Communication method and communication apparatus

Abstract

The present application provides a communication method and a communication apparatus. The communication method comprises: a first device obtains at least one first model, wherein the first model may be a bilateral model or a unilateral model, the bilateral model is used for data encoding and data decoding, and the unilateral model is used for data encoding or data decoding; and the first device can determine a second model on the basis of the at least one first model obtained, wherein the second model is a unilateral model. By means of the method, the first device can determine the second model on the basis of the at least one first model obtained, and only one AI model for data encoding and / or one AI model for data decoding needs to be retained in a network, thereby reducing the overhead for storing AI models in a terminal device and / or a network device.

Description

Communication methods and communication devices

[0001] This application claims priority to Chinese Patent Application No. 202410956762.2, filed on July 17, 2024, entitled "Communication Method and Communication Apparatus", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of communication technology, and more specifically, to a communication method and a communication device. Background Technology

[0003] To optimize air interface transmission, a working mode of a bilateral channel state information (CSI) compression feedback-enhanced artificial intelligence (AI) model is proposed. In this mode, the terminal device collects CSI and compresses it using a pre-trained AI compression model before transmitting it to the network device. The network device then decompresses the CSI using a pre-trained AI decompression model and performs downlink precoding based on the decompressed CSI. Regarding the selection of the AI ​​compression and decompression models, two schemes exist: Scheme 1: Each network device manufacturer can train a bilateral model suitable for its network devices. This bilateral model includes an AI compression model for the terminal device and an AI decompression model for the network device. The trained AI compression model for the terminal device is registered in the network for terminal devices to download. Each terminal device downloads multiple AI compression models trained by various network device manufacturers and selects the AI ​​compression model trained by that network device manufacturer when establishing a connection with a particular network device. Option 2: Each terminal device manufacturer can train a two-sided model applicable to its manufactured terminal devices. This two-sided model includes an AI compression model on the terminal device side and an AI decompression model on the network device side. The trained network device-side AI decompression model is registered in the network for network devices to download. Each network device will download multiple AI decompression models trained by multiple terminal device manufacturers and select the AI ​​decompression model trained by that terminal device manufacturer when establishing a connection with a particular terminal device. Since there are many manufacturers of both network devices and terminal devices, the terminal devices in Option 1 need to download and store a large number of AI compression models, and the network devices in Option 2 need to download and store a large number of AI decompression models, resulting in high storage overhead. Summary of the Invention

[0004] This application provides a communication method and a communication device that can reduce the overhead of storing AI models in terminal devices and network devices.

[0005] Firstly, a communication method is provided. This method can be applied to a first device; that is, the method can be executed by the first device or by components of the first device (such as a chip, chip system, circuit, or communication module). This application does not limit the scope of the method. The following description mainly uses the first device as an example.

[0006] The method may include: a first device acquiring at least one first model, the first model being a bilateral model or a first unilateral model, the bilateral model being used for data encoding and data decoding, and the first unilateral model being used for data encoding or data decoding; the first device determining a second model based on at least one first model, the second model being used for data encoding or data decoding; and the first device sending first information, the first information being used to indicate the second model.

[0007] Using the above method, the first device can determine a second model based on at least one first model obtained. Only one AI model for data encoding and / or one AI model for data decoding can be retained in the network, thereby reducing the overhead of storing AI models in terminal devices and / or network devices.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, when the first model is a first unilateral model, the first device determines a second model based on at least one first model, including: the first device determines a second model based on at least one first model and at least one local model, wherein the second model belongs to at least one first model and the local model is a second unilateral model.

[0009] Wherein, if the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data decoding.

[0010] Optionally, the first device may also determine a second model and a second model A based on at least one first model and at least one local model, wherein the second model belongs to at least one first model and the second model A belongs to at least one local model.

[0011] Using the above method, the first device can select a second model from at least one first model for data encoding or data decoding, thereby reducing the overhead of storing AI models for data encoding or data decoding in terminal devices or network devices.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, when the first model is a first unilateral model, the above method further includes: the first device acquiring at least one third model, the third model being a second unilateral model; the first device determining a second model based on at least one first model including: the first device determining a second model and a first fourth model based on at least one first model and at least one third model, the second model belonging to at least one first model and the fourth model belonging to at least one third model; the sending of the first information including: sending first information and second information, the first information being used to indicate the second model and the second information being used to indicate the fourth model.

