there is no restriction on how the first device obtains the preset threshold.

[0699] In the following manner, the first device determines the estimated downlink channel information based on the downlink channel information, the first compressor, and the first decompressor, including the following possibilities:

[0700] Possibility 1: The first device is a terminal device, and the downlink channel information is the downlink channel without compression loss obtained by the terminal device (denoted as H d ). The first compressor is denoted as function f en , the first decompressor is denoted as function f de In the embodiments of the present application, there is no limitation on how the terminal device obtains a downlink channel without compression loss. The existing solution for the terminal device to obtain a downlink channel without compression loss may be referred to, and no further details will be given here.

[0701] In the following example, the process of estimating downlink channel information based on downlink channel information, the first compressor and the first decompressor is as follows: Figure 6 As shown in (a), Figure 6 This is a schematic diagram of determining and estimating downlink channel information provided by an embodiment of the present application.

[0702] from Figure 6 It can be seen from (a) that the downlink channel information is compressed by the first compressor and then decompressed by the first decompressor to obtain the above-mentioned estimated downlink channel information.

[0703] It should be noted that, if possible, the terminal device is aware of the above-mentioned first compressor and first decompressor (for example, the terminal device locally stores the first compressor and the first decompressor, or the terminal device is aware of the first compressor and the network device sends the above-mentioned first decompressor to the terminal device), so after the terminal device obtains a downlink channel with no compression loss, it can determine the estimated downlink channel information based on the downlink channel with no compression loss and the known first compressor and first decompressor.

[0704] The terminal device determines the estimated downlink channel information by calculating the estimated downlink channel information, or by looking up the estimated downlink channel information in a table.

[0705] For example, from Figure 6 It can be seen from (a) that the terminal device can calculate the estimated downlink channel information based on the following formula:

[0706] Among them, H d represents the compression-lossless downlink channel obtained by the terminal device, f en represents the first compressor, f en (H d ) represents the result of compression of the downlink channel without compression loss obtained by the terminal device through the first compressor, f de represents the first decompressor, Indicates the estimated downlink channel information calculated by the terminal device.

[0707] It should be understood that the above formula This is just an example and does not constitute any limitation on the scope of protection of this application. Simple variations of this formula are also within the scope of protection of this application, such as: Δ is the offset, which may be a preset value and will not be described in detail here.

[0708] For another example, the terminal device may obtain estimated downlink channel information based on Table 1.

[0709] Table 1

[0710]

[0711] It should be understood that the above-mentioned method for the terminal device to determine the estimated downlink channel information is only an example and does not constitute any limitation to the scope of protection of this application. Other methods in which the terminal device can determine the estimated downlink channel information are also within the scope of protection of this application, and this application will no longer give examples one by one.

[0712] Possibility 2: The first device is a network device, and the downlink channel information is the downlink channel with compression loss obtained by the network device (denoted as H d_NB ). The first compressor is denoted as function f en , the first decompressor is denoted as function f de In the embodiment of the present application, there is no restriction on how the network device obtains the downlink channel with compression loss. The solution of obtaining the downlink channel without compression loss by using the existing AI method or the traditional channel feedback method can be referred to, which will not be described in detail here.

[0713] In the second possibility, the process of estimating the downlink channel information based on the downlink channel information, the first compressor and the first decompressor is as follows: Figure 6 As shown in (b) in Figure 6 It can be seen from (b) that the downlink channel information is compressed by the first compressor and then decompressed by the first decompressor to obtain the above-mentioned estimated downlink channel information.

[0714] It should be noted that, in the second possibility, the network device is already aware of the above-mentioned first compressor and first decompressor (for example, the network device locally stores the first compressor and the first decompressor, or the network device is already aware of the first decompressor and the terminal device sends the above-mentioned first compressor to the network device), so after the network device obtains the downlink channel with compression loss, it can determine the estimated downlink channel information based on the downlink channel with compression loss and the known first compressor and first decompressor.

[0715] The network device determines the estimated downlink channel information by calculating the estimated downlink channel information, or by looking up the estimated downlink channel information in a table.

[0716] For example, from Figure 6 From (b) in the figure, it can be seen that the network device can calculate the estimated downlink channel information based on the following formula:

[0717] Among them, H d_NB Indicates the downlink channel with compression loss obtained by the network device, f en represents the first compressor, f en (H d_NB ) represents the result of compression of the downlink channel with compression loss obtained by the network device through the first compressor, f de represents the first decompressor, Indicates the estimated downlink channel information calculated by the network device.

[0718] It should be understood that the above formula This is just an example and does not constitute any limitation on the scope of protection of this application. Simple variations of this formula are also within the scope of protection of this application, such as: Δ is the offset, which may be a preset value and will not be described in detail here.

[0719] For another example, the terminal device may obtain estimated downlink channel information based on Table 2.

[0720] Table 2

[0721]

[0722] It should be understood that the above-mentioned method for network devices to determine estimated downlink channel information is only an example and does not constitute any limitation on the scope of protection of this application. Other methods in which network devices can determine estimated downlink channel information are also within the scope of protection of this application, and this application will no longer give examples one by one.

[0723] Method 2:

[0724] The first device determines estimated downlink channel information based on the downlink channel information, the uplink channel information, the first compressor, and the first decompressor. Furthermore, determining whether to update the first compressor and the first decompressor based on the downlink channel information and the estimated downlink channel information is similar to the first embodiment described above and will not be further described here. Therefore, the second embodiment primarily describes how to determine the estimated downlink channel information based on the downlink channel information, the uplink channel information, the first compressor, and the first decompressor.

[0725] In the second mode, the first device determines the estimated downlink channel information based on the downlink channel information, the uplink channel information, the first compressor, and the first decompressor, including the following possibilities:

[0726] Possibility 1: The first ...

Claims

1. A method for adjusting a neural network, comprising: a first neural network and a second neural network, wherein: The first neural network is applied to the first device side, and the second neural network is applied to the second device side, wherein the method comprises: The first device determines a third neural network and a fourth neural network, wherein the third neural network and the fourth neural network are respectively the updated or adjusted first neural network and the second neural network; The first device sends information of the fourth neural network to the second device, The first neural network or the third neural network is used for the first device to perform first signal processing, and the second neural network or the fourth neural network is used for the second device to perform second signal processing, and the second signal processing corresponds to the first signal processing.

2. The method according to claim 1, characterized in that The first neural network or the third neural network is used by the first device to perform first signal processing, including: The first neural network is used for the first device to process the first signal to obtain the second signal; The third neural network is used for the first device to process the third signal to obtain a fourth signal; The second neural network or the fourth neural network is used by the second device to perform second signal processing, including: The second neural network is used for the second device to process the second signal to obtain a fifth signal; The fourth neural network is used for the second device to process the fourth signal to obtain a sixth signal; or, The second neural network or the fourth neural network is used by the second device to perform second signal processing, including: The second neural network is used for the second device to process the first signal to obtain the second signal; The fourth neural network is used for the second device to process the third signal to obtain the fourth signal; The first neural network or the third neural network is used by the first device to perform first signal processing, including: The first neural network is used for the first device to process the second signal to obtain the fifth signal; The third neural network is used by the first device to process the fourth signal to obtain the sixth signal.

3. The method according to claim 2, characterized in that The difference between the first training signal related to the third signal and the second training signal related to the sixth signal satisfies a first condition; or, The similarity between the first training signal and the second training signal satisfies a second condition.

4. The method according to claim 3, characterized in that The difference degree includes any of the following: Difference, mean square error, normalized mean square error, or mean absolute error; The similarity includes a correlation coefficient.

