Transmission method and device

Abstract

An embodiment of the present invention provides a transmission method and device, the method comprising: determining a neural network model of a transmitter or receiver of the first device according to the interaction between the first device and the second device, the neural network model of the transmitter or receiver being any one of a set of neural network models of the transmitter or receiver of the first device; and performing data transmission with the second device through the neural network model of the transmitter or receiver. In an embodiment of the present invention, a transmitter or receiver in a communication system can select a suitable neural network model for data transmission according to the current channel environment, realize the function of a transceiver through the neural network model, improve the flexibility of the system structure, and reduce the deployment cost of the system.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of communication technology, and in particular to a transmission method and device. Background Art

[0002] The design of the physical layer of wireless communications has undergone decades of development and is currently highly modular: the transmission link is composed of a cascade of several independent modules with different functions, and each module is responsible for completing a specific signal processing process.

[0003] See also Figure 1 The original message passes through the channel after source coding, channel coding, modulation and other processes, and then reaches the receiving end and goes through channel equalization (including estimation of various offset factors in the channel), demodulation, channel decoding, source decoding and other processes to obtain the final received message. In the actual system, considering the signaling interaction with the upper layer, resource mapping, and the use of different transmission technologies (such as multi-antenna technology, etc.), the specific module distribution and functions will change accordingly.

[0004] With the evolution of technology, the number of modules in the physical layer link has gradually increased, which has greatly increased the complexity of the interface and signal flow between modules, reduced the flexibility of the system structure, and increased the deployment cost of the system. Summary of the invention

[0005] An object of an embodiment of the present invention is to provide a transmission method and device to solve the problem that the modularization of physical layer links leads to poor system structure flexibility and increased system deployment costs.

[0006] In a first aspect, an embodiment of the present invention provides a transmission method, applied to a first device, including:

[0007] Determine a transmitting end or receiving end neural network model of the first device according to the interaction between the first device and the second device, wherein the transmitting end or receiving end neural network model is any group in a transmitting end or receiving end neural network model set of the first device;

[0008] Data is transmitted to the second device via the transmitting or receiving neural network model.

[0009] Optionally, the method further comprises:

[0010] Acquire a channel type and communication parameters corresponding to the channel type;

[0011] According to the channel type and communication parameters, simulated channel training is performed to obtain a set of transmitting end or receiving end neural network models corresponding to the channel type.

[0012] Optionally, the channel type includes one of the following: an indoor channel, an outdoor channel, a line-of-sight channel, a non-line-of-sight channel, a low-frequency band channel, and a high-frequency band channel.

[0013] Optionally, the communication parameters include one or more of the following: signal-to-noise ratio, supported rate, number of supported data streams, and number of supported channels.

[0014] Optionally, the first device is a receiving end, and the second device is a sending end;

[0015] The determining, according to the interaction between the first device and the second device, a receiving end neural network model of the first device includes:

[0016] receiving a first signal from the second device;

[0017] Selecting a set of receiving-end neural network models from a set of receiving-end neural network models of the first device according to the first signal;

[0018] or,

[0019] Selecting, according to the first signal, a plurality of groups of receiving-end neural network models from a set of receiving-end neural network models of the first device;

[0020] Obtaining a set of transmitting-end neural network models selected by the second device from a set of transmitting-end neural network models of the second device;

[0021] According to a group of transmitting-end neural network models selected by the second device, a group of receiving-end neural network models is selected from the multiple groups of receiving-end neural network models.

[0022] Optionally, the method further comprises:

[0023] Sending first information to the second device;

[0024] The first information includes one or more of the following:

[0025] Information of one or more groups of receiving-end neural network models selected by the first device based on the first signal;

[0026] The first device configures the second device with information on whether to adjust the neural network model of the transmitting end;

[0027] Information used to adjust the loss function of the receiving neural network model.

[0028] Optionally, the method further comprises:

[0029] receiving second information from the second device;

[0030] The second information includes one or more of the following:

[0031] Information about a set of transmitting-end neural network models selected by the second device;

[0032] The second device configures the first device with information on whether to adjust the neural network model of the transmitting end;

[0033] Information about a reference signal used to adjust a neural network model at the receiving end;

[0034] Information used to adjust the loss function of the receiving neural network model.

[0035] Optionally, the first device is a transmitting end, and the second device is a receiving end;

[0036] The determining, according to the interaction between the first device and the second device, a transmitting end neural network model of the first device includes:

[0037] Obtaining one or more groups of receiving-end neural network models selected by the second device from a set of receiving-end neural network models of the second device;

[0038] According to one or more groups of receiving-end neural network models selected by the second device, a group of transmitting-end neural network models is selected from the set of transmitting-end neural network models of the first device.

[0039] Optionally, the method further comprises:

[0040] Sending third information to the second device, where the third information includes one or more of the following:

[0041] Information about a set of transmitting-end neural network models selected by the first device;

[0042] Information about a reference signal used to adjust a neural network model at the receiving end;

[0043] Information used to adjust the loss function of the receiving neural network model;

[0044] The first device configures the second device with information on whether to adjust the neural network model of the receiving end.

[0045] Optionally, transmitting data to the second device through the transmitting end or receiving end neural network model includes:

[0046] Adjusting the selected neural network model of the transmitting end or the receiving end;

[0047] Data is transmitted to the second device through the adjusted transmitting end or receiving end neural network model.

[0048] Optionally, the first device is a receiving end, the second device is a sending end, and the adjusting the selected sending end neural network model includes:

[0049] receiving a reference signal for adjusting a receiving-end neural network model from the second device, wherein the reference signal is sent by the second device using a set of transmitting-end neural network models selected by the second device;

[0050] Adjusting a set of receiving-end neural network models selected by the first device according to a loss function for adjusting the receiving-end neural network model and the reference signal;

[0051] The loss function is provided by the second device.

[0052] Optionally, the method further comprises:

[0053] If the first device configures the second device to adjust the transmitting end neural network model, a loss function value is sent to the second device, where the loss function value is calculated by the first device based on the loss function and the reference signal.

[0054] Optionally, the first device is a transmitting end, the second device is a receiving end, and the adjusting the selected transmitting end neural network model includes:

[0055] receiving a loss function value from the second device;

[0056] Adjusting a set of transmitting-end neural network models selected by the first device according to the loss function value;

[0057] Among them, the loss function value is calculated by the second device based on the loss function used to adjust the transceiver neural network model and the received reference signal used to adjust the transceiver neural network model; the loss function is provided by the first device; and the reference signal is sent to the second device by the first device using a group of transceiver neural network models selected by it.

[0058] Optionally, the performing data transmission with the second device through the adjusted transmitting end or receiving end neural network model includes:

[0059] When preset conditions are met, data is transmitted with the second device through the adjusted transmitting or receiving neural network model.

[0060] Optionally, the preset condition includes one or more of the following:

[0061] The number of adjustments of the neural network model at the transmitting end or the receiving end reaches a preset value;

[0062] Receiving an instruction to stop adjusting the neural network model of the transmitting end or the receiving end;

[0063] The adjusted transmitting or receiving end neural network model meets the transmission requirements of the first device.

[0064] In a second aspect, an embodiment of the present invention further provides a communication device, wherein the communication device is a first device, and the first device includes:

[0065] A determination module, configured to determine a sender or receiver neural network model of the first device according to the interaction between the first device and the second device, wherein the sender or receiver neural network model is any one of a sender or receiver neural network model set of the first device;

[0066] A transmission module is used to transmit data with the second device through the transmitting end or receiving end neural network model.

[0067] In a third aspect, an embodiment of the present invention further provides a communication device, wherein the communication device is a first device, and the first device includes: a processor;

[0068] The processor reads the program in the memory and performs the following operations: determining a transmitting or receiving neural network model of the first device according to the interaction between the first device and the second device, wherein the transmitting or receiving neural network model is any group in a set of transmitting or receiving neural network models of the first device; and transmitting data with the second device through the transmitting or receiving neural network model.

[0069] In a fourth aspect, an embodiment of the present invention further provides a communication device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, steps of the transmission method as described in the first aspect are implemented.

[0070] In a fifth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the transmission method described in the first aspect are implemented.