[0013] Wherein, if the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data decoding.

[0014] Optionally, the first device may also determine a second model and a fourth model based on at least one first model, at least one third model and at least one local model, wherein the at least one local model includes at least one first unilateral model and / or at least one second unilateral model, wherein the second model belongs to at least one first model or the second model belongs to at least one first unilateral model included in at least one local model, and the fourth model belongs to at least one third model or the fourth model belongs to at least one second unilateral model included in at least one local model.

[0015] Using the above method, the first device can select a second model and a fourth model from at least one first unilateral model and at least one second unilateral model respectively to cooperate in completing data encoding and data decoding, thereby reducing the overhead of terminal devices and network devices storing AI models for data encoding or data decoding.

[0016] In conjunction with the first aspect, in some implementations of the first aspect, the above-mentioned at least one third model is trained based on at least one first model, or at least one first model is trained based on the third model.

[0017] For example, if at least one first model is a first one-sided model for data encoding, then model B in at least one third model can be a second one-sided model for data decoding trained based on model A in at least one first model; or, if at least one first model is a first one-sided model for data decoding, then model B in at least one third model can be a second one-sided model for data encoding trained based on model A in at least one first model.

[0018] For example, if at least one third model is a second one-sided model for data encoding, then model A in at least one first model can be a first one-sided model for data decoding trained based on model B in at least one third model. Alternatively, if at least one third model is a second one-sided model for data decoding, then model A in at least one first model can be a first one-sided model for data encoding trained based on model B in at least one third model.

[0019] The above method enables the first device to achieve high performance in data encoding and decoding by cooperating at least one first model and at least one third model, thereby improving the accuracy or complexity of the second and fourth models determined by the first device.

[0020] In conjunction with the first aspect, in some implementations of the first aspect, when the first model is the first unilateral model, the second model is the second unilateral model.

[0021] Wherein, if the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data decoding.

[0022] Using the above method, the first device can determine a second unilateral model based on at least one first unilateral model, which can ensure that the terminal devices and network devices in the network retain only one AI model for data encoding or data decoding, thereby reducing the overhead of storing AI models for terminal devices and network devices.

[0023] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the first device sending third information, the third information being used to request an update of the first model; the first device obtaining at least one first model includes: the first device obtaining at least one first model based on the third information, the first model being an updated model.

[0024] For example, the first device may periodically send third information.

[0025] Using the above method, the first device can trigger real-time updates to the AI ​​model in the network, ensuring the accuracy of data encoding and decoding.

[0026] In conjunction with the first aspect, in certain implementations of the first aspect, the aforementioned first information includes at least one of the following:

[0027] The second model includes its model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the second model can be obtained.

[0028] The above methods can provide multiple ways to send models, increasing the flexibility of model sending.

[0029] Secondly, a communication method is provided. This method can be applied to a second device; that is, the method can be executed by the second device or by components of the second device (such as a chip, chip system, circuit, or communication module). This application does not limit the scope of the method. The following description mainly uses a second device as an example.

[0030] The method may include: a second device sending fourth information, the fourth information being used to indicate a first model, the first model being a bilateral model or a first unilateral model, the bilateral model being used for data encoding and data decoding, the first unilateral model being used for data encoding or data decoding, the first model being used by the first device to determine a second model, the second model being used for data encoding or data decoding.

[0031] Using the above method, the first device can determine a second model based on at least one first model obtained. Only one AI model for data encoding and / or one AI model for data decoding can be retained in the network, thereby reducing the overhead of storing AI models in terminal devices and / or network devices.

[0032] In conjunction with the second aspect, in some implementations of the second aspect, when the first model is a first unilateral model, the above method further includes: the second device receiving first information, which is used to indicate the second model, which is also a first unilateral model.

[0033] Using the above method, the first device can select a second model from at least one first model for data encoding or data decoding, thereby reducing the overhead of storing AI models for data encoding or data decoding in terminal devices or network devices.

[0034] In conjunction with the second aspect, in some implementations of the second aspect, when the first model is the first unilateral model, the second model is the second unilateral model.

[0035] Wherein, if the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data decoding.