5. The method according to any one of claims 2 to 4, characterized in that: The first neural network or the second neural network is used to process the first signal to obtain the second signal, including: The first neural network or the second neural network and the first functional module are used to process the first signal to obtain the second signal; and / or, The first neural network or the second neural network is used to process the second signal to obtain the fifth signal, including: The first neural network or the second neural network and the second functional module are used to process the second signal to obtain the fifth signal.

6. The method according to any one of claims 2 to 4, characterized in that: The third neural network or the fourth neural network is used to process the third signal to obtain the fourth signal, including: The third neural network or the fourth neural network and the third functional module are used to process the third signal to obtain the fourth signal; and / or, The third neural network or the fourth neural network is used to process the fourth signal to obtain the sixth signal, including: The third neural network or the fourth neural network and the fourth functional module are used to process the fourth signal to obtain the sixth signal.

7. The method according to any one of claims 1 to 4, characterized in that The first device determines the third neural network and the fourth neural network including: The first device updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively.

8. The method according to claim 7, characterized in that Before the first device updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively, the method further includes: The first device receives first indication information from the second device, where the first indication information is used to instruct the first device to update the first neural network and the second neural network; or The first device periodically or non-periodically determines whether the first neural network and the second neural network need to be updated, and the first device determines that the first neural network and the second neural network need to be updated; or, The first device receives second indication information from the second device, the second indication information is used to instruct the first device to determine whether the first neural network and the second neural network need to be updated, and the first device determines that the first neural network and the second neural network need to be updated; or The first device determines whether the first neural network and the second neural network need to be updated based on a preset condition trigger, and the first device determines that the first neural network and the second neural network need to be updated.

9. The method according to claim 8, characterized in that The first indication information and / or the second indication information is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

10. The method according to claim 7, characterized in that The first device updating the first neural network and the second neural network comprises: The first device periodically or aperiodically updates the first neural network and the second neural network.

11. The method according to claim 8, characterized in that The first device determines whether the first neural network and the second neural network need to be updated, including: The first device determines whether the first neural network and the second neural network need to be updated based on whether the difference or similarity between the first information and the second information meets a third condition, wherein the first information includes a first judgment signal related to the first signal, and the second information includes a second judgment signal related to the fifth signal.

12. The method according to claim 11, characterized in that The method further comprises: The first device determines the second judgment signal based on the first judgment signal, the first neural network and the second neural network; or, The first device determines the second judgment signal based on the first judgment signal, the first neural network and the second neural network, and the first functional module and / or the second functional module.

13. The method according to claim 11, characterized in that The first determination signal includes the first device receiving downlink channel information from the second device, or, The first determination signal includes channel information measured by the first device based on receiving a sounding reference signal SRS from the second device; The downlink channel information is carried on a preset resource or the downlink channel information is indicated to be used to determine the first judgment signal.

14. The method according to any one of claims 1 to 4, characterized in that The first device determines the third neural network and the fourth neural network including: The first device selects the third neural network and the fourth neural network from a first set based on communication system data, wherein the first set includes a plurality of neural networks; or The first device adjusts the network weights corresponding to the first neural network and the second neural network respectively based on the communication system data to obtain the third neural network and the fourth neural network; or The first device selects a fifth neural network and a sixth neural network from the second set based on the communication system data, and then adjusts the network weights corresponding to the fifth neural network and the sixth neural network respectively based on the communication system data to obtain the third neural network and the fourth neural network.

15. The method according to claim 14, characterized in that The communication system data includes at least one of channel characteristics, signal-to-noise ratio, downlink channel information, uplink channel information and hybrid automatic repeat request HARQ information.

16. The method according to any one of claims 1 to 4, characterized in that The method further comprises: The first device sends information of the third neural network to the second device.

17. The method according to claim 16, characterized in that The information of the third neural network includes at least one of the following information: The identification of the third neural network, information on the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, the identification of the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, the identification of the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, information on the network weights corresponding to the third neural network, the amount of change between the network weights corresponding to the third neural network and the network weights corresponding to the first neural network, the identification of the network weights corresponding to the third neural network, and third indication information, wherein the third indication information is used to indicate whether the network structure corresponding to the third neural network is the same as the network structure corresponding to the first neural network.

18. The method of claim 1, wherein: The information of the fourth neural network includes at least one of the following information: The identification of the fourth neural network, information on the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identification of the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identification of the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, information on the network weights corresponding to the fourth neural network, the amount of change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network, the identification of the network weights corresponding to the fourth neural network, and fourth indication information, wherein the fourth indication information is used to indicate whether the network structure corresponding to the fourth neural network and the network structure corresponding to the second neural network are the same.

19. The method according to claim 17 or 18, characterized in that The amount of change between the network weights corresponding to the third neural network and the network weights corresponding to the first neural network, and / or the amount of change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network include one or more of the following information: Weight position information, weight interpretation rule indication information, weight value arrangement information and weight value change amount arrangement information; The information of the network structure corresponding to the third neural network and / or the information of the network structure corresponding to the fourth neural network includes: an indication of the type of neural network and an indication of the structure of the neural network; The information of the network weights corresponding to the third neural network and / or the information of the network weights corresponding to the fourth neural network include: Weight interpretation rule indication information and / or weight value arrangement information.

20. The method according to any one of claims 1 to 4, characterized in that Before the first device sends the information of the fourth neural network to the second device, or before sending the information of the fourth neural network and the information of the third neural network, the method further includes: The first device sends first information to the second device, where the first information is used to request the second device to receive information about the fourth neural network, or to request the second device to receive information about the fourth neural network and information about the third neural network; The first device receives a first response message from the second device, where the first response message is used to instruct the first device to send information about the fourth neural network, or to instruct the first device to send information about the fourth neural network and information about the third neural network.

21. The method of claim 20, wherein: The first information and / or the first response message is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

22. The method according to any one of claims 1 to 4, characterized in that Before the first device sends the information of the fourth neural network to the second device, or before sending the information of the fourth neural network and the information of the third neural network, the method further includes: The first device receives second information from the second device, where the second information is used to indicate that the second device can receive information about the fourth neural network, or to indicate that the second device can receive information about the fourth neural network and information about the third neural network; The first device sending the information of the fourth neural network to the second device, or sending the information of the fourth neural network and the information of the third neural network includes: After receiving the second information, the first device sends the information of the fourth neural network to the second device, or sends the information of the fourth neural network and the information of the third neural network.

23. The method of claim 22, wherein: After receiving the second information, the first device sends the information of the fourth neural network to the second device, or sends the information of the fourth neural network and the information of the third neural network, including: After the first device receives the second information, it starts sending the information of the fourth neural network to the second device within a first preset time period after the first starting moment or after the first preset time period, or sends the information of the fourth neural network and the information of the third neural network.

24. The method of claim 22, wherein: The second information is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

25. The method according to any one of claims 1 to 4, characterized in that The first device sending the information of the fourth neural network to the second device, or sending the information of the fourth neural network and the information of the third neural network includes: After the first device updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively, it sends information of the fourth neural network to the second device, or sends information of the fourth neural network and information of the third neural network.

26. The method of claim 25, wherein: The first device sending the information of the fourth neural network to the second device, or sending the information of the fourth neural network and the information of the third neural network includes: After the first device updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively, it starts sending information of the fourth neural network to the second device within a second preset time after the second starting moment or after the second preset time, or sends information of the fourth neural network and information of the third neural network.