[0071] In an embodiment of the present invention, a transmitting end or a receiving end in a communication system can select a suitable neural network model for data transmission according to the current channel environment, realize the functions of a transceiver through the neural network model, improve the flexibility of the system structure, and reduce the deployment cost of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0073] Figure 1 It is a schematic diagram of the physical layer link module;

[0074] Figure 2a and Figure 2b Schematic diagram of an end-to-end communication system based on a neural network;

[0075] Figure 3 is a schematic diagram of a wireless communication system to which an embodiment of the present invention is applied;

[0076] Figure 4 This is a flow chart of a transmission method according to an embodiment of the present invention;

[0077] Figure 5a and Figure 5b The second flowchart of the transmission method according to the embodiment of the present invention;

[0078] Figure 6 This is a schematic diagram of the signal processing flow at the traditional LTE transmitter;

[0079] Figure 7 It is a schematic diagram of a signal processing flow of a physical layer link downlink transmitting end based on a neural network in an embodiment of the present invention;

[0080] Figure 8 Schematic diagram of an offline training model set in an embodiment of the present invention;

[0081] Fig. 9 The third flowchart of the transmission method according to the embodiment of the present invention;

[0082] Fig.10 is one of the schematic diagrams of the first device according to an embodiment of the present invention;

[0083] Fig.11 This is a second schematic diagram of the first device according to an embodiment of the present invention;

[0084] Fig.12 This is the third schematic diagram of the first device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0085] See also Figure 2a and Figure 2b , illustrating the implementation framework of the end-to-end communication system implemented by the neural network, where Figure 2a The transmitter and receiver are completely realized by neural network. Figure 2bThe transmitter and receiver are realized by combining neural network and traditional signal processing modules.

[0086] It should be noted that in the actual system, it is not necessary to press Figure 2a As shown, the entire transmitter (receiver) is implemented using a neural network; alternatively, only some physical layer functions, such as channel coding-modulation-MIMO, can be implemented using a neural network, while the remaining functions, such as OFDM modulation, are implemented in a traditional way, such as Figure 2b The main advantage of using neural networks to implement physical layer transceivers is that it can achieve joint optimization among multiple modules, and neural networks are naturally highly parallel and can be accelerated by dedicated hardware.

[0087] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0088] The term "comprise" and any variation thereof in the specification and claims of the present application are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices. In addition, the use of "and / or" in the specification and claims to indicate at least one of the connected objects, such as A and / or B, means that A alone, B alone, and A and B exist in three cases.

[0089] In the embodiments of the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0090] The technology described in this article is not limited to Long Time Evolution (LTE) / LTE-Advanced (LTE-A) systems, and can also be used in various wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency Division Multiple Access (SC-FDMA) and other systems.

[0091] The terms "system" and "network" are often used interchangeably. A CDMA system may implement radio technologies such as CDMA2000, Universal Terrestrial Radio Access (UTRA). UTRA includes Wideband Code Division Multiple Access (WCDMA) and other CDMA variants. A TDMA system may implement radio technologies such as Global System for Mobile Communication (GSM). An OFDMA system may implement radio technologies such as Ultra Mobile Broadband (UMB), Evolution-UTRA (E-UTRA), IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20, Flash-OFDM, etc. UTRA and E-UTRA are parts of Universal Mobile Telecommunications System (UMTS). LTE and more advanced LTE (such as LTE-A) are new UMTS versions that use E-UTRA. UTRA, E-UTRA, UMTS, LTE, LTE-A, and GSM are described in documents from an organization named "3rd Generation Partnership Project" (3GPP). CDMA2000 and UMB are described in documents from an organization named "3rd Generation Partnership Project 2" (3GPP2). The techniques described herein may be used for the systems and radio technologies mentioned above as well as for other systems and radio technologies.

[0092] See also Figure 3 The following describes an embodiment of the present invention in conjunction with the accompanying drawings. A transmission method and device provided by the embodiment of the present invention can be applied to a wireless communication system. Figure 3 , is a schematic diagram of the architecture of a wireless communication system provided by an embodiment of the present invention. Figure 3 As shown, the wireless communication system may include: a network device 31 and a terminal 32, the terminal 32 may be denoted as UE32, and the terminal 32 may communicate with the network device 31 (transmit signaling or transmit data). In practical applications, the connection between the above-mentioned devices may be a wireless connection. In order to conveniently and intuitively represent the connection relationship between the devices, Figure 3 Solid lines are used to indicate.

[0093] The network device 31 provided in an embodiment of the present invention may be a base station, which may be a commonly used base station, an evolved node base station (eNB), or a network device in a 5G system (for example, a next generation node base station (gNB) or a transmission and reception point (TRP)).

[0094] The terminal 32 provided in the embodiment of the present invention can be a mobile phone, a tablet computer, a laptop computer, an ultra-mobile personal computer (UMPC), a netbook or a personal digital assistant (PDA), a mobile Internet device (MID), a wearable device (Wearable Device) or a vehicle-mounted device, etc.

[0095] See also Figure 4 , an embodiment of the present invention provides a transmission method, the execution subject of the method may be a first device, the first device may be a transmitting end, or the first device may also be a receiving end:

[0096] Step 401: Determine a transmitting end or receiving end neural network model of the first device according to the interaction between the first device and the second device, wherein the transmitting end or receiving end neural network model is any group in a transmitting end or receiving end neural network model set of the first device;

[0097] In an embodiment of the present invention, the neural network model of the transmitting end or the receiving end refers to a model established by using a neural network to implement some functions in the transmitting end or the receiving end. For example, some functions include but are not limited to: coding, scrambling, modulation mapping, layer mapping, and precoding.

[0098] The set of neural network models of the transmitting end or the receiving end is obtained by the first device based on simulation training in different communication scenarios, where different communication scenarios correspond to different channel types and / or communication parameters. Exemplarily, the set of neural network models of the transmitting end or the receiving end is obtained by the first device through offline training.

[0099] Step 402: Transmit data with the second device through the transmitting or receiving neural network model.

[0100] For example, in a downlink system, a first device (such as a base station) transmits data to a second device (such as a terminal) through a transmitting neural network model, or, in an uplink system, a first device (such as a terminal) transmits data to a second device (such as a base station) through a transmitting neural network model.

[0101] In some embodiments, the set of neural network models at the transmitting end or the receiving end is obtained by the first device based on simulation training in different communication scenarios, where different communication scenarios correspond to different channel types and / or communication parameters.

[0102] Optionally, a set of neural network models of the transmitting end or the receiving end is obtained by offline training, specifically including: obtaining a channel type and communication parameters corresponding to the channel type; performing simulation channel training according to the channel type and communication parameters to obtain a set of neural network models of the transmitting end or the receiving end corresponding to the channel type.

[0103] For example, simulation channel training is performed based on the simulation channel model corresponding to the channel type and the communication parameters. It can be understood that during the training process, the communication parameters can be adjusted to obtain multiple transmitter or receiver neural network models of the corresponding channel type (i.e., a set of transmitter or receiver neural network models).

[0104] The channel type is used to represent the communication environment, and the channel type may include one of the following: indoor channel, outdoor channel, line-of-sight channel, non-line-of-sight channel, low-frequency band channel, high-frequency band channel, etc.

[0105] The communication parameters are used to represent relevant parameters of the communication environment, and the communication parameters may include one or more of the following: signal-to-noise ratio, supported rate, number of supported data streams, number of supported channels, etc.

[0106] The transmitter or receiver neural network model set trained in the above manner can meet the transmission performance requirements of different wireless channels, and can cope with complex, changeable and rapidly changing wireless channels. The transmitter or receiver neural network model set is obtained by adjusting the communication parameters during the training process, thereby reducing the training overhead and the difficulty of deploying the transmitter or receiver neural network model set.

[0107] In an embodiment of the present invention, the first device may determine a set of transmitting-end or receiving-end neural network models for data transmission based on interaction with the second device.

[0108] The interaction includes: receiving a signal sent by the second device, which is a signal provided by the transmitter to the receiver for channel estimation or channel detection, such as a reference signal. Alternatively, the interaction may also include: receiving a result of the second device selecting a transmitter or receiver neural network model fed back by the second device.

[0109] (I) Taking the first device as the receiving end and the second device as the sending end as an example, this article introduces how the first device selects the receiving end neural network model.

[0110] In some embodiments, step 401 may include: receiving a first signal from the second device; and selecting a set of receiving-end neural network models from a set of receiving-end neural network models of the first device based on the first signal.

[0111] Optionally, the first signal includes: a system broadcast message and / or a reference signal.

[0112] In other embodiments, step 401 may include: selecting multiple groups of receiving-end neural network models from the set of receiving-end neural network models of the first device based on the first signal, and obtaining a group of transmitting-end neural network models selected by the second device from the set of transmitting-end neural network models of the second device; and selecting a group of receiving-end neural network models from the multiple groups of receiving-end neural network models based on the group of transmitting-end neural network models.