[0036] Using the above method, the first device can determine a second unilateral model based on at least one first unilateral model, which can ensure that the terminal devices and network devices in the network retain only one AI model for data encoding or data decoding, thereby reducing the overhead of storing AI models for terminal devices and network devices.

[0037] In conjunction with the second aspect, in some implementations of the second aspect, the above method further includes: the second device receiving third information, the third information being used to request an update to the first model; the second device sending fourth information includes: the second device sending fourth information based on the third information, the fourth information indicating that the first model is the updated model.

[0038] For example, the first device can periodically send third information, and correspondingly, the second device can periodically receive the third information.

[0039] Using the above method, the first device can trigger real-time updates to the AI ​​model in the network, ensuring the accuracy of data encoding and decoding.

[0040] In conjunction with the second aspect, in some implementations of the second aspect, the aforementioned fourth information includes at least one of the following:

[0041] The first model's model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the first model can be obtained.

[0042] The above methods can provide multiple ways to send models, increasing the flexibility of model sending.

[0043] Thirdly, a communication device is provided. The device includes: a transceiver unit configured to acquire at least one first model, the first model being a bilateral model or a first unilateral model, the bilateral model being used for data encoding and data decoding, and the first unilateral model being used for data encoding or data decoding; the device further includes: a processing unit configured to determine a second model based on at least one first model, the second model being used for data encoding or data decoding; the transceiver unit is further configured to transmit first information, the first information being used to indicate the second model.

[0044] In conjunction with the third aspect, in some implementations of the third aspect, when the first model is a first unilateral model, the above-mentioned processing unit is used to determine a second model based on at least one first model, including: the processing unit is used to determine a second model based on at least one first model and at least one local model, the second model belonging to at least one first model, and the local model being a second unilateral model.

[0045] Wherein, if the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data decoding.

[0046] Optionally, the above processing unit can also be used to determine a second model and a second model A based on at least one first model and at least one local model, wherein the second model belongs to at least one first model and the second model A belongs to at least one local model.

[0047] In conjunction with the third aspect, in some implementations of the third aspect, when the first model is a first unilateral model, the aforementioned transceiver unit is further configured to acquire at least one third model, which is a second unilateral model; the aforementioned processing unit is configured to determine a second model based on at least one first model, including: the aforementioned processing unit is configured to determine a second model and a first fourth model based on at least one first model and at least one third model, wherein the second model belongs to at least one first model and the fourth model belongs to at least one third model; the aforementioned transceiver unit is configured to send first information, including: the aforementioned transceiver unit is configured to send first information and second information, wherein the first information is used to indicate the second model and the second information is used to indicate the fourth model.

[0048] Wherein, if the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data decoding.

[0049] Optionally, the above processing unit can also be used to determine a second model and a fourth model based on at least one first model, at least one third model and at least one local model, wherein the at least one local model includes at least one first unilateral model and / or at least one second unilateral model, wherein the second model belongs to at least one first model or the second model belongs to at least one first unilateral model included in at least one local model, and the fourth model belongs to at least one third model or the fourth model belongs to at least one second unilateral model included in at least one local model.

[0050] In conjunction with the third aspect, in some implementations of the third aspect, the aforementioned at least one third model is trained based on at least one first model, or at least one first model is trained based on the third model.

[0051] In conjunction with the third aspect, in some implementations of the third aspect, when the first model is the first unilateral model, the second model is the second unilateral model.

[0052] Wherein, if the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data decoding.

[0053] In conjunction with the third aspect, in some implementations of the third aspect, the aforementioned transceiver unit is further configured to send third information, which is used to request an update of the first model; the aforementioned transceiver unit is configured to obtain at least one first model, which is an updated model, based on the third information.

[0054] In conjunction with the third aspect, in certain implementations of the third aspect, the aforementioned first information includes at least one of the following:

[0055] The second model includes its model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the second model can be obtained.

[0056] The explanation and beneficial effects of the third aspect and any implementation thereof can be referred to the first aspect above, and the third aspect will not be elaborated further.

[0057] Fourthly, a communication device is provided. The device includes: a transceiver unit for transmitting fourth information, the fourth information indicating a first model, the first model being a bilateral model or a first unilateral model, the bilateral model being used for data encoding and data decoding, the first unilateral model being used for data encoding or data decoding, the first model being used by a first device to determine a second model, the second model being used for data encoding or data decoding.