27. The method according to any one of claims 1 to 4, characterized in that Before the first device sends the information of the fourth neural network to the second device, or before sending the information of the fourth neural network and the information of the third neural network, the method further includes: The first device sends first information to the second device, where the first information is used to notify the second device to receive information about the fourth neural network, or information about the fourth neural network and information about the third neural network; The first device sending the information of the fourth neural network to the second device, or sending the information of the fourth neural network and the information of the third neural network includes: The first device starts sending the information of the fourth neural network to the second device within a third preset time period after sending the first information to the second device or after the third preset time period, or sends the information of the fourth neural network and the information of the third neural network.

28. The method according to any one of claims 1 to 4, characterized in that After the first device sends the information of the fourth neural network to the second device, or sends the information of the fourth neural network and the information of the third neural network, the method further includes: The first device receives a first notification message from the second device, where the first notification message is used to instruct the second device to adopt the fourth neural network, or to adopt the fourth neural network and the third neural network for information transmission.

29. The method of claim 28, wherein: The first notification message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

30. The method of claim 20, wherein: The method further comprises: The first device sends a second notification message to the second device, or, The first device receives a second notification message from the second device, The second notification message is used to notify that the information transmission mode changes from a first mode to a second mode, the first mode includes information transmission based on a neural network, the neural network includes the first neural network and the second neural network, and the second mode includes information transmission based on a non-neural network.

31. The method of claim 30, wherein: The neural network also includes a seventh neural network and an eighth neural network.

32. The method of claim 30, wherein: The method further comprises: The first device switches the information transmission mode from the first mode to the second mode.

33. The method of claim 32, wherein: The first device switching the information transmission mode from the first mode to the second mode includes: The first device starts to switch the information transmission mode from the first mode to the second mode within a fourth preset time period after a fourth starting moment or after the fourth preset time period, wherein the fourth starting moment includes the moment when the first device sends the second notification message to the second device, or the moment when the first device receives the second notification message from the second device.

34. The method of claim 23, wherein: The first starting time and the second starting time include: The time when the first device receives the first indication information from the second device; or The time when the first device sends the second notification message to the second device; or The time when the first device receives the second notification message from the second device; or The time when the first device sends the first signaling to the second device; or a time point at which the first device receives a second signaling from the second device, wherein the first signaling is used to indicate a timing start, and the second signaling is used to indicate an update start; The second notification message is used to notify that the information transmission mode is changed from the first mode to the second mode.

35. The method of claim 30, wherein: The second notification message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

36. The method according to any one of claims 1 to 4, characterized in that The method further comprises: The first device receives a second response message from the second device, where the second response message is used to indicate a refusal to receive information about the fourth neural network, or to indicate a refusal to receive information about the fourth neural network and information about the third neural network.

37. The method of claim 36, wherein: The second response message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

38. The method according to any one of claims 1 to 4, characterized in that The first device includes a terminal device or a network device, the second device includes the terminal device or the network device, When the first device is the terminal device, the second device is the network device, When the first device is the network device, the second device is the terminal device.

39. The method according to any one of claims 1 to 4, characterized in that The method further comprises: The first device receives fifth indication information from the second device, where the fifth indication information is used to indicate the capability of the first device to send the first device; The first device sends first capability information to the second device, where the first capability information includes at least one of the following information: computing capability information of the first device, storage capability information of the first device or information on whether the first device stores a first AI model, a neural network structure supported by the first device, a neural network scale supported by the first device, and an AI communication function supported by the first device, wherein the first AI model is any AI model; or The method further comprises: The first device sends sixth indication information to the second device, where the sixth indication information is used to indicate the second device to send a capability of the second device; The first device receives second capability information from the second device, where the second capability information includes at least one of the following information: The computing power information of the second device, the storage capacity information of the second device or whether the second device stores information of a second AI model, the neural network structure supported by the second device, the neural network scale supported by the second device, and the AI ​​communication function supported by the second device, wherein the second AI model is any AI model.

40. The method according to any one of claims 2 to 4, characterized in that The first signal includes channel information; The first neural network includes a first compressor, the second neural network includes a first decompressor, the third neural network includes a second compressor, and the fourth neural network includes a second decompressor; or, The first neural network includes a first decompressor, the second neural network includes a first compressor, the third neural network includes a second decompressor, and the fourth neural network includes a second compressor.

41. The method of claim 40, wherein: The channel information includes channel information received by multiple receiving antennas or channel information of multiple layers.

42. A method for adjusting a neural network, involving a first neural network and a second neural network, wherein: The first neural network is applied to the first device side, and the second neural network is applied to the second device side, wherein the method comprises: The second device receives information of a fourth neural network from the first device, wherein the fourth neural network is a neural network updated or adjusted by the second neural network; The second device determines the fourth neural network based on the information of the fourth neural network, The first neural network or the third neural network is used for the first device to perform first signal processing, the second neural network or the fourth neural network is used for the second device to perform second signal processing, the second signal processing corresponds to the first signal processing, and the third neural network is the updated or adjusted first neural network determined by the first device.

43. The method of claim 42, wherein: The method further comprises: The second device receives information of a third neural network from the first device, where the third neural network is a neural network corresponding to the first neural network; The first neural network or the third neural network is used for the first device to perform first signal processing, and the second neural network or the fourth neural network is used for the second device to perform second signal processing, and the second signal processing corresponds to the first signal processing.

44. The method of claim 43, wherein: The first neural network or the third neural network is used by the first device to perform first signal processing, including: The first neural network is used for the first device to process the first signal to obtain the second signal; The third neural network is used for the first device to process the third signal to obtain a fourth signal; The second neural network or the fourth neural network is used by the second device to perform second signal processing, including: The second neural network is used for the second device to process the second signal to obtain a fifth signal; The fourth neural network is used for the second device to process the fourth signal to obtain a sixth signal; or, The second neural network or the fourth neural network is used by the second device to perform second signal processing, including: The second neural network is used for the second device to process the first signal to obtain the second signal; The fourth neural network is used for the second device to process the third signal to obtain the fourth signal; The first neural network or the third neural network is used by the first device to perform first signal processing, including: The first neural network is used for the first device to process the second signal to obtain the fifth signal; The third neural network is used by the first device to process the fourth signal to obtain the sixth signal.

45. The method of claim 44, wherein: The difference between the first training signal related to the third signal and the second training signal related to the sixth signal satisfies a first condition; or, The similarity between the first training signal and the second training signal satisfies a second condition.

46. ​​The method of claim 45, wherein: The difference degree includes any of the following: Difference, mean square error, normalized mean square error, or mean absolute error; The similarity includes a correlation coefficient.

47. The method according to any one of claims 44 to 46, characterized in that The first neural network or the second neural network is used to process the first signal to obtain the second signal, including: The first neural network or the second neural network and the first functional module are used to process the first signal to obtain the second signal; and / or, The first neural network or the second neural network is used to process the second signal to obtain the fifth signal, including: The first neural network or the second neural network and the second functional module are used to process the second signal to obtain the fifth signal.

48. The method according to any one of claims 44 to 46, characterized in that The third neural network or the fourth neural network is used to process the third signal to obtain the fourth signal, including: The third neural network or the fourth neural network and the third functional module are used to process the third signal to obtain the fourth signal; and / or, The third neural network or the fourth neural network is used to process the fourth signal to obtain the sixth signal, including: The third neural network or the fourth neural network and the fourth functional module are used to process the fourth signal to obtain the sixth signal.

49. The method according to any one of claims 42 to 46, characterized in that Before the second device receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network from the first device, the method further includes: The second device determines that the first neural network and the second neural network need to be updated; The second device sends first indication information to the first device, where the first indication information is used to instruct the first device to update the first neural network and the second neural network; or The method further comprises: The second device sends second indication information to the first device, where the second indication information is used to instruct the first device to determine whether the first neural network and the second neural network need to be updated.