[0113] In some embodiments, after the first device selects one or more sets of receiving-end neural network models, the method further includes:

[0114] Sending first information to the second device, where the first information includes one or more of the following:

[0115] (1) Information about one or more groups of receiving-end neural network models selected by the first device based on the first signal, and the information may be an index of one or more groups of receiving-end neural network models in a collection.

[0116] It is understandable that the second device can select a corresponding set of transceiver neural network models from the set of transmitter neural network models of the second device based on the information.

[0117] (2) Information about the loss function used to adjust the receiving end neural network model. For example, the information about the loss function may be the loss function itself, or may be an index of the loss function in a set.

[0118] It can be understood that the loss function is used to represent the optimization target for adjusting the receiving end neural network model. Optionally, the loss function includes but is not limited to: receiving end bit-level log-likelihood ratio (LLR); receiving end symbol-level log-likelihood ratio, etc.

[0119] (3) The first device configures the second device to adjust the information of the neural network model of the sending end

[0120] The first information may be control information, such as uplink control information, but is certainly not limited thereto.

[0121] The above adjustment can be understood as fine-tuning or optimization, that is, according to the sending or receiving conditions of the real-time signal, the previously selected receiving end neural network model is fine-tuned or optimized, so that the fine-tuned or optimized receiving end neural network model can better meet the channel transmission between the sending end and the receiving end.

[0122] In some embodiments, the method may further include:

[0123] Receive second information from the second device; wherein the second information includes one or more of the following:

[0124] (1) Information about a group of transmitting-end neural network models selected by the second device, where the group of transmitting-end neural network models is selected by the second device from a set of transmitting-end neural network models of the second device according to a result of selecting a receiving-end neural network model reported by the first device.

[0125] It can be understood that this step is an optional step, and only when the first device selects multiple groups of receiving-end neural network models based on the first signal, the first device needs to obtain the selection result of the second device.

[0126] (2) information about a reference signal used to adjust a neural network model at the receiving end, which information may be a type of the reference signal;

[0127] That is, the second device can specify a reference signal for adjusting the receiving-end neural network model to the first device.

[0128] (3) Information about the loss function used to adjust the receiving end neural network model. For example, the information about the loss function may be the loss function itself, or may be an index of the loss function in the set.

[0129] It can be understood that the loss function is used as the optimization target for adjusting the neural network model at the receiving end. Optionally, the loss function includes but is not limited to: bit-level log-likelihood ratio (LLR) at the receiving end; symbol-level log-likelihood ratio at the receiving end, etc.

[0130] (4) The second device configures the first device with information on whether to adjust the neural network model at the receiving end.

[0131] That is, the first device can allow the second device to adjust the receiving end neural network model, or prohibit the second device from adjusting the receiving end neural network model, so that the computational burden of the second device can be flexibly adjusted.

[0132] The above adjustment can be understood as fine-tuning or optimization, that is, according to the sending or receiving conditions of the real-time signal, the previously selected receiving end neural network model is fine-tuned or optimized, so that the fine-tuned or optimized receiving end neural network model can better meet the channel transmission between the sending end and the receiving end.

[0133] It can be understood that the second information can be control information, but is certainly not limited to this.

[0134] In order to make the receiving end or sending end neural network model selected by the first device and the second device respectively meet the channel communication requirements between the sending end and the receiving end, in an embodiment of the present invention, the first device can feed back its selection result to the second device, and the second device selects the sending end neural network model according to the selection result of the first device. If the first device has previously selected multiple groups of receiving end neural network models, the second device will feed back its selected sending end neural network model to the first device, and the first device will select one from the multiple groups of receiving end neural network models according to the group of sending end neural network models selected by the second device.

[0135] (ii) Taking the first device as the transmitter and the second device as the receiver as an example, this paper introduces how the first device determines the neural network model of the transmitter.

[0136] In some embodiments, step 401 may include: obtaining one or more groups of receiving-end neural network models selected by the second device from a set of receiving-end neural network models of the second device; and selecting a group of transmitting-end neural network models from a set of transmitting-end neural network models of the first device based on the one or more groups of receiving-end neural network models.

[0137] Optionally, after selecting a set of transmitting-end neural network models, the method further comprises:

[0138] Sending third information to the second device, where the third information includes one or more of the following:

[0139] (1) information of a group of transmitting-end neural network models selected by the first device;

[0140] It can be understood that (1) is optional.

[0141] (2) information of a reference signal used to adjust a neural network model at the receiving end;

[0142] (3) information used to adjust the loss function of the receiving neural network model;

[0143] (4) The first device configures the second device with information on whether to adjust the neural network model at the receiving end.

[0144] The third information mentioned above may be uplink control information, but is certainly not limited thereto.

[0145] It can be understood that after the first device selects the transmitting neural network model for transmitting data, the data can be transmitted through the selected transmitting neural network model, or the first device can also adjust the previously selected transmitting neural network model through real-time data, and then transmit the data through the adjusted transmitting neural network model.

[0146] Thus, in the embodiment of the present invention, the transmitting end neural network model is trained in at least two stages, that is, the transmitting end neural network model set is first trained offline using the simulated channel model, and then the transmitting end neural network model is fine-tuned and optimized using the actual channel data. On the one hand, the complexity of online training and the amount of data required are reduced, thereby controlling the cost of online training within an acceptable range; on the other hand, the analytical form of the simulated channel model enables gradient transfer.

[0147] (I) Taking the first device as the receiving end and the second device as the sending end as an example, how the first device adjusts the transceiver neural network model is introduced.

[0148] Optionally, the first device adjusts the selected receiving end neural network model in a manner including:

[0149] receiving a reference signal for adjusting a receiving-end neural network model from the second device, the reference signal being sent by the second device using a group of transmitting-end neural network models selected by the second device; adjusting a group of receiving-end neural network models selected by the first device according to a loss function for adjusting the receiving-end neural network model and the reference signal;

[0150] The loss function includes an objective function required to adjust the neural network model at the receiving end, and the loss function is provided by the second device.

[0151] Optionally, if the first device configures the second device to adjust the transmitting end neural network model, a loss function value is sent to the second device, where the loss function value is calculated by the first device according to the loss function and the reference signal;

[0152] If the first device configures the second device not to adjust the sending end neural network model, the loss function value is not fed back to the second device or a null value is fed back.

[0153] (ii) Taking the first device as the transmitting end and the second device as the receiving end as an example, this paper introduces how the first device adjusts the transceiver neural network model.

[0154] Optionally, adjusting the selected sending end neural network model includes:

[0155] receiving a loss function value from the second device;

[0156] Adjusting a set of transmitting-end neural network models selected by the first device according to the loss function value;

[0157] The loss function value is calculated by the second device based on the loss function used to adjust the receiving end neural network model and the received reference signal used to adjust the receiving end neural network model; the loss function is provided by the first device; and the reference signal is sent to the second device by the first device using a group of transmitting end neural network models selected by it.

[0158] It should be noted that, in the embodiment of the present invention, there is no limit on the number of times the first device adjusts the transmitting end neural network model. The above embodiment introduces that the first device adjusts the transmitting end neural network model according to the reference signal sent or received in real time. If the adjusted transmitting end neural network model cannot meet the data transmission requirements, the step of adjusting the transmitting end neural network model can be performed again.

[0159] In some embodiments, the method further includes: determining whether a preset condition is met; if the preset condition is met, transmitting data through the adjusted transmitting or receiving end neural network model; otherwise, continuing to adjust the data neural network model.

[0160] Optionally, the preset condition includes one or more of the following:

[0161] (1) The number of adjustments of the neural network model at the transmitting end or the receiving end reaches a preset value;

[0162] (2) receiving an instruction to stop adjustment;

[0163] (3) The adjusted transmitting or receiving neural network model meets the transmission requirements of the first device.

[0164] In an embodiment of the present invention, a transmitting end or a receiving end in a communication system can select a suitable transceiver neural network model for data transmission according to the current channel environment, realize the functions of the transceiver through the neural network model, improve the flexibility of the system structure, and reduce the deployment cost of the system.

[0165] In the embodiment of the present invention, a suitable neural network model is selected for data transmission according to the current channel environment. The specific process may include: offline training of the neural network model set at the transmitting end or the receiving end and selection and optimization of the neural network model at the transmitting end or the receiving end, which will be introduced in turn below:

[0166] (i) Offline training of a set of neural network models at the sender or receiver;

[0167] In this step, the sender and the receiver can establish a sender or receiver neural network model set for use in the data transmission stage.