[0058] In conjunction with the fourth aspect, in some implementations of the fourth aspect, when the first model is a first unilateral model, the aforementioned transceiver unit is also used to receive first information, which is used to indicate the second model, which is also a first unilateral model.

[0059] In conjunction with the fourth aspect, in some implementations of the fourth aspect, when the first model is the first unilateral model, the second model is the second unilateral model.

[0060] Wherein, if the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data decoding.

[0061] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the aforementioned transceiver unit is further configured to receive third information, which is used to request an update to the first model; the transceiver unit is configured to send the fourth information, which is based on the third information, and the first model indicated by the fourth information is the updated model.

[0062] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the aforementioned fourth information includes at least one of the following:

[0063] The first model's model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the first model can be obtained.

[0064] The explanation and beneficial effects of the fourth aspect and any implementation thereof can be found in the second aspect above, and the fourth aspect will not be elaborated further.

[0065] Fifthly, a communication apparatus is provided, which is used to perform the method in the first aspect or any possible implementation thereof, or the apparatus is used to perform the method in the second aspect or any possible implementation thereof. Specifically, the apparatus may include units and / or modules for performing the method in the first aspect or any possible implementation thereof, or the apparatus may include units and / or modules for performing the method in the second aspect or any possible implementation thereof, such as a processing unit and / or a communication unit.

[0066] In one implementation, the device is a communication device (such as a first device, or a second device). When the device is a communication device, the communication unit can be a transceiver, or an input / output interface; the processing unit can be at least one processor. Optionally, the transceiver can be a transceiver circuit. Optionally, the input / output interface can be an input / output circuit.

[0067] In another implementation, the device is a chip, chip system, circuit, or communication module for a communication device (such as the first device or the second device). When the device is a chip, chip system, or circuit for a communication device, the communication unit may be an input / output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip, chip system, or circuit; the processing unit may be at least one processor, processing circuit, or logic circuit.

[0068] A sixth aspect provides a communication device, comprising: at least one processor for executing a computer program or instructions to perform the method in any possible implementation of the first or second aspect described above. Optionally, the device further comprises a memory for storing the computer program or instructions. Optionally, the device further comprises a communication interface coupled to the processor, which can be used to input the computer program or instructions to the processor or to output information from the processor.

[0069] In one implementation, the device is a communication device (such as a first device, or a second device).

[0070] In another implementation, the device is a chip, chip system, circuit, or communication module for a communication device (such as the first device or the second device).

[0071] In a seventh aspect, a processor is provided for performing the methods provided in the first or second aspect described above.

[0072] Unless otherwise specified, or if it does not contradict its actual function or internal logic in the relevant description, the transmission and acquisition / reception operations involved in the processor can be understood as processor output and reception, input and other operations, or as transmission and reception operations performed by radio frequency circuits and antennas. This application does not limit them in this regard.

[0073] Optionally, the device further includes: a memory for storing a program; correspondingly, at least one processor for executing the computer program or instructions in the memory.

[0074] Optionally, the device also includes a communication interface. The communication interface is coupled to the processor and can be used to input information to the processor or output information from the processor.

[0075] Eighthly, a computer-readable storage medium is provided that stores program code for execution by a device, the program code including methods for performing any possible implementation of the first or second aspect described above.

[0076] Ninth aspect, a computer program product containing instructions is provided, which, when run on a computer, causes the computer to perform the method in any possible implementation of the first or second aspect described above.

[0077] In a tenth aspect, a chip is provided, the chip including a processor and a communication interface, the processor reading instructions from a memory through the communication interface and executing the method provided by any of the above implementations of the first or second aspect.

[0078] Optionally, the chip is a modem chip, also known as a baseband chip, or a system-on-chip (SoC) chip containing a modem core or a system-in-package (SIP) chip.

[0079] Optionally, as one implementation, the chip also includes a memory storing computer programs or instructions, and a processor for executing the computer programs or instructions in the memory. When the computer programs or instructions are executed, the processor is used to perform the method provided by any of the above implementations of the first or second aspect.

[0080] Eleventhly, a computer program product containing instructions is provided, which, when run on a computer, causes the computer to perform the method provided by any of the above implementations of the first or second aspect.