50. The method of claim 49, wherein: The first indication information and / or the second indication information is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

51. The method of claim 49, wherein: The second device determining that the first neural network and the second neural network need to be updated includes: The second device periodically or non-periodically determines whether the first neural network and the second neural network need to be updated; or, The second device receives seventh indication information from the first device, where the seventh indication information is used to instruct the second device to determine whether the first neural network and the second neural network need to be updated; or The second device determines whether the first neural network and the second neural network need to be updated based on a preset condition trigger.

52. The method of claim 51, wherein: The seventh indication information is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

53. The method of claim 49, wherein: The second device determines whether the first neural network and the second neural network need to be updated, including: The second device determines whether the first neural network and the second neural network need to be updated according to whether the difference or similarity between the first information and the second information meets a third condition, wherein the first information includes a first judgment signal related to the first signal, and the second information includes a second judgment signal related to the fifth signal.

54. The method of claim 53, wherein: The method further comprises: The second device determines the second judgment signal based on the first judgment signal, the first neural network and the second neural network; or, The second device determines the second judgment signal based on the first judgment signal, the first neural network and the second neural network, and the first functional module and / or the second functional module.

55. The method of claim 53, wherein: The first determination signal includes downlink channel information sent by the first device to the second device, or, The first determination signal includes channel information measured based on a sounding reference signal SRS sent by the first device; The downlink channel information is carried on a preset resource or the downlink channel information is indicated to be used to determine the first judgment signal.

56. The method of any one of claims 43 to 46, wherein: The information of the third neural network includes at least one of the following information: The identification of the third neural network, information on the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, the identification of the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, the identification of the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, information on the network weights corresponding to the third neural network, the amount of change between the network weights corresponding to the third neural network and the network weights corresponding to the first neural network, the identification of the network weights corresponding to the third neural network, and third indication information, wherein the third indication information is used to indicate whether the network structure corresponding to the third neural network is the same as the network structure corresponding to the first neural network.

57. The method according to any one of claims 42 to 46, characterized in that The information of the fourth neural network includes at least one of the following information: The identification of the fourth neural network, information on the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identification of the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identification of the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, information on the network weights corresponding to the fourth neural network, the amount of change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network, the identification of the network weights corresponding to the fourth neural network, and fourth indication information, wherein the fourth indication information is used to indicate whether the network structure corresponding to the fourth neural network and the network structure corresponding to the second neural network are the same.

58. The method of claim 56, wherein: The amount of change between the network weights corresponding to the third neural network and the network weights corresponding to the first neural network, and / or the amount of change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network include one or more of the following information: Weight position information, weight interpretation rule indication information, weight value arrangement information and weight value change amount arrangement information; The information of the network structure corresponding to the third neural network and / or the information of the network structure corresponding to the fourth neural network includes: an indication of the type of neural network and an indication of the structure of the neural network; The information of the network weights corresponding to the third neural network and / or the information of the network weights corresponding to the fourth neural network include: Weight interpretation rule indication information and / or weight value arrangement information.

59. The method according to any one of claims 42 to 46, characterized in that Before the second device receives information of the fourth neural network from the first device, or before the second device receives information of the fourth neural network and information of the third neural network from the first device, the method further includes: The second device receives first information from the first device, where the first information is used to notify the second device to receive information about the fourth neural network, or to notify the second device to receive information about the fourth neural network and information about the third neural network; The second device sends a first response message to the first device, where the first response message is used to instruct the first device to send information about the fourth neural network, or to instruct the first device to send information about the fourth neural network and information about the third neural network.

60. The method of claim 59, wherein: The first information and / or the first response message is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

61. The method according to any one of claims 42 to 46, wherein: Before the second device receives information of the fourth neural network from the first device, or before the second device receives information of the fourth neural network and information of the third neural network from the first device, the method further includes: The second device sends second information to the first device, where the second information is used to indicate that the second device can receive information about the fourth neural network, or to indicate that the second device can receive information about the fourth neural network and information about the third neural network; The second device receiving information of the fourth neural network from the first device, or receiving information of the fourth neural network and information of the third neural network from the first device includes: After sending the second information, the second device receives information about the fourth neural network from the first device, or receives information about the fourth neural network and information about the third neural network from the first device.

62. The method of claim 61, wherein: After the second device sends the second information, receiving the information of the fourth neural network from the first device, or receiving the information of the fourth neural network and the information of the third neural network from the first device includes: After the second device sends the second information, it starts to receive information about the fourth neural network from the first device within a first preset time period after a first starting moment or after the first preset time period, or receives information about the fourth neural network and information about the third neural network from the first device.

63. The method of claim 61, wherein: The second information is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

64. The method of any one of claims 42 to 46, wherein: The second device receiving information of the fourth neural network from the first device, or receiving information of the fourth neural network and information of the third neural network from the first device includes: The method comprises: starting to receive information about the fourth neural network from the first device within a second preset time period after the second starting moment or after the second preset time period, or receiving information about the fourth neural network and information about the third neural network from the first device.

65. The method of any one of claims 42 to 46, wherein: Before the second device receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network from the first device, the method further includes: The second device receives first information from the first device, where the first information is used to notify the second device to receive information about the fourth neural network, or to notify the second device to receive information about the fourth neural network and information about the third neural network; The second device receiving the information of the fourth neural network from the first device, or receiving the information of the fourth neural network and the information of the third neural network from the first device includes: The second device starts receiving information about the fourth neural network from the first device within a third preset time period after receiving the first information from the first device or after the third preset time period, or receives information about the fourth neural network and information about the third neural network from the first device.

66. The method of any one of claims 42 to 46, wherein: After the second device receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network, the method further includes: The second device sends a first notification message to the first device, where the first notification message is used to instruct the second device to adopt the fourth neural network, or to adopt the fourth neural network and the third neural network to transmit information.

67. The method of claim 66, wherein: The first notification message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

68. The method of claim 59, wherein: The method further comprises: The second device sends a second notification message to the first device, or, The second device receives a second notification message from the first device, The second notification message is used to notify that the information transmission mode changes from a first mode to a second mode, the first mode includes information transmission based on a neural network, the neural network includes the first neural network and the second neural network, and the second mode includes information transmission based on a non-neural network.

69. The method of claim 68, wherein: The neural network also includes a seventh neural network and an eighth neural network.

70. The method of claim 68, wherein: The method further comprises: The second device switches the information transmission mode from the first mode to the second mode.

71. The method of claim 70, wherein: The second device switching the information transmission mode from the first mode to the second mode includes: The second device starts to switch the information transmission mode from the first mode to the second mode within a fourth preset time period after a fourth starting moment or after the fourth preset time period, wherein the fourth starting moment includes the moment when the second device sends the second notification message to the first device, or the moment when the second device receives the second notification message from the first device.

72. The method of claim 62, wherein: The first starting time and the second starting time include: The time when the second device sends the first indication information to the first device; or The moment when the second device receives the second notification message sent by the first device; or The time when the second device sends the second notification message to the first device; or The time when the second device receives the first signaling from the first device; or a time point at which the second device sends a second signaling to the first device, wherein the first signaling is used to indicate a timing start, and the second signaling is used to indicate an update start; The second notification message is used to notify that the information transmission mode is changed from the first mode to the second mode.

73. The method of claim 68, wherein: The second notification message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

74. The method of any one of claims 42 to 46, wherein: The method further comprises: The second device sends a second response message to the first device, where the second response message is used to indicate a refusal to receive information about the fourth neural network, or to indicate a refusal to receive information about the fourth neural network and information about the third neural network.