[0168] The set of neural network models at the transmitting or receiving end is obtained based on simulated channel training under different communication scenarios. For example, possible wireless channels are classified in advance according to parameters such as indoor or outdoor, line-of-sight or non-line-of-sight, high frequency band or low frequency band, and then a series of neural network models are obtained through simulated channel training under different communication scenarios, and the training results are stored in a storage device.

[0169] The above process of training the neural network model in different communication scenarios can be performed using existing methods, including but not limited to stochastic gradient descent (SGD) and the like.

[0170] During the online data transmission stage, the sender and receiver can read the neural network model set obtained through offline training from the corresponding storage devices.

[0171] When training a model set offline, different communication parameters (such as signal-to-noise ratio, model input dimension, model output dimension, number of supported data streams, number of supported channels, etc.) need to be adjusted to achieve the purpose of training multiple transceiver neural network models.

[0172] When adjusting the above communication parameters, different sets of parameters need to be adapted to different transmission environments and transmission requirements. Figure 2b When the system uses neural networks to implement coding-modulation functions, only two parameters, signal-to-noise ratio and supported transmission rate, can be used to distinguish each element. When the system uses neural networks to implement coding-modulation-multiple antenna functions, four parameters, signal-to-noise ratio, model input dimension, model output dimension, number of supported data streams, and number of supported channels, can be used to distinguish each element. For a transceiver implemented using a neural network, the supported transmission rate is actually calculated by the ratio of the model's output dimension to its input dimension.

[0173] There is a certain correspondence between the sender neural network model and the receiver neural network model trained by the above method, that is, one sender neural network model corresponds to one receiver neural network model. When a receiver neural network model is selected at the receiver (transmitter), it means that the sender needs to select another sender neural network model in the binary.

[0174] It is understandable that the number of elements in the neural network model set at the sender and receiver can be flexibly adjusted, and the default value can be set at the same time. Finally, the elements in the set are not fixed after they are determined, but can be flexibly adjusted according to the historical transmission results, but it should be noted that the corresponding relationship between the elements of the sender and receiver sets must remain unchanged after the adjustment.

[0175] In addition, the results of offline training need to be unified in base stations and terminal devices of different manufacturers. To achieve this, a third party (such as a standard organization such as 3GPP) can be used to unify the offline mode and stipulate the offline training results. It should be noted that since the transceiver structure is almost unchanged, the above process can be performed offline, and the offline training results can be used for a considerable period of time without rapid adjustment.

[0176] In some implementations, the results of the above offline training process are not directly used in the online data transmission process. From the perspective of ensuring transmission performance, the neural network models at the transmitting end and the receiving end can be trained based on data generated by the actual channel.

[0177] However, the existing training using actual channel data faces two major difficulties. First, the training process of a neural network often requires a large amount of data and has huge computational overhead. Putting it directly online often occupies a large amount of wireless channel resources and has huge delays. In addition, online training needs to solve the problem of gradient transmission in the channel. Specifically, the existing neural network training algorithms all use the gradient transmission method, but the wireless channels in the real environment are very complex, and it is impossible to guarantee whether the gradient is differentiable or even exists. Therefore, the gradient on the receiving side cannot be directly back-propagated through the channel to the sending side.

[0178] In order to address the above-mentioned problems, in an embodiment of the present invention, a two-stage training method can be adopted, that is, firstly offline using a simulated channel model to roughly train a set of neural network models of the transmitter or receiver, and then using actual channel data to fine-tune and optimize the neural network model of the transmitter or receiver. On the one hand, it is to reduce the complexity of online training and the amount of data required, thereby controlling the overhead of online training within an acceptable range; on the other hand, the analytical form of the simulated channel model enables gradient transfer. In addition, in order to solve the problem of gradient transfer during online fine-tuning, in an embodiment of the present invention, relevant signaling is designed to transfer necessary parameters between the transmitter and the receiver. For details, please refer to the process. Figure 5a and Figure 5b Related to steps 5 to 9.

[0179] (II) Selection and optimization of neural network models at the transmitter or receiver

[0180] After establishing the roughly optimized neural network model set of the transmitter or receiver, the system will determine the neural network model used by the transmitter or receiver through signaling interaction, and further optimize the neural network model of the transmitter or receiver using real-time data. In the implementation process, only the use of neural networks to realize the transmission of data channels can be considered, while the control channel and other parts are still implemented using traditional methods. Figure 5a and 5b There are two ways to send reference signals, namely sending through the traditional physical layer process and sending reference signals through the neural network. The specific steps are as follows (since the uplink and downlink signaling processes are different, the uplink and downlink situations are described separately):

[0181] See also Figure 5a , the sending end is a base station, the receiving end is a terminal, and includes steps 1 to 10.

[0182] Step 1: The transmitter sends a system broadcast message or a reference signal to the receiver.

[0183] The reference signal is a signal provided by the transmitting end to the receiving end for channel estimation or channel detection. Exemplarily, the base station enables the terminal to measure the real-time channel by sending a system broadcast message or a reference signal.

[0184] It can be understood that the format and transmission of the system broadcast message or reference signal can be performed with reference to the existing methods.

[0185] Step 2: The receiving end selects the receiving end neural network model.

[0186] For example, the terminal selects one or more groups of receiving-end neural network models from a set of receiving-end neural network models obtained through offline training based on a received system broadcast message or reference signal.

[0187] It is understandable that the method for the receiving end to select the receiving end neural network model in step 2 includes but is not limited to:

[0188] (1) traversal search, for example, the receiving end searches for part or all of the receiving end neural network models in the set of receiving end neural network models obtained through offline training in a predetermined order, and selects one or more sets of receiving end neural network models;

[0189] (2) Parameter matching, for example, the receiving end matches one or more sets of receiving end neural network models from the set of receiving end neural network models obtained through offline training according to predetermined parameters.

[0190] It is understandable that the method for determining the number of receiving-end neural network models that need to be selected in step 2 includes but is not limited to: (1) autonomous determination by the receiving end, for example, the receiving end determines the number based on a local algorithm; (2) indication by the sending end, for example, the sending end indicates the number through signaling.

[0191] Step 3: The receiving end feeds back to the sending end one or more groups of receiving end neural network models selected by the receiving end;

[0192] For example, the terminal sends uplink control information (UCI) to the base station, where the UCI indicates one or more groups of receiving end neural network models selected by the terminal.

[0193] It is understandable that the content fed back by the receiving end to the sending end includes but is not limited to: identification of one or more groups of receiving end neural network models, for example, indexes of one or more groups of receiving end neural network models in a collection.

[0194] Step 4: The sender selects a sender neural network model;

[0195] For example, the base station selects a group of transmitting-end neural network models from a set of transmitting-end neural network models obtained through offline training based on one or more groups of receiving-end neural network models fed back by the terminal.

[0196] It is understandable that the method for the sender to select the sender neural network model in step 4 includes but is not limited to:

[0197] (1) traversal search, for example, the sender searches for part or all of the sender neural network models in the sender neural network model set obtained through offline training in a predetermined order, and selects a set of sender neural network models from them;

[0198] (2) Parameter matching, for example, the sender matches a corresponding set of sender neural network models in the sender neural network model set obtained through offline training according to predetermined parameters.

[0199] Step 5: The sending end sends indication information to the receiving end, where the indication information indicates one or more of the following:

[0200] (a) A set of sender neural network models selected by the sender;

[0201] For example, when the terminal feeds back multiple groups of receiving end neural network models, the base station informs the terminal of a group of transmitting end neural network models selected by it, and the content indicated by the indication information includes but is not limited to the index of the transmitting end neural network model in the set.

[0202] (b) reference signal;

[0203] That is, the reference signal used by the receiver to adjust the receiver neural network model.

[0204] Adjusting the receiving end neural network model can also be called fine-tuning the receiving end neural network model, or optimizing the receiving end neural network model, that is, the receiving end adjusts (fine-tunes or optimizes) the selected receiving end neural network model according to the received reference signal.

[0205] (c) Type of loss function;

[0206] That is, the type of loss function used by the receiver when adjusting the neural network model at the receiver.

[0207] (d) The sending end configures whether the receiving end adjusts the receiving end neural network model.

[0208] It can be understood that the indication information can be downlink control information (Downlink Control Information, DCI), (a) is an optional option and is only included when the receiving end feeds back multiple groups of receiving end neural network models.

[0209] The above loss function refers to the objective function required for the adjustment (fine-tuning or optimization) process of the neural network model at the transmitter or receiver. Optionally, the types of loss functions include but are not limited to one or more of the following:

[0210] i) Receiver bit-level log-likelihood ratio (LLR);

[0211] ii) Symbol-level log-likelihood ratio at the receiving end.