[0081] In a twelfth aspect, a communication system is provided, including the aforementioned first device and / or second device. Attached Figure Description

[0082] Figure 1 is a schematic diagram of a communication architecture applicable to this application.

[0083] Figure 2 is a schematic flowchart of a communication method 200 provided in an embodiment of this application.

[0084] Figure 3 is a schematic flowchart of a communication method 300 provided in an embodiment of this application.

[0085] Figure 4 is a schematic flowchart of a communication method 400 provided in an embodiment of this application.

[0086] Figure 5 is a schematic flowchart of a communication method 500 provided in an embodiment of this application.

[0087] Figure 6 is a schematic block diagram of a communication device 600 provided in an embodiment of this application.

[0088] Figure 7 is a schematic block diagram of another communication device 700 provided in an embodiment of this application.

[0089] Figure 8 is a schematic block diagram of a chip system 800 provided in an embodiment of this application. Detailed Implementation

[0090] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0091] To facilitate understanding of the embodiments of this application, the following points will be explained before introducing the embodiments of this application.

[0092] (1) In this application, “for indicating” or “indication” can include both direct and indirect indication, or in other words, “for indicating” or “indication” can be explicit and / or implicit. For example, when describing information for indicating information I, it can include the information directly indicating I or indirectly indicating I, but does not necessarily mean that the information carries I.

[0093] (2) In the embodiments shown below, the terms "first," "second," "third," "fourth," and various other designations are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application. For example, distinguishing different messages, etc.

[0094] (3) In the embodiments of this application, the words “exemplary,” “for example,” “exemplary,” “as another example,” etc., are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design that is described as an “exemplary” in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word “exemplary” is intended to present the concept in a concrete manner.

[0095] (4) The terms “including,” “contains,” “have,” and variations thereof all mean “including but not limited to,” unless otherwise specifically emphasized.

[0096] (5) In the embodiments of this application, the relevant descriptions involving A sending messages, information or data to B, and B receiving messages, information or data from A are intended to indicate which object the message, information or data is to be sent to, and do not limit whether they are sent directly or indirectly through other nodes.

[0097] The technical solutions of this application can be applied to various communication systems, including but not limited to: 5th generation (5G) systems or new radio (NR) systems, long term evolution (LTE) systems, long term evolution-advanced (LTE-A) systems, wireless local area network (WLAN) systems, satellite communication systems, optical communication systems, microwave communication systems, etc. They can also be applied to future communication systems, such as future communication networks and converged systems of multiple systems. Furthermore, they can be applied to device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-to-machine (M2M) communication, machine-type communication (MTC), and Internet of Things (IoT) communication systems or other communication systems. Furthermore, it can be extended to similar wireless communication systems, such as wireless-fidelity (Wi-Fi), worldwide interoperability for microwave access (WIMAX), and communication systems related to the 3rd generation partnership project (3GPP), without limitation.

[0098] The communication architecture applicable to the embodiments of this application will be described next.

[0099] Figure 1 is a schematic diagram of an applicable communication architecture 100 according to an embodiment of this application. As shown in Figure 1, the communication architecture 100 may include: a network management system (NMS) or service management and orchestration (SMO) function, an element management system (EMS), over-the-top (OTT) transmission, network devices, and terminal devices.

[0100] Specifically, the NMS is responsible for the operation, management, and maintenance of the network. It can also be called a cross-domain management system. In the 3rd Generation Partnership Project (3GPP) network domain, the NMS directly manages the EMS.

[0101] The SMO plays a similar role to the NMS in the network architecture, responsible for the operation, management, and maintenance of various network services and orchestration functions. In the Open Radio Access Network (ORAN) domain, the network elements directly managed by the SMO can be heterogeneous, such as the EMS, network devices, and network data analytics functions (NWDAF).

[0102] OTT generally refers to the vendor server of network equipment or the vendor server of terminal equipment. AI models or data in the network can be made available to the vendor server's OTT via NMS / SMO, and the vendor server's OTT can interact with NMS / SMO. The AI ​​models or data of the terminal device can also interact with the vendor server's OTT.

[0103] EMS is used to manage one or more network elements of a certain category. It can also be called a domain management system or a single-domain management system. In the radio access network (RAN) domain, EMS can provide comprehensive management of individual network devices.