75. The method of claim 74, wherein: The second response message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

76. The method of any one of claims 42 to 46, wherein: The first device includes a terminal device or a network device, the second device includes the terminal device or the network device, When the first device is the terminal device, the second device is the network device, When the first device is the network device, the second device is the terminal device.

77. The method of any one of claims 42 to 46, wherein: The method further comprises: The second device sends fifth indication information to the first device, where the fifth indication information is used to indicate a capability of the first device to send the first device; The second device receives first capability information from the first device, where the first capability information includes at least one of the following information: computing capability information of the first device, storage capability information of the first device or information on whether the first device stores a first AI model, a neural network structure supported by the first device, a neural network scale supported by the first device, and an AI communication function supported by the first device, wherein the first AI model is any AI model; or The method further comprises: The second device receives sixth indication information from the first device, where the sixth indication information is used to indicate the second device to send a capability of the second device; The second device sends second capability information to the first device, where the second capability information includes at least one of the following information: The computing power information of the second device, the storage capacity information of the second device or whether the second device stores information of a second AI model, the neural network structure supported by the second device, the neural network scale supported by the second device, and the AI ​​communication function supported by the second device, wherein the second AI model is any AI model.

78. The method of any one of claims 44 to 46, wherein: The first signal includes channel information; The first neural network includes a first compressor, the second neural network includes a first decompressor, the third neural network includes a second compressor, and the fourth neural network includes a second decompressor; or, The first neural network includes a first decompressor, the second neural network includes a first compressor, the third neural network includes a second decompressor, and the fourth neural network includes a second compressor.

79. The method of claim 78, wherein: The channel information includes channel information received by multiple receiving antennas or channel information of multiple layers.

80. An apparatus for adjusting a neural network, comprising a first neural network and a second neural network, wherein: The first neural network is applied to the device side, and the second neural network is applied to the second device side, wherein the device comprises: A processing unit, configured to determine a third neural network and a fourth neural network, wherein the third neural network and the fourth neural network are respectively the first neural network and the second neural network after being updated or adjusted; a sending unit, configured to send information of the fourth neural network to the second device, The first neural network or the third neural network is used by the processing unit to perform first signal processing, and the second neural network or the fourth neural network is used by the second device to perform second signal processing, and the second signal processing corresponds to the first signal processing.

81. The device of claim 80, wherein: The first neural network or the third neural network is used by the processing unit to perform the first signal processing, including: The first neural network is used for the processing unit to process the first signal to obtain the second signal; The third neural network is used for the processing unit to process the third signal to obtain a fourth signal; The second neural network or the fourth neural network is used by the second device to perform second signal processing, including: The second neural network is used for the second device to process the second signal to obtain a fifth signal; The fourth neural network is used for the second device to process the fourth signal to obtain a sixth signal; or, The second neural network or the fourth neural network is used by the second device to perform second signal processing, including: The second neural network is used for the second device to process the first signal to obtain the second signal; The fourth neural network is used for the second device to process the third signal to obtain the fourth signal; The first neural network or the third neural network is used by the processing unit to perform the first signal processing, including: The first neural network is used for the processing unit to process the second signal to obtain the fifth signal; The third neural network is used by the processing unit to process the fourth signal to obtain the sixth signal.

82. The device according to claim 81, characterized in that The difference between the first training signal related to the third signal and the second training signal related to the sixth signal satisfies a first condition; or, The similarity between the first training signal and the second training signal satisfies a second condition.

83. The device of claim 82, wherein: The difference degree includes any of the following: Difference, mean square error, normalized mean square error, or mean absolute error; The similarity includes a correlation coefficient.

84. The device according to any one of claims 81 to 83, characterized in that The first neural network or the second neural network is used to process the first signal to obtain the second signal, including: The first neural network or the second neural network and the first functional module are used to process the first signal to obtain the second signal; and / or, The first neural network or the second neural network is used to process the second signal to obtain the fifth signal, including: The first neural network or the second neural network and the second functional module are used to process the second signal to obtain the fifth signal.

85. The device according to any one of claims 81 to 83, characterized in that The third neural network or the fourth neural network is used to process the third signal to obtain the fourth signal, including: The third neural network or the fourth neural network and the third functional module are used to process the third signal to obtain the fourth signal; and / or, The third neural network or the fourth neural network is used to process the fourth signal to obtain the sixth signal, including: The third neural network or the fourth neural network and the fourth functional module are used to process the fourth signal to obtain the sixth signal.

86. The device according to any one of claims 80 to 83, characterized in that The processing unit determines the third neural network and the fourth neural network including: The processing unit updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively.

87. The device of claim 86, wherein: Before the processing unit updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively, the device further includes: a receiving unit, configured to receive first indication information from the second device, wherein the first indication information is used to instruct the processing unit to update the first neural network and the second neural network; or The processing unit periodically or non-periodically determines whether the first neural network and the second neural network need to be updated, and the processing unit determines that the first neural network and the second neural network need to be updated; or, The receiving unit receives second indication information from a second device, the second indication information is used to instruct the processing unit to determine whether the first neural network and the second neural network need to be updated, and the processing unit determines that the first neural network and the second neural network need to be updated; or The processing unit determines whether the first neural network and the second neural network need to be updated based on a preset condition trigger, and the processing unit determines that the first neural network and the second neural network need to be updated.

88. The device of claim 87, wherein: The first indication information and / or the second indication information is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

89. The device of claim 86, wherein: The processing unit updates the first neural network and the second neural network including: The processing unit periodically or aperiodically updates the first neural network and the second neural network.

90. The device of claim 87, wherein: The processing unit determines whether the first neural network and the second neural network need to be updated, including: The processing determines whether the first neural network and the second neural network need to be updated based on whether the difference or similarity between the first information and the second information meets a third condition, wherein the first information includes a first judgment signal related to the first signal, and the second information includes a second judgment signal related to the fifth signal.

91. The device of claim 90, wherein: The processing unit is further configured to determine the second judgment signal based on the first judgment signal, the first neural network, and the second neural network; or The processing unit is further used to determine the second judgment signal based on the first judgment signal, the first neural network, the second neural network, and the first functional module and / or the second functional module.

92. The device of claim 90, wherein: The first determination signal includes the receiving unit receiving downlink channel information from the second device, or, The first determination signal includes channel information measured by the processing unit based on receiving a sounding reference signal SRS from the second device; The downlink channel information is carried on a preset resource or the downlink channel information is indicated to be used to determine the first judgment signal.

93. The device according to any one of claims 80 to 83, characterized in that The processing unit determines the third neural network and the fourth neural network including: The processing unit selects the third neural network and the fourth neural network from a first set based on the communication system data, wherein the first set includes a plurality of neural networks; or The processing unit adjusts the network weights corresponding to the first neural network and the second neural network respectively based on the communication system data to obtain the third neural network and the fourth neural network; or The processing unit selects a fifth neural network and a sixth neural network from the second set based on the communication system data, and then adjusts the network weights corresponding to the fifth neural network and the sixth neural network respectively based on the communication system data to obtain the third neural network and the fourth neural network.

94. The device of claim 93, wherein: The communication system data includes at least one of channel characteristics, signal-to-noise ratio, downlink channel information, uplink channel information and hybrid automatic repeat request HARQ information.

95. The device according to any one of claims 80 to 83, characterized in that The sending unit is also used to send information of the third neural network to the second device.