[0212] Optionally, the loss function may be indicated in one or more of the following ways:

[0213] i) Issue a specific loss function;

[0214] ii) Send the index of the loss function in the set.

[0215] It is understandable that the above-mentioned loss function does not refer to the objective function itself used for adjustment, as long as the sending end can calculate the objective function used for adjustment based on the feedback "loss function".

[0216] Step 6: The transmitter sends a reference signal to the receiver through the selected transmitter neural network model.

[0217] For example, the base station uses the selected transmitting end neural network model to send the reference signal determined in step 5 to the terminal.

[0218] Step 7: The receiver adjusts the selected receiver neural network model.

[0219] For example, the terminal uses the loss function sent by the base station in step 5 as the optimization target and uses the feedback signal received in step 6 as data to adjust the terminal receiving end neural network model.

[0220] It can be understood that if the receiving end is configured not to adjust the receiving end neural network model in step 5, then in this step the receiving end only calculates the loss function required for feedback without adjusting its own receiving end neural network model.

[0221] Step 8: The receiving end feeds back the loss function value to the sending end;

[0222] For example, the terminal feeds back a loss function value to the base station through the UCI, where the loss function value is calculated by the receiving end according to the reference signal received in step 6 and the loss function.

[0223] Step 9: The sender adjusts the sender neural network model.

[0224] For example, the base station adjusts the neural network model of the transmitting end according to the loss function value fed back by the terminal.

[0225] It is understandable that the neural network model adjustment process of the sending and receiving ends (steps 6 to 9) can be performed multiple times until the receiving end confirms that it is no longer necessary to adjust the receiving end neural network model (or the maximum number of adjustments is reached). The receiving end will then feed back a message to the sending end, indicating whether formal data transmission can or cannot be carried out.

[0226] Step 10: The sender uses the adjusted sender neural network model for data transmission.

[0227] For example, if the terminal feedback indicates that data transmission is possible, the base station uses the adjusted transmitting or receiving neural network model to send and receive data.

[0228] If the terminal feedback indicates that formal data transmission is not possible, it means that the channel may not meet the transmission requirements, and the process will return to step 1, or the neural network model selection process of the transmitter and receiver will be started again from step 1 after waiting for a period of time.

[0229] See also Figure 5b , the sending end is a terminal, the receiving end is a base station, and includes steps 1 to 10.

[0230] Step 1: The transmitter sends a reference signal to the receiver;

[0231] The reference signal is a signal provided by the transmitting end to the receiving end for channel estimation or channel detection. For example, the terminal sends the reference signal to enable the base station to measure the channel in real time. It can be understood that the format and transmission of the reference signal can refer to the existing method.

[0232] Step 2: The receiver selects one or more sets of receiver neural network models from the receiver model set obtained through offline training based on the reference signal received in step 1.

[0233] For example, the base station selects one or more groups of receiving end neural network models from a set of receiving end models obtained through offline training based on the received reference signal.

[0234] It is understandable that the method for the receiving end to select the receiving end neural network model in step 2 includes but is not limited to:

[0235] (1) traversal search, for example, the receiving end searches for part or all of the receiving end neural network models in the set of receiving end neural network models obtained through offline training in a predetermined order, and selects one or more sets of receiving end neural network models;

[0236] (2) Parameter matching, for example, the receiving end matches one or more sets of receiving end neural network models from the set of receiving end neural network models obtained through offline training according to predetermined parameters.

[0237] It is understandable that the method for determining the number of receiving-end neural network models that need to be selected in step 2 includes but is not limited to: (1) autonomous determination by the receiving end, for example, the receiving end determines the number based on a local algorithm; (2) indication by the sending end.

[0238] Step 3: The sending end sends indication information to the receiving end, where the indication information indicates one or more of the following:

[0239] (a) one or more groups of sender neural network models selected by the sender;

[0240] The content indicated by the indication information includes but is not limited to the index of the selected one or more groups of sending-end neural network models in the collection.

[0241] (b) The sender configures whether the receiver adjusts the neural network model at the receiver;

[0242] For example, the base station configures the terminal whether to adjust (fine-tune or optimize) the transceiver neural network model.

[0243] (c) Type of loss function;

[0244] If (b) the receiving end is configured to adjust the receiving end neural network model, the indication information also indicates the loss function type.

[0245] Optionally, the base station sends the indication information via DCI.

[0246] The loss function refers to the objective function required for the adjustment (fine-tuning or optimization) process of the receiving end neural network model. Optionally, the types of loss functions include but are not limited to one or more of the following:

[0247] i) receiving end bit-level log-likelihood ratio;

[0248] ii) Symbol-level log-likelihood ratio at the receiving end.

[0249] Optionally, the loss function may be indicated in one or more of the following ways:

[0250] i) Issue a specific loss function;

[0251] ii) Send the index of the loss function in the set.

[0252] It is understandable that the above-mentioned loss function does not refer to the objective function itself used for adjustment, as long as the sending end can calculate the objective function used for adjustment based on the feedback "loss function".

[0253] Step 4: The sender selects a set of sender neural network models.

[0254] Exemplarily, the terminal selects a set of transmitting-end neural network models from a set of transmitting-end neural network models obtained through offline training based on one or more sets of receiving-end neural network models fed back by the base station. The model selection methods that can be used here include but are not limited to: (1) traversal search; (2) parameter matching.

[0255] Step 5: The sending end sends indication information to the receiving end, where the indication information indicates one or more of the following:

[0256] (1) a set of sending-end neural network models;

[0257] For example, when the base station sends multiple groups of receiving-end neural network models, the terminal informs the base station of the group of transmitting-end neural network models it has selected, and the content indicated by the indication information includes but is not limited to the index of the group of transmitting-end neural network models in the collection.

[0258] The above item (1) is optional and is only included when the base station sends multiple sets of receiving end neural network models.

[0259] (2) The type of the first signal, such as the type of the reference signal.

[0260] The receiving end can adjust its selected receiving end neural network model according to the first signal.

[0261] For example, the terminal sends indication information to the base station through UCI.

[0262] Step 6: The transmitter uses the selected transmitter neural network model to send the reference signal indicated in step 5.

[0263] For example, the terminal uses the selected transmitting neural network model to send the reference signal specified in step 5 to the base station.

[0264] Step 7: The receiving end adjusts the receiving end neural network model;

[0265] For example, the base station uses the loss function sent in step 5 as the optimization target and uses the reference signal in step 6 as data to adjust the neural network model of the base station receiver.

[0266] Step 8: The receiving end sends the loss function value to the sending end;

[0267] For example, the base station feeds back the loss function value to the terminal via DCI, and the loss function value is calculated by the base station according to the reference signal received in step 6. If the base station configures the terminal not to adjust the transmitting end neural network model in step 3, no feedback is performed in this step or a null value is fed back.

[0268] Step 9: The sender adjusts the sender neural network model.

[0269] It is understandable that step 9 is an optional step. For example, when the base station configures the terminal to adjust the neural network model of the transmitting end, the terminal adjusts the neural network model of the transmitting end according to the loss function value fed back by the base station.

[0270] It can be understood that the transceiver neural network model adjustment process (steps 6 to 9) of the sending and receiving ends can be performed multiple times until the receiving end confirms that the neural network model adjustment is no longer required (or the maximum number of adjustments is reached). The receiving end will then feed back a message to the sending end, indicating whether formal data transmission can or cannot be carried out.

[0271] Step 10: The sender uses the adjusted sender neural network model for data transmission.

[0272] For example, if the base station feedback indicates that data transmission is possible, the terminal uses the adjusted sender neural network model to send and receive data. If the base station feedback indicates that formal output transmission is not possible, it means that the channel may not meet the transmission requirements, and the terminal will return to step 1, or wait for a period of time and start again from step 1 to execute the sender and receiver neural network model selection process.

[0273] It should be noted that Figure 5a and Figure 5b The difference is that the base station can configure whether to fine-tune the model on the terminal side according to the terminal conditions (such as computing power, power consumption, etc.). In downlink, the base station, as the transmitting end, notifies the terminal side in step 5 whether to fine-tune the neural network model. If the terminal side is set not to fine-tune the neural network model, it only calculates the loss function required for feedback in step 7, and does not adjust its own neural network model. In uplink, the base station, as the receiving end, notifies the terminal side in step 3 whether to fine-tune the neural network model. If the terminal side is set not to fine-tune the neural network model, the base station does not feedback the loss function or feeds back a null value to the terminal side in step 8. The purpose of setting this mechanism is to flexibly adjust the computational burden on the terminal side.

[0274] The embodiment of the present invention is described below in conjunction with the first embodiment.