[0104] A network device is a device with wireless transceiver capabilities used to communicate with terminal devices. Network devices can be nodes in a Radio Area Network (RAN), and can be called base stations or RAN nodes. They can be base stations in 5G networks such as Evolved Node B (eNB or eNodeB) of Long Term Evolution (LTE) and gNodeB (gNB), or base stations, broadband network gateways (BNGs), aggregation switches, or 3GPP access equipment in public land mobile networks (PLMNs) that evolve after 5G.

[0105] The aforementioned RAN can be configured as a RAN, O-RAN, or cloud radio access network (C-RAN) as defined by the 3GPP protocol. Network equipment can also include various forms of base stations, such as: macro base stations, micro base stations (also known as small stations), relay stations, transmission reception points (TRPs), transmitting points (TPs), mobile switching centers, and equipment that performs base station functions in device-to-device (D2D), vehicle-to-everything (V2X), and machine-to-machine (M2M) communications, as well as network equipment in non-terrestrial networks (NTNs), etc., without specific limitations.

[0106] Network equipment can also be network elements or modules capable of implementing some of the functions of a base station. For example, network equipment can be one or more of the following: a centralized unit (CU), a distributed unit (DU), or a radio unit (RU). The functions of the CU and DU can be implemented by different network elements, or simultaneously by the base band unit (BBU) of the base station. Optionally, the CU can be further separated into a CU-control plane (CP) and a CU-user plane (UP). The functions of the RU can be implemented by the base station's radio frequency equipment. For example, the base station's radio frequency equipment can be a remote radio unit (RRU), a pico remote radio unit (pRRU), an active antenna unit (AAU), or other units, modules, or devices with radio frequency processing capabilities. The communication interface protocol between the BBU and the radio frequency equipment can be the Common Public Radio Interface (CPRI) protocol, the Enhanced Common Public Radio Interface (eCPRI) protocol, or the fronthaul interface protocol between the DU and RU in the O-RAN system, etc., without restriction.

[0107] The communication device used to implement the functions of a network device can be a network device itself, or a device capable of supporting the network device in implementing those functions, such as a chip system. This device can be installed in or used in conjunction with the network device. The chip system in this application embodiment can be composed of chips, or it can include chips and other discrete components.

[0108] Terminal equipment is a device with wireless transceiver capabilities. It can refer to user equipment (UE), access terminal, subscriber unit, user station, mobile station, remote station, remote terminal, mobile device, user terminal, wireless communication equipment, user agent, or user device. Terminal devices can also be satellite phones, cellular phones, smartphones, wireless data cards, wireless modems, machine-type communication devices, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), customer-premises equipment (CPEs), point-of-sale (POS) machines, handheld devices with wireless communication capabilities, computing devices or other processing devices connected to a wireless modem, in-vehicle devices, communication devices mounted on high-altitude aircraft, wearable devices, drones, robots, terminals in device-to-device (D2D) communication, terminals in vehicle-to-everything (V2X) communication, virtual reality (VR) terminal devices, augmented reality (AR) terminal devices, wireless terminals in industrial control, wireless terminals in self-driving vehicles, wireless terminals in remote medical care, and smart grids. This application does not limit the types of wireless terminals, such as those in grids, transportation safety, smart cities, smart homes, or communication networks that evolve after 5G.

[0109] The communication device used to implement the functions of a terminal device can be the terminal device itself, or it can be a device that supports the terminal device in implementing those functions, such as a chip system. This device can be installed in the terminal device or used in conjunction with the terminal device. In this application, the chip system can be composed of chips, or it can include chips and other discrete components.

[0110] It should be noted that the above-mentioned communication architecture 100 may include any number of EMS, any number of OTT Servers, any number of network devices, or any number of terminal devices, and this application does not limit it.

[0111] It should be noted that the communication architecture 100 applicable to the embodiments of this application is merely an example, and the communication architecture applicable to this application is not limited thereto. Any communication architecture capable of implementing the functions of the aforementioned network elements is applicable to this application. In other words, the communication architecture described in this application is for the purpose of more clearly illustrating the technical solutions of this application and does not constitute a limitation on the technical solutions provided in this application. Those skilled in the art will understand that with the evolution of system architecture or the emergence of new business scenarios, the technical solutions provided in this application are also applicable to similar technical problems.