96. The device of claim 95, wherein: The information of the third neural network includes at least one of the following information: The identification of the third neural network, information on the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, the identification of the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, the identification of the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, information on the network weights corresponding to the third neural network, the amount of change between the network weights corresponding to the third neural network and the network weights corresponding to the first neural network, the identification of the network weights corresponding to the third neural network, and third indication information, wherein the third indication information is used to indicate whether the network structure corresponding to the third neural network is the same as the network structure corresponding to the first neural network.

97. The device according to any one of claims 80 to 83, characterized in that The information of the fourth neural network includes at least one of the following information: The identification of the fourth neural network, information on the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identification of the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identification of the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, information on the network weights corresponding to the fourth neural network, the amount of change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network, the identification of the network weights corresponding to the fourth neural network, and fourth indication information, wherein the fourth indication information is used to indicate whether the network structure corresponding to the fourth neural network and the network structure corresponding to the second neural network are the same.

98. The device of claim 96, wherein: The amount of change between the network weights corresponding to the third neural network and the network weights corresponding to the first neural network, and / or the amount of change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network include one or more of the following information: Weight position information, weight interpretation rule indication information, weight value arrangement information and weight value change amount arrangement information; The information of the network structure corresponding to the third neural network and / or the information of the network structure corresponding to the fourth neural network includes: an indication of the type of neural network and an indication of the structure of the neural network; The information of the network weights corresponding to the third neural network and / or the information of the network weights corresponding to the fourth neural network include: Weight interpretation rule indication information and / or weight value arrangement information.

99. The device according to any one of claims 80 to 83, characterized in that The sending unit sends the information of the fourth neural network to the second device, or, before sending the information of the fourth neural network and the information of the third neural network, the sending unit is further used to send first information to the second device, wherein the first information is used to request the second device to receive the information of the fourth neural network, or to request the second device to receive the information of the fourth neural network and the information of the third neural network; The apparatus also includes: a receiving unit, configured to receive a first response message from the second device, wherein the first response message is used to instruct the sending unit to send information about the fourth neural network, or to instruct the sending unit to send information about the fourth neural network and information about the third neural network.

100. The device of claim 99, wherein: The first information and / or the first response message is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

101. The device according to any one of claims 80 to 83, characterized in that The sending unit sends the information of the fourth neural network to the second device, or before sending the information of the fourth neural network and the information of the third neural network, The apparatus further includes: a receiving unit, configured to receive second information from the second device, wherein the second information is used to indicate that the second device can receive information about the fourth neural network, or to indicate that the second device can receive information about the fourth neural network and information about the third neural network; The sending unit is used to send the information of the fourth neural network to the second device, or the sending of the information of the fourth neural network and the information of the third neural network includes: After the receiving unit receives the second information, the sending unit sends the information of the fourth neural network to the second device, or sends the information of the fourth neural network and the information of the third neural network.

102. The device according to claim 101, characterized in that After the receiving unit receives the second information, the sending unit sends the information of the fourth neural network to the second device, or sending the information of the fourth neural network and the information of the third neural network includes: After the receiving unit receives the second information, the sending unit starts sending the information of the fourth neural network to the second device within a first preset time period after the first starting moment or after the first preset time period, or sends the information of the fourth neural network and the information of the third neural network.

103. The device according to claim 101, characterized in that The second information is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

104. The device according to any one of claims 80 to 83, characterized in that The sending unit sending the information of the fourth neural network to the second device, or sending the information of the fourth neural network and the information of the third neural network includes: The sending unit sends information of the fourth neural network to the second device, or sends information of the fourth neural network and information of the third neural network, after the processing unit updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively.

105. The device of claim 104, wherein: The sending unit, after the processing unit updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively, sends the information of the fourth neural network to the second device, or the sending of the information of the fourth neural network and the information of the third neural network includes: The sending unit, after the processing unit updates the first neural network and the second neural network to obtain the third neural network and the fourth neural network respectively, starts sending information of the fourth neural network to the second device within a second preset time period after a second starting moment or after the second preset time period, or sends information of the fourth neural network and information of the third neural network.

106. The device according to any one of claims 80 to 83, characterized in that The sending unit sends the information of the fourth neural network to the second device, or, before sending the information of the fourth neural network and the information of the third neural network, the sending unit is further used to send first information to the second device, where the first information is used to notify the second device to receive the information of the fourth neural network, or to notify the second device to receive the information of the fourth neural network and the information of the third neural network; The sending unit sending the information of the fourth neural network to the second device, or sending the information of the fourth neural network and the information of the third neural network includes: The sending unit starts sending the information of the fourth neural network to the second device within a third preset time period after sending the first information to the second device or after the third preset time period, or sends the information of the fourth neural network and the information of the third neural network.

107. The device according to any one of claims 80 to 83, characterized in that After the sending unit sends the information of the fourth neural network to the second device, or sends the information of the fourth neural network and the information of the third neural network, The apparatus further includes: a receiving unit, configured to receive a first notification message from the second device, wherein the first notification message is configured to instruct the second device to adopt the fourth neural network, or to adopt the fourth neural network and the third neural network for information transmission.

108. The device of claim 107, wherein: The first notification message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

109. The device of claim 99, wherein: The sending unit is further configured to send a second notification message to the second device, or The apparatus further includes: a receiving unit, configured to receive a second notification message from the second device, The second notification message is used to notify that the information transmission mode changes from a first mode to a second mode, the first mode includes information transmission based on a neural network, the neural network includes the first neural network and the second neural network, and the second mode includes information transmission based on a non-neural network.

110. The device of claim 109, wherein: The neural network also includes a seventh neural network and an eighth neural network.

111. The device of claim 109, wherein: The processing unit is further used to switch the information transmission mode from the first mode to the second mode.

112. The device according to claim 111, characterized in that The processing unit is further configured to switch the information transmission mode from the first mode to the second mode, comprising: The processing unit is also used to switch the information transmission mode from the first mode to the second mode within a fourth preset time period after a fourth starting moment or after a fourth preset time period, wherein the fourth starting moment includes the moment when the sending unit sends the second notification message to the second device, or the moment when the receiving unit receives the second notification message from the second device.

113. The device according to claim 102, characterized in that The first starting time and the second starting time include: The time when the receiving unit receives the first indication information from the second device; or The time at which the sending unit sends the second notification message to the second device; or The time when the receiving unit receives the second notification message from the second device; or The time when the sending unit sends the first signaling to the second device; or The moment when the receiving unit receives the second signaling from the second device, wherein the first signaling is used to indicate the start of timing, and the second signaling is used to indicate the start of updating; The second notification message is used to notify that the information transmission mode is changed from the first mode to the second mode.

114. The device of claim 109, wherein: The second notification message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

115. The device according to any one of claims 80 to 83, characterized in that The apparatus further includes: a receiving unit, configured to receive a second response message from the second device, wherein the second response message is configured to indicate a refusal to receive information of the fourth neural network, or to indicate a refusal to receive information of the fourth neural network and information of the third neural network.

116. The device according to claim 115, characterized in that The second response message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

117. The device according to any one of claims 80 to 83, characterized in that The apparatus includes a terminal device or a network device, the second device includes the terminal device or the network device, When the apparatus is the terminal device, the second device is the network device, When the apparatus is the network device, the second device is the terminal device.