[0275] Embodiment 1:

[0276] Taking the LTE system as an example, Figure 6 The block diagram of the transmitter implemented by the traditional method is shown in Figure 1. The coding, scrambling, modulation mapper, layer mapper, precoding (MIMO processing) and other functions in the traditional link of the transmitter are implemented by the neural network model, see Figure 7 , and the subsequent resource mapping and OFDM waveform generation modules are implemented in the traditional way. The receiving end is implemented in the form of duality with the transmitting end, which will not be elaborated here. In addition, the reference signal and other contents mentioned below can also be implemented in the traditional way.

[0277] Step 1: Offline training process of the neural network model at the sending or receiving end.

[0278] During the offline training process, the wireless channels are first classified according to parameters such as indoor or outdoor, line-of-sight or non-line-of-sight, high frequency band or low frequency band, and then the simulation channel model is used to generate data for the above different channel types, and the parameters such as signal-to-noise ratio, transmission rate, number of supported data streams, and number of supported channels are adjusted. Offline training obtains multiple transmitter or receiver neural network models, and the offline training results are stored according to the simulation channel model type. After the communication system is turned on, the system can obtain the above offline training results by reading the data in the access device. Figure 8 A schematic diagram of the model set results obtained through offline training is given in FIG.

[0279] Step 2: Online selection and optimization of the neural network model at the sending or receiving end.

[0280] After obtaining the offline trained sender or receiver neural network model set, the system will select the appropriate neural network model at the sender and receiver, optimize it using real-time data, and finally use it for actual data transmission. Fig. 9 The signaling process of step 2 above is given, which mainly includes:

[0281] Step 1: The base station (transmitter) sends a cell reference signal (CRS) to the terminal.

[0282] Step 2: The terminal (receiver) selects one or more sets of receiving end neural network models based on the received CRS.

[0283] Step 3: The terminal reports one or more groups of receiving end neural network models to the base station through UCI signaling, and the feedback content includes but is not limited to the index of one or more groups of receiving end neural network models in the set.

[0284] Step 4: When the terminal reports a set of receiving-end neural network models, the base station selects a transmitting-end neural network model corresponding to this model; and when the terminal reports multiple sets of receiving-end neural network models, the base station selects a set of transmitting-end neural network models from the transceiver neural network model set based on the multiple sets of receiving-end neural network models.

[0285] Step 5: The base station sends the following information to the terminal via DCI:

[0286] i) If the terminal reports multiple sets of receiving end neural network models in step 3, the base station informs the terminal of a set of sending end neural network models selected by it, and the feedback content includes but is not limited to: the index of the sending end neural network model in the set;

[0287] ii) the reference signal used when adjusting the neural network model at the receiving end;

[0288] iii) The loss function used when adjusting the neural network model at the receiving end;

[0289] iv) Indication information on whether the terminal adjusts the receiving end neural network model.

[0290] Optionally, the types of loss functions include but are not limited to: a) receiving end bit-level log likelihood ratio (LLR); b) receiving end symbol-level log likelihood ratio, etc. The ways of indicating the loss function include but are not limited to: a) sending a specific loss function; b) sending the index of the loss function in the set, etc.

[0291] Step 6: The base station uses the transmitting neural network model selected in step 4 to send the reference signal determined in step 5 to the terminal, and the terminal uses the receiving neural network model selected in step 3 to receive it.

[0292] Step 7: The terminal adjusts the receiving end neural network model according to the loss function sent in step 5 and the reference signal received in step 6, and calculates the loss function value that needs to be fed back to the base station. If the base station configures the terminal not to adjust the receiving end neural network model in step 5, the terminal only calculates the loss function value required for feedback in this step without adjusting its own receiving end neural network model.

[0293] Step 8: The terminal feeds back the loss function value to the base station through UCI.

[0294] Step 9: The base station optimizes and fine-tunes the transmitter neural network model based on the feedback loss function value.

[0295] It is understandable that the base station and the terminal repeat the process from step 6 to step 8 multiple times until the terminal confirms that no fine-tuning is needed (or the maximum number of repetitions is reached), and the terminal feeds back a confirmation message to the sender, indicating whether formal data transmission can be performed.

[0296] Step 10: If the terminal feedback indicates that data transmission is possible, the terminal or base station uses the optimized neural network model to send and receive data. If the terminal feedback indicates that formal output transmission is not possible, return to step 1, or wait for a period of time and then start the transceiver model selection process again from step 1.

[0297] Step 11: If the base station and the terminal find that the channel has changed and the previously used transmitting or receiving neural network model is no longer suitable for the current data transmission (or after the preset amount of data has been transmitted), the above selection process can be reactivated until new data transmission can be performed.

[0298] exist Figure 7In the process shown, since only the neural network models at both ends of the transmitter and receiver need to be determined when transmitting data, the transmission of CRS, UCI, and DCI can be implemented in the traditional LTE system mode, without the problem of "nested" implementation that requires the selection of a neural network model to send a reference signal. In addition, Figure 8 The time granularity of reporting the neural network model through UCI signaling feedback can be the same as MCS and PMI.

[0299] See also Fig.10 The embodiment of the present invention provides a communication device, which is a first device. The first device 1000 includes:

[0300] A determination module 1001 is used to determine a sender or receiver neural network model of the first device according to the interaction between the first device and the second device, where the sender or receiver neural network model is any group in a sender or receiver neural network model set of the first device;

[0301] The transmission module 1002 is used to transmit data with the second device through the transmitting end or receiving end neural network model.

[0302] In some implementations, the first device 1000 further includes:

[0303] The training module is used to obtain the channel type and the communication parameters corresponding to the channel type; perform simulation channel training according to the channel type and the communication parameters to obtain a set of transmitting or receiving end neural network models corresponding to the channel type.

[0304] In some implementations, the channel type includes one of: an indoor channel, an outdoor channel, a line-of-sight channel, a non-line-of-sight channel, a low-band channel, and a high-band channel.

[0305] In some implementations, the communication parameters include one or more of the following: signal-to-noise ratio, supported rate, number of supported data streams, and number of supported channels.

[0306] In some embodiments, the determination module 1001 is further used to: receive a first signal from the second device; select a group of receiving-end neural network models from the set of receiving-end neural network models of the first device based on the first signal; or select multiple groups of receiving-end neural network models from the set of receiving-end neural network models of the first device based on the first signal; obtain a group of transmitting-end neural network models selected by the second device from the set of transmitting-end neural network models of the second device; and select a group of receiving-end neural network models from the multiple groups of receiving-end neural network models based on the group of transmitting-end neural network models selected by the second device.

[0307] In some implementations, the first device 1000 further includes: a first sending module, configured to send first information to the second device;

[0308] The first information includes one or more of the following:

[0309] (1) information of one or more groups of receiving-end neural network models selected by the first device based on the first signal;

[0310] (2) the first device configures the second device to adjust the information of the neural network model of the sending end;

[0311] (3) Information used to adjust the loss function of the receiving end neural network model.

[0312] In some implementations, the first device 1000 further includes: a first receiving module, configured to receive second information from the second device;

[0313] The second information includes one or more of the following:

[0314] (1) information of a group of transmitting-end neural network models selected by the second device;

[0315] (2) the second device configures the first device with information on whether to adjust the neural network model of the receiving end;

[0316] (3) information of a reference signal used to adjust a neural network model at the receiving end;

[0317] (4) Information used to adjust the loss function of the receiving end neural network model.

[0318] In some embodiments, the determination module 1001 is further used to: obtain one or more groups of receiving-end neural network models selected by the second device from the set of receiving-end neural network models of the second device; and select a group of transmitting-end neural network models from the set of transmitting-end neural network models of the first device based on the one or more groups of receiving-end neural network models selected by the second device.

[0319] In some implementations, the first device 1000 further includes:

[0320] The second sending module is configured to send third information to the second device, where the third information includes one or more of the following:

[0321] (1) information of a group of transmitting-end neural network models selected by the first device;

[0322] (2) information of a reference signal used to adjust a neural network model at the receiving end;

[0323] (3) information used to adjust the loss function of the receiving neural network model;

[0324] (4) The first device configures the second device with information on whether to adjust the neural network model at the receiving end.

[0325] In some embodiments, the transmission module 1002 is further used to: adjust the selected transmitting end or receiving end neural network model; and transmit data with the second device through the adjusted transmitting end or receiving end neural network model.

[0326] In some embodiments, the transmission module 1002 is further used to: receive a reference signal for adjusting the receiving-end neural network model from the second device, wherein the reference signal is sent by the second device using a group of transmitting-end neural network models selected by it; adjust a group of receiving-end neural network models selected by the first device according to a loss function for adjusting the receiving-end neural network model and the reference signal; wherein the loss function is provided by the second device.