[0112] Furthermore, the network element names and message / information names mentioned in this application are merely examples. In future communication networks, these network elements and messages / information may also use other names, as long as they have the same or similar functions as those described in this application and achieve the same or similar technical objectives, they should all fall within the technical scope covered by this application. For example, in future communication networks, some or all of the names of the aforementioned network elements may be retained from 4G / 5G, or new names may be adopted.

[0113] To optimize air interface transmission, a working mode for a bilateral CSI compression feedback enhancement AI model is proposed. In this mode, the terminal device collects CSI data and compresses it using a pre-trained AI compression model before transmitting it to the network device. The network device then decompresses the CSI using a pre-trained AI decompression model and performs downlink precoding based on the decompressed CSI. Regarding the selection of the AI ​​compression and decompression models, two schemes exist: Scheme 1: Each network device manufacturer can train a bilateral model suitable for its network devices. This bilateral model includes an AI compression model for the terminal device and an AI decompression model for the network device. The trained AI compression model for the terminal device is registered in the network for terminal devices to download. Each terminal device downloads multiple AI compression models trained by various network device manufacturers and selects the AI ​​compression model trained by that network device manufacturer when establishing a connection with a particular network device. Option 2: Each terminal device manufacturer can train a two-sided model applicable to its manufactured terminal devices. This two-sided model includes an AI compression model on the terminal device side and an AI decompression model on the network device side. The trained network device-side AI decompression model is registered in the network for network devices to download. Each network device will download multiple AI decompression models trained by multiple terminal device manufacturers and select the AI ​​decompression model trained by that terminal device manufacturer when establishing a connection with a particular terminal device. Since there are many manufacturers of both network devices and terminal devices, the terminal devices in Option 1 need to download and store a large number of AI compression models, and the network devices in Option 2 need to download and store a large number of AI decompression models, resulting in high storage overhead.

[0114] Based on the above technical problems, this application provides a communication 200 that can reduce the overhead of storing AI models in terminal devices and network devices.

[0115] Figure 2 is a schematic flowchart of a communication method 200 provided in an embodiment of this application. This embodiment uses a first device and a second device as examples of the execution entities in the interaction illustration to illustrate the method, but this application does not limit the execution entities of this interaction illustration. For example, the first device in Figure 2 can also be a chip, chip system, or processor that supports the methods implemented by the first device, or it can also be a logic module or software that can implement all or part of the first device; the second device can also be a chip, chip system, or processor that supports the methods implemented by the second device, or it can also be a logic module or software that can implement all or part of the second device.

[0116] Communication method 200 may include the following steps:

[0117] S210, the second device sends fourth information to the first device, which is used to instruct the first model. Accordingly, the first device receives the fourth information.

[0118] Specifically, the first model is either a bilateral model or a first unilateral model. The bilateral model is used for data encoding and data decoding, while the first unilateral model is used for data encoding or data decoding.

[0119] The various models in the embodiments of this application are AI models used for information transmission and deployed on both sides of the communication device. The various models involved in the embodiments of this application are illustrated using data encoding and / or data decoding as examples. The various models in the embodiments of this application can also be used for data compression and / or data decompression, etc., and this application does not limit them.

[0120] For example, the first device may be a registry center for model registration, such as an NMS.

[0121] For example, the second device may be a manufacturer's server for the terminal device or an EMS for managing individual network devices; this application does not limit this to any particular device.

[0122] If the second device is a terminal device manufacturer's server, the second device can also be replaced with network functions, network elements, or devices related to the terminal device manufacturer. This application embodiment uses a terminal device manufacturer's server as an example for illustration. If the second device is an EMS used to manage a single network device, the second device can also be replaced with network functions, network elements, or devices related to the network device manufacturer. This application embodiment uses an EMS used to manage a single network device as an example for illustration. This application does not limit this.

[0123] For example, the fourth information is used to indicate the method of obtaining the first model, and the fourth information may include at least one of the following:

[0124] The first model includes its model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the first model can be obtained.

[0125] S210 is illustrated using a second device as an example. There may be at least one second device, each of which can instruct the first model to the first device.

[0126] S212, the first device acquires at least one first model.

[0127] S214, the first device determines a second model based on at least one first model.