118. The device according to any one of claims 80 to 83, characterized in that The apparatus further includes: a receiving unit, configured to receive fifth indication information from the second device, wherein the fifth indication information is used to indicate the capability of the sending unit to send the apparatus; The sending unit sends first capability information to the second device, where the first capability information includes at least one of the following information: computing capability information of the device, storage capability information of the device, or information on whether the device stores a first AI model, a neural network structure supported by the device, a neural network scale supported by the device, and an AI communication function supported by the device, wherein the first AI model is any one AI model; or The device also includes: The sending unit sends sixth indication information to the second device, where the sixth indication information is used to indicate the capability of the second device to send the second device; The receiving unit receives second capability information from the second device, where the second capability information includes at least one of the following information: The computing power information of the second device, the storage capacity information of the second device or whether the second device stores information of a second AI model, the neural network structure supported by the second device, the neural network scale supported by the second device, and the AI ​​communication function supported by the second device, wherein the second AI model is any AI model.

119. The device according to any one of claims 81 to 83, characterized in that The first signal includes channel information; The first neural network includes a first compressor, the second neural network includes a first decompressor, the third neural network includes a second compressor, and the fourth neural network includes a second decompressor; or, The first neural network includes a first decompressor, the second neural network includes a first compressor, the third neural network includes a second decompressor, and the fourth neural network includes a second compressor.

120. The device of claim 119, wherein: The channel information includes channel information received by multiple receiving antennas or channel information of multiple layers.

121. An apparatus for adjusting a neural network, comprising a first neural network and a second neural network, wherein: The first neural network is applied to the first device side, and the second neural network is applied to the device side, wherein the device comprises: a receiving unit, configured to receive information of a fourth neural network from the first device, wherein the fourth neural network is a neural network after the second neural network is updated or adjusted; a processing unit, configured to determine the fourth neural network based on information of the fourth neural network, Among them, the first neural network or the third neural network is used for the first device to perform first signal processing, the second neural network or the fourth neural network is used for the device to perform second signal processing, the second signal processing corresponds to the first signal processing, and the third neural network is the updated or adjusted first neural network determined by the first device.

122. The device according to claim 121, characterized in that The receiving unit is further used to receive information of a third neural network from the first device, where the third neural network is a neural network corresponding to the first neural network; The first neural network or the third neural network is used for the first device to perform first signal processing, and the second neural network or the fourth neural network is used for the processing unit to perform second signal processing, and the second signal processing corresponds to the first signal processing.

123. The device according to claim 122, characterized in that The first neural network or the third neural network is used by the first device to perform first signal processing, including: The first neural network is used for the first device to process the first signal to obtain the second signal; The third neural network is used for the first device to process the third signal to obtain a fourth signal; The second neural network or the fourth neural network is used by the processing unit to perform the second signal processing, including: The second neural network is used for the processing unit to process the second signal to obtain a fifth signal; The fourth neural network is used for the processing unit to process the fourth signal to obtain a sixth signal; or, The second neural network or the fourth neural network is used by the processing unit to perform the second signal processing, including: The second neural network is used for the processing unit to process the first signal to obtain the second signal; The fourth neural network is used for the processing unit to process the third signal to obtain the fourth signal; The first neural network or the third neural network is used by the first device to perform first signal processing, including: The first neural network is used for the first device to process the second signal to obtain the fifth signal; The third neural network is used by the first device to process the fourth signal to obtain the sixth signal.

124. The device according to claim 123, characterized in that The difference between the first training signal related to the third signal and the second training signal related to the sixth signal satisfies a first condition; or, The similarity between the first training signal and the second training signal satisfies a second condition.

125. The device of claim 124, wherein: The difference degree includes any of the following: Difference, mean square error, normalized mean square error, or mean absolute error; The similarity includes a correlation coefficient.

126. The device according to any one of claims 123 to 125, characterized in that The first neural network or the second neural network is used to process the first signal to obtain the second signal, including: The first neural network or the second neural network and the first functional module are used to process the first signal to obtain the second signal; and / or, The first neural network or the second neural network is used to process the second signal to obtain the fifth signal, including: The first neural network or the second neural network and the second functional module are used to process the second signal to obtain the fifth signal.

127. The device according to any one of claims 123 to 125, characterized in that The third neural network or the fourth neural network is used to process the third signal to obtain the fourth signal, including: The third neural network or the fourth neural network and the third functional module are used to process the third signal to obtain the fourth signal; and / or, The third neural network or the fourth neural network is used to process the fourth signal to obtain the sixth signal, including: The third neural network or the fourth neural network and the fourth functional module are used to process the fourth signal to obtain the sixth signal.

128. The device according to any one of claims 121 to 123, characterized in that Before the receiving unit receives the information of the fourth neural network from the first device, or before receiving the information of the fourth neural network and the information of the third neural network from the first device, the processing unit is further used to determine whether the first neural network and the second neural network need to be updated; The device also includes: a sending unit, configured to send first indication information to the first device, wherein the first indication information is used to instruct the first device to update the first neural network and the second neural network; or The sending unit is further used to send second indication information to the first device, where the second indication information is used to instruct the first device to determine whether the first neural network and the second neural network need to be updated.

129. The device of claim 128, wherein: The first indication information and / or the second indication information is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

130. The device of claim 128, wherein: The processing unit determines that the first neural network and the second neural network need to be updated, including: The processing unit periodically or non-periodically determines whether the first neural network and the second neural network need to be updated; or, The receiving unit receives seventh indication information from the first device, where the seventh indication information is used to instruct the processing unit to determine whether the first neural network and the second neural network need to be updated; or The processing unit determines whether the first neural network and the second neural network need to be updated based on a preset condition trigger.

131. The device of claim 130, wherein: The seventh indication information is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

132. The device of claim 128, wherein: The processing unit determines whether the first neural network and the second neural network need to be updated, including: The processing unit determines whether the first neural network and the second neural network need to be updated according to whether the difference or similarity between the first information and the second information meets a third condition, wherein the first information includes a first judgment signal related to the first signal, and the second information includes a second judgment signal related to the fifth signal.

133. The device of claim 132, wherein: The processing unit determines the second judgment signal based on the first judgment signal, the first neural network and the second neural network; or The processing unit determines the second judgment signal based on the first judgment signal, the first neural network and the second neural network, and the first functional module and / or the second functional module.

134. The device of claim 132, wherein: The first determination signal includes downlink channel information received by the receiving unit, or, The first determination signal includes channel information measured based on a sounding reference signal SRS received by the receiving unit; The downlink channel information is carried on a preset resource or the downlink channel information is indicated to be used to determine the first judgment signal.

135. The device according to any one of claims 122 to 125, characterized in that The information of the third neural network includes at least one of the following information: The identification of the third neural network, information on the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, the identification of the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, the identification of the network structure corresponding to the third neural network and information on the network weights corresponding to the third neural network, information on the network weights corresponding to the third neural network, the amount of change between the network weights corresponding to the third neural network and the network weights corresponding to the first neural network, the identification of the network weights corresponding to the third neural network, and third indication information, wherein the third indication information is used to indicate whether the network structure corresponding to the third neural network is the same as the network structure corresponding to the first neural network.

136. The device according to any one of claims 121 to 125, characterized in that The information of the fourth neural network includes at least one of the following information: The identification of the fourth neural network, information on the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identification of the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, the identification of the network structure corresponding to the fourth neural network and information on the network weights corresponding to the fourth neural network, information on the network weights corresponding to the fourth neural network, the amount of change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network, the identification of the network weights corresponding to the fourth neural network, and fourth indication information, wherein the fourth indication information is used to indicate whether the network structure corresponding to the fourth neural network and the network structure corresponding to the second neural network are the same.