[0327] In some implementations, the first device 1000 further includes:

[0328] The third sending module is used to send a loss function value to the second device if the first device configures the second device to adjust the transmitting end neural network model, and the loss function value is calculated by the first device based on the loss function and the reference signal; if the first device configures the second device not to adjust the transceiver neural network model, the loss function value is not fed back to the second device.

[0329] In some embodiments, the transmission module 1002 is further used to: receive a loss function value from the second device; adjust a group of transmitting-end neural network models selected by the first device according to the loss function value; wherein the loss function value is calculated by the second device based on the loss function used to adjust the receiving-end neural network model and the received reference signal used to adjust the receiving-end neural network model; the loss function is provided by the first device; and the reference signal is sent to the second device by the first device using a group of transmitting-end neural network models selected by it.

[0330] In some embodiments, the transmission module 1002 is further used to: when a preset condition is met, transmit data to the second device through the adjusted transmitting end or receiving end neural network model.

[0331] In some implementations, the preset conditions include one or more of the following:

[0332] (1) The number of adjustments of the neural network model at the transmitting end or the receiving end reaches a preset value;

[0333] (2) receiving an instruction to stop adjusting the neural network model of the transmitting end or the receiving end;

[0334] (3) The adjusted neural network model of the transmitting end or the receiving end meets the transmission requirements of the first device or the second device.

[0335] The first device provided by the embodiment of the present invention can execute the above Figure 4 The implementation principle and technical effect of the method embodiment shown are similar, and this embodiment will not be repeated here.

[0336] See also Fig.11 An embodiment of the present invention provides a communication device, which is a first device. The first device 1100 includes: a processor 1101.

[0337] The processor 1101 reads the program in the memory and performs the following operations: determining a transmitting or receiving neural network model of the first device according to the interaction between the first device and the second device, wherein the transmitting or receiving neural network model is any group in a set of transmitting or receiving neural network models of the first device; and transmitting data with the second device through the transmitting or receiving neural network model.

[0338] In some embodiments, the processor 1101 reads the program in the memory to perform the following operations: obtain a channel type and communication parameters corresponding to the channel type; perform simulation channel training based on the channel type and communication parameters to obtain a set of transmitting or receiving end neural network models corresponding to the channel type.

[0339] In some implementations, the channel type includes one of: an indoor channel, an outdoor channel, a line-of-sight channel, a non-line-of-sight channel, a low-band channel, and a high-band channel.

[0340] In some implementations, the communication parameters include one or more of the following: signal-to-noise ratio, supported rate, number of supported data streams, and number of supported channels.

[0341] Optionally, the processor 1101 reads the program in the memory to perform the following operations: receive a first signal from the second device; select a group of receiving-end neural network models from the set of receiving-end neural network models of the first device based on the first signal; or select multiple groups of receiving-end neural network models from the set of receiving-end neural network models of the first device based on the first signal; obtain a group of transmitting-end neural network models selected by the second device from the set of transmitting-end neural network models of the second device; and select a group of receiving-end neural network models from the multiple groups of receiving-end neural network models based on the group of transmitting-end neural network models selected by the second device.

[0342] Optionally, the processor 1101 reads a program in the memory to perform the following operations: sending first information to the second device;

[0343] The first information includes one or more of the following:

[0344] (1) information of one or more groups of receiving-end neural network models selected by the first device based on the first signal;

[0345] (2) the first device configures the second device to adjust the information of the neural network model of the sending end;

[0346] (3) Information used to adjust the loss function of the receiving end neural network model.

[0347] Optionally, the processor 1101 reads a program in the memory to perform the following operations: receiving second information from the second device;

[0348] The second information includes one or more of the following:

[0349] (1) information of a group of transmitting-end neural network models selected by the second device;

[0350] (2) the second device configures the first device with information on whether to adjust the neural network model of the receiving end;

[0351] (3) information of a reference signal used to adjust a neural network model at the receiving end;

[0352] (4) Information used to adjust the loss function of the receiving end neural network model.

[0353] Optionally, the processor 1101 reads the program in the memory to perform the following operations: obtain one or more groups of receiving-end neural network models selected by the second device from the set of receiving-end neural network models of the second device; and select a group of transmitting-end neural network models from the set of transmitting-end neural network models of the first device based on the one or more groups of receiving-end neural network models selected by the second device.

[0354] Optionally, the processor 1101 reads a program in the memory to perform the following operations: sending third information to the second device, where the third information includes one or more of the following:

[0355] (1) information of a group of transmitting-end neural network models selected by the first device;

[0356] (2) information of a reference signal used to adjust a neural network model at the receiving end;

[0357] (3) information used to adjust the loss function of the receiving neural network model;

[0358] (4) The first device configures the second device with information on whether to adjust the neural network model at the receiving end.

[0359] Optionally, the processor 1101 reads a program in the memory to perform the following operations: adjusting the selected neural network model of the transmitting end or the receiving end; and transmitting data with the second device through the adjusted neural network model of the transmitting end or the receiving end.

[0360] Optionally, the processor 1101 reads the program in the memory and performs the following operations: receiving a reference signal for adjusting the receiving-end neural network model from the second device, the reference signal being sent by the second device using a group of transmitting-end neural network models selected by it; adjusting a group of receiving-end neural network models selected by the first device according to a loss function for adjusting the receiving-end neural network model and the reference signal; wherein the loss function is provided by the second device.

[0361] Optionally, the processor 1101 reads the program in the memory and performs the following operations: if the first device configures the second device to adjust the transmitting end neural network model, a loss function value is sent to the second device, and the loss function value is calculated by the first device based on the loss function and the reference signal; if the first device configures the second device not to adjust the transceiver neural network model, the loss function value is not fed back to the second device.

[0362] Optionally, the processor 1101 reads the program in the memory to perform the following operations: receive a loss function value from the second device; adjust a group of transmitting-end neural network models selected by the first device according to the loss function value; wherein the loss function value is calculated by the second device based on the loss function used to adjust the receiving-end neural network model and the received reference signal used to adjust the receiving-end neural network model; the loss function is provided by the first device; and the reference signal is sent by the first device to the second device using a group of transmitting-end neural network models selected by it.

[0363] Optionally, the processor 1101 reads a program in the memory and performs the following operations: when a preset condition is met, data is transmitted with the second device through the adjusted transmitting or receiving neural network model.

[0364] In some implementations, the preset conditions include one or more of the following:

[0365] (1) The number of adjustments of the neural network model at the transmitting end or the receiving end reaches a preset value;

[0366] (2) receiving an instruction to stop adjusting the neural network model of the transmitting end or the receiving end;

[0367] (3) The adjusted neural network model of the transmitting end or the receiving end meets the transmission requirements of the first device or the second device.

[0368] The first device provided by the embodiment of the present invention can execute the above Figure 4 The implementation principle and technical effect of the method embodiment shown are similar, and this embodiment will not be repeated here.

[0369] See also Fig.12 , Fig.12 is a structural diagram of a communication device to which an embodiment of the present invention is applied, such as Fig.12 As shown, the communication device 1200 includes: a processor 1201, a transceiver 1202, a memory 1203 and a bus interface, wherein:

[0370] In one embodiment of the present invention, the communication device 1200 further includes: a computer program stored in the memory 1203 and executable on the processor 1201, the computer program being executed by the processor 1201 to implement Figure 4 Steps in the illustrated embodiment.

[0371] exist Fig.12In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking together various circuits of one or more processors represented by processor 1201 and memory represented by memory 1203. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The transceiver 1202 may be a plurality of components, namely, a transmitter and a receiver, providing a unit for communicating with various other devices over a transmission medium.

[0372] The processor 1201 is responsible for managing the bus architecture and general processing, and the memory 1203 can store data used by the processor 1201 when performing operations.

[0373] The communication device provided in the embodiment of the present invention can execute the above Figure 4 The implementation principle and technical effect of the method embodiment shown are similar, and this embodiment will not be repeated here.

[0374] The steps of the method or algorithm described in conjunction with the disclosure of the present invention may be implemented in hardware or by a processor executing software instructions. The software instructions may be composed of corresponding software modules, and the software modules may be stored in RAM, flash memory, ROM, EPROM, EEPROM, registers, hard disks, mobile hard disks, read-only optical disks, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium may also be a component of the processor. The processor and the storage medium may be located in an ASIC. In addition, the ASIC may be located in a core network interface device. Of course, the processor and the storage medium may also exist in a core network interface device as discrete components.