137. The device of claim 135, wherein: The amount of change between the network weights corresponding to the third neural network and the network weights corresponding to the first neural network, and / or the amount of change between the network weights corresponding to the fourth neural network and the network weights corresponding to the second neural network include one or more of the following information: Weight position information, weight interpretation rule indication information, weight value arrangement information and weight value change amount arrangement information; The information of the network structure corresponding to the third neural network and / or the information of the network structure corresponding to the fourth neural network includes: an indication of the type of neural network and an indication of the structure of the neural network; The information of the network weights corresponding to the third neural network and / or the information of the network weights corresponding to the fourth neural network include: Weight interpretation rule indication information and / or weight value arrangement information.

138. The device according to any one of claims 121 to 125, characterized in that Before the receiving unit receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network from the first device, the receiving unit is further used to receive first information from the first device, wherein the first information is used to notify the receiving unit to receive the information of the fourth neural network, or to notify the receiving unit to receive the information of the fourth neural network and the information of the third neural network; The apparatus also includes: a sending unit, configured to send a first response message to the first device, wherein the first response message is used to instruct the first device to send information about the fourth neural network, or to instruct the first device to send information about the fourth neural network and information about the third neural network.

139. The device of claim 138, wherein: The first information and / or the first response message is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

140. The device according to any one of claims 121 to 125, characterized in that Before the receiving unit receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network from the first device, the sending unit sends second information to the first device, where the second information is used to indicate that the receiving unit can receive the information of the fourth neural network, or receive the information of the fourth neural network and the information of the third neural network; The receiving unit receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network from the first device, including: After the sending unit sends the second information, the receiving unit receives information about the fourth neural network from the first device, or receives information about the fourth neural network and information about the third neural network from the first device.

141. The device of claim 140, wherein: After the sending unit sends the second information, the receiving unit receives the information of the fourth neural network from the first device, or the receiving of the information of the fourth neural network and the information of the third neural network from the first device includes: After the sending unit sends the second information, the receiving unit starts to receive the information of the fourth neural network from the first device within a first preset time period after a first starting moment or after the first preset time period, or receives the information of the fourth neural network and the information of the third neural network from the first device.

142. The device of claim 140, wherein: The second information is carried in an uplink physical control channel PUCCH, a downlink physical control channel PDCCH, a physical uplink shared channel PUSCH, a physical downlink shared channel PDSCH, an uplink grant UL grant or a random access channel RACH.

143. The device according to any one of claims 121 to 125, characterized in that The receiving unit receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network from the first device, including: The receiving unit starts receiving information about the fourth neural network from the first device within a second preset time period after a second starting moment or after the second preset time period, or receives information about the fourth neural network and information about the third neural network from the first device.

144. The device according to any one of claims 121 to 125, characterized in that Before the receiving unit receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network from the first device, the receiving unit receives first information from the first device, wherein the first information is used to notify the receiving unit to receive the information of the fourth neural network, or to receive the information of the fourth neural network and the information of the third neural network; The receiving unit receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network from the first device, including: The receiving unit starts receiving information of the fourth neural network from the first device within a third preset time period or after the third preset time period after receiving the first information from the first device, or receives information of the fourth neural network and information of the third neural network from the first device.

145. The device according to any one of claims 121 to 125, characterized in that After the receiving unit receives the information of the fourth neural network from the first device, or receives the information of the fourth neural network and the information of the third neural network, The apparatus also includes: a sending unit, configured to send a first notification message to the first device, wherein the first notification message is used to instruct the processing unit to adopt the fourth neural network, or to adopt the fourth neural network and the third neural network for information transmission.

146. The device of claim 145, wherein: The first notification message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

147. The device of claim 138, wherein: The sending unit sends a second notification message to the first device, or, The receiving unit receives a second notification message from the first device, The second notification message is used to notify that the information transmission mode changes from a first mode to a second mode, the first mode includes information transmission based on a neural network, the neural network includes the first neural network and the second neural network, and the second mode includes information transmission based on a non-neural network.

148. The device of claim 147, wherein: The neural network also includes a seventh neural network and an eighth neural network.

149. The device of claim 147, wherein: The processing unit switches the information transmission mode from the first mode to the second mode.

150. The device of claim 149, wherein: The processing unit switches the information transmission mode from the first mode to the second mode, comprising: The processing unit starts to switch the information transmission mode from the first mode to the second mode within a fourth preset time period after a fourth starting moment or after the fourth preset time period, wherein the fourth starting moment includes the moment when the sending unit sends the second notification message to the first device, or the moment when the receiving unit receives the second notification message from the first device.

151. The device of claim 141, wherein: The first starting time and the second starting time include: The time when the sending unit sends the first indication information to the first device; or The receiving unit receives the time when the first device sends the second notification message; or The time when the sending unit sends the second notification message to the first device; or The time when the receiving unit receives the first signaling from the first device; or A time at which the sending unit sends a second signaling to the first device, wherein the first signaling is used to indicate a timing start, and the second signaling is used to indicate an update start; The second notification message is used to notify that the information transmission mode is changed from the first mode to the second mode.

152. The device of claim 147, wherein: The second notification message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

153. The device according to any one of claims 121 to 125, characterized in that The apparatus further includes: a sending unit, configured to send a second response message to the first device, wherein the second response message is used to indicate a refusal to receive information of the fourth neural network, or to indicate a refusal to receive information of the fourth neural network and information of the third neural network.

154. The device of claim 153, wherein: The second response message is carried in the uplink physical control channel PUCCH, the downlink physical control channel PDCCH, the physical uplink shared channel PUSCH, the physical downlink shared channel PDSCH, the uplink grant UL grant or the random access channel RACH.

155. The device according to any one of claims 121 to 125, characterized in that The first device includes a terminal device or a network device, the apparatus includes the terminal device or the network device, When the first device is the terminal device, the apparatus is the network device, When the first device is the network device, the apparatus is the terminal device.

156. The device according to any one of claims 121 to 125, characterized in that The apparatus further includes: a sending unit, configured to send fifth indication information to the first device, wherein the fifth indication information is used to indicate the capability of the first device to send the first device; The receiving unit receives first capability information from the first device, where the first capability information includes at least one of the following information: computing capability information of the first device, storage capability information of the first device or information on whether the first device stores a first AI model, a neural network structure supported by the first device, a neural network scale supported by the first device, and an AI communication function supported by the first device, wherein the first AI model is any AI model; or The device also includes: The receiving unit receives sixth indication information from the first device, where the sixth indication information is used to indicate the sending unit to send the capability of the device; The sending unit sends second capability information to the first device, where the second capability information includes at least one of the following information: The computing power information of the device, the storage capacity information of the device or whether the device stores information of a second AI model, the neural network structure supported by the device, the neural network scale supported by the device, and the AI ​​communication function supported by the device, wherein the second AI model is any one AI model.

157. The device according to any one of claims 123 to 125, characterized in that The first signal includes channel information; The first neural network includes a first compressor, the second neural network includes a first decompressor, the third neural network includes a second compressor, and the fourth neural network includes a second decompressor; or, The first neural network includes a first decompressor, the second neural network includes a first compressor, the third neural network includes a second decompressor, and the fourth neural network includes a second compressor.

158. The device of claim 157, wherein: The channel information includes channel information received by multiple receiving antennas or channel information of multiple layers.

159. A communication system comprising the apparatus for adjusting a neural network as claimed in any one of claims 80-120 and the apparatus for adjusting a neural network as claimed in any one of claims 121-158.

160. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed, enables the computer to implement the method according to any one of claims 1 to 41, or enables the computer to implement the method according to any one of claims 42 to 79.

161. A chip device, characterized in that: It comprises a processing circuit, which is used to call and run a program from a memory, so that a communication device equipped with the chip device executes a method as described in any one of claims 1 to 41, or so that the communication device equipped with the chip device executes a method as described in any one of claims 42 to 79.

Images