[0375] Those skilled in the art will appreciate that in one or more of the above examples, the functions described in the present invention may be implemented in hardware, software, firmware, or any combination thereof. When implemented using software, the functions may be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any media that facilitates the transmission of a computer program from one place to another. The storage medium may be any available medium that a general or special-purpose computer can access.

[0376] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solution of the present invention should be included in the scope of protection of the present invention.

[0377] Those skilled in the art will appreciate that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the embodiments of the present invention may take the form of complete hardware embodiments, complete software embodiments, or embodiments combining software and hardware. Moreover, the embodiments of the present invention may take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0378] The embodiments of the present invention are described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0379] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0380] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0381] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A transmission method, applied to a first device, characterized in that: include: Determine, according to the interaction between the first device and the second device, a transmitting end or a receiving end neural network model of the first device, wherein the transmitting end or the receiving end neural network model is any group in a transmitting end or a receiving end neural network model set of the first device; Performing data transmission with the second device through the transmitting end or receiving end neural network model; The performing data transmission with the second device through the transmitting end or receiving end neural network model includes: Adjusting the selected neural network model of the transmitting end or the receiving end; Performing data transmission with the second device through the adjusted transmitting end or receiving end neural network model; The first device is a receiving end, the second device is a sending end, and the adjusting the selected sending end neural network model includes: receiving a reference signal for adjusting a receiving-end neural network model from the second device, wherein the reference signal is sent by the second device using a set of transmitting-end neural network models selected by the second device; Adjusting a set of receiving-end neural network models selected by the first device according to a loss function for adjusting the receiving-end neural network model and the reference signal; Alternatively, the first device is a transmitting end, the second device is a receiving end, and the adjusting the selected transmitting end neural network model includes: receiving a loss function value from the second device; Adjusting a set of transmitting-end neural network models selected by the first device according to the loss function value; Among them, the loss function value is calculated by the second device based on the loss function used to adjust the receiving end neural network model and the received reference signal used to adjust the receiving end neural network model; the reference signal is sent to the second device by the first device using a group of transmitting end neural network models selected by it.

2. The method according to claim 1, characterized in that The method further comprises: Acquire a channel type and communication parameters corresponding to the channel type; According to the channel type and communication parameters, simulated channel training is performed to obtain a set of transmitting end or receiving end neural network models corresponding to the channel type.

3. The method according to claim 2, characterized in that: The channel type includes one of the following: an indoor channel, an outdoor channel, a line-of-sight channel, a non-line-of-sight channel, a low-frequency band channel, and a high-frequency band channel.

4. The method according to claim 2, characterized in that: The communication parameters include one or more of the following: signal-to-noise ratio, supported rate, number of supported data streams, and number of supported channels.

5. The method according to claim 1, characterized in that The first device is a receiving end, and the second device is a sending end; The determining, according to the interaction between the first device and the second device, a receiving end neural network model of the first device includes: receiving a first signal from the second device; Selecting a set of receiving-end neural network models from a set of receiving-end neural network models of the first device according to the first signal; or, Selecting, according to the first signal, a plurality of groups of receiving-end neural network models from a set of receiving-end neural network models of the first device; Obtaining a set of transmitting-end neural network models selected by the second device from a set of transmitting-end neural network models of the second device; According to a group of transmitting-end neural network models selected by the second device, a group of receiving-end neural network models is selected from the multiple groups of receiving-end neural network models.

6. The method according to claim 5, characterized in that The method further comprises: Sending first information to the second device; The first information includes one or more of the following: Information of one or more groups of receiving-end neural network models selected by the first device based on the first signal; Information about whether the second device adjusts the neural network model of the sending end; Information used to adjust the loss function of the receiving neural network model.

7. The method according to claim 5, characterized in that The method further comprises: receiving second information from the second device; The second information includes one or more of the following: Information about a set of transmitting-end neural network models selected by the second device; Information about whether the first device adjusts the receiving end neural network model; Information about a reference signal used to adjust a neural network model at the receiving end; Information used to adjust the loss function of the receiving neural network model.

8. The method according to claim 1, characterized in that The first device is a transmitting end, and the second device is a receiving end; The determining, according to the interaction between the first device and the second device, a transmitting end neural network model of the first device includes: Obtaining one or more groups of receiving-end neural network models selected by the second device from a set of receiving-end neural network models of the second device; According to one or more groups of receiving-end neural network models selected by the second device, a group of transmitting-end neural network models is selected from the set of transmitting-end neural network models of the first device.

9. The method according to claim 8, characterized in that The method further comprises: Sending third information to the second device, where the third information includes one or more of the following: Information about a set of transmitting-end neural network models selected by the first device; Information about a reference signal used to adjust a neural network model at the receiving end; Information used to adjust the loss function of the receiving neural network model; Information about whether the second device adjusts the receiving end neural network model.

10. The method according to claim 1, characterized in that The method further comprises: The first device sends a loss function value to the second device, which is used by the second device to adjust the transmitting end neural network model. The loss function value is calculated by the first device based on the loss function and the reference signal.

11. The method according to claim 1, characterized in that: The data transmission with the second device through the adjusted transmitting end or receiving end neural network model includes: When preset conditions are met, data is transmitted with the second device through the adjusted transmitting or receiving neural network model.

12. The method according to claim 11, characterized in that The preset conditions include one or more of the following: The number of adjustments of the neural network model at the transmitting end or the receiving end reaches a preset value; Receiving an instruction to stop adjusting the neural network model of the transmitting end or the receiving end; The adjusted transmitting or receiving end neural network model meets the transmission requirements of the first device.

13. A communication device, the communication device being a first device, characterized in that: The first device comprises: A determination module, configured to determine a sender or receiver neural network model of the first device according to the interaction between the first device and the second device, wherein the sender or receiver neural network model is any one of a sender or receiver neural network model set of the first device; A transmission module, used for transmitting data with the second device through the transmitting end or receiving end neural network model; The performing data transmission with the second device through the transmitting end or receiving end neural network model includes: Adjusting the selected neural network model of the transmitting end or the receiving end; Performing data transmission with the second device through the adjusted transmitting end or receiving end neural network model; The first device is a receiving end, the second device is a sending end, and the adjusting the selected sending end neural network model includes: receiving a reference signal for adjusting a receiving-end neural network model from the second device, wherein the reference signal is sent by the second device using a set of transmitting-end neural network models selected by the second device; Adjusting a set of receiving-end neural network models selected by the first device according to a loss function for adjusting the receiving-end neural network model and the reference signal; Alternatively, the first device is a transmitting end, the second device is a receiving end, and the adjusting the selected transmitting end neural network model includes: receiving a loss function value from the second device; Adjusting a set of transmitting-end neural network models selected by the first device according to the loss function value; Among them, the loss function value is calculated by the second device based on the loss function used to adjust the receiving end neural network model and the received reference signal used to adjust the receiving end neural network model; the reference signal is sent to the second device by the first device using a group of transmitting end neural network models selected by it.

14. A communication device, the communication device being a first device, characterized in that: The first device includes: a processor; The processor reads the program in the memory and performs the following operations: determining a transmitting end or receiving end neural network model of the first device according to the interaction between the first device and the second device, wherein the transmitting end or receiving end neural network model is any group in a transmitting end or receiving end neural network model set of the first device; and performing data transmission with the second device through the transmitting end or receiving end neural network model; The performing data transmission with the second device through the transmitting end or receiving end neural network model includes: Adjusting the selected neural network model of the transmitting end or the receiving end; Performing data transmission with the second device through the adjusted transmitting end or receiving end neural network model; The first device is a receiving end, the second device is a sending end, and the adjusting the selected sending end neural network model includes: receiving a reference signal for adjusting a receiving-end neural network model from the second device, wherein the reference signal is sent by the second device using a set of transmitting-end neural network models selected by the second device; Adjusting a set of receiving-end neural network models selected by the first device according to a loss function for adjusting the receiving-end neural network model and the reference signal; Alternatively, the first device is a transmitting end, the second device is a receiving end, and the adjusting the selected transmitting end neural network model includes: receiving a loss function value from the second device; Adjusting a set of transmitting-end neural network models selected by the first device according to the loss function value; Among them, the loss function value is calculated by the second device based on the loss function used to adjust the receiving end neural network model and the received reference signal used to adjust the receiving end neural network model; the reference signal is sent to the second device by the first device using a group of transmitting end neural network models selected by it.

15. A communication device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the transmission method according to any one of claims 1 to 12 are implemented.

16. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the transmission method according to any one of claims 1 to 12 are implemented.

Images