[0128] The first and second models can be divided into the following three possible cases:

[0129] Case 1: The first model is a bilateral model, and the second model is either the first unilateral model or the second unilateral model.

[0130] Specifically, the first one-sided model can be used for data encoding or data decoding. If the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data encoding.

[0131] Case 2: The first model is the first unilateral model, and the second model is the first unilateral model.

[0132] In one implementation of case 2, the first device determines a second model based on at least one first model and at least one local model, wherein the at least one local model is a second unilateral model and the second model belongs to at least one first model.

[0133] Specifically, the first one-sided model can be used for data encoding or data decoding. If the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data encoding.

[0134] Case 3: The first model is the first unilateral model, and the second model is the second unilateral model.

[0135] Specifically, the first one-sided model can be used for data encoding or data decoding. If the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data encoding.

[0136] S216, the first device sends first information, which is used to instruct the second model.

[0137] For example, the first information is used to indicate the method of obtaining the second model, and the first information may include at least one of the following:

[0138] The second model includes its model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the second model can be obtained.

[0139] In cases 1 and 2 above, the first device sends the first information to the second device; in case 3 above, the first device sends the first information to the third device.

[0140] In one example, if the second device is a vendor server for terminal devices and the third device is an EMS for managing individual network devices, for cases 1 and 2 above, the second device can deploy the second model indicated by the first device on the vendor's terminal devices for data encoding or decoding by the terminal devices; for case 3 above, the third device can deploy the second model indicated by the first device on the network devices for data encoding or decoding by the network devices.

[0141] In another example, if the second device is an EMS for managing a single network device and the third device is a vendor server for the terminal device, for cases 1 and 2 above, the second device can deploy the second model indicated by the first device on the network device for the network device to perform data encoding or data decoding; for case 3 above, the third device can deploy the second model indicated by the first device on the vendor's terminal device for the terminal device to perform data encoding or data decoding.

[0142] Through the aforementioned communication method 200, only one AI model can be retained on both the terminal device side and the network device side, which can reduce the overhead of storing AI models on the terminal device and the network device.

[0143] The embodiments below provide detailed descriptions of the technical solutions applicable to Situations 1, 2, and 3. The embodiments below are uniformly illustrated with NMS as the first device, OTT#1 as the second device, and EMS#1 as the third device. However, it should be understood that the first device could also be NMS, the second device could be EMS#1, and the third device could be OTT#1; this application does not limit this to any particular device.

[0144] The communication method 300 shown below is an example of a technical solution applicable to the above situation 1. Figure 3 is an exemplary flowchart of the communication method 300.

[0145] Communication method 300 may include the following steps:

[0146] S310, OTT#1 trains the model based on the first dataset to obtain the first model.

[0147] Specifically, the first model is a two-sided model, which can be used for both data encoding and data decoding.

[0148] Specifically, the data included in the first dataset corresponds to the function of the first model; that is, the data in the first dataset is collected according to the function of the first model to be trained. For example, if the first model is a model used for CSI compression or CSI decompression, the first dataset may include raw channel data collected from terminal devices or network devices, such as the raw channel time-frequency resource matrix.

[0149] Communication method 300 is illustrated using an OTT as an example; other OTTs may exist. Other OTTs will also perform similar receiving or sending actions as OTT#1, which will not be elaborated here.

[0150] S312, OTT#1 sends a fourth message to the NMS, which indicates the first model. Accordingly, the NMS receives the fourth message.

[0151] For example, the fourth information may include at least one of the following:

[0152] The first model includes its model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the first model can be obtained.

[0153] For example, OTT#1 can indicate the first model to the NMS in the following four ways:

[0154] 1) Send the model file of the first model. For example, send model files in formats such as .bin, .pt, .json, .onnx, etc.

[0155] 2) Send the training dataset, model structure, loss function type, and training method of the first model. For example, the training dataset of the first model can be the first dataset mentioned above, the loss function type can be the function expression of the loss function formula, parameter values, etc., and the training method can be the type of optimizer used for training (Adam, SGD, etc.).

[0156] 3) Send the model structure and parameter indication information of the first model. For example, the parameter indication information can be a list of parameter matrices.

[0157] 4) Sending the first model can obtain the device address.

[0158] S314, NMS acquires at least one first model.

[0159] S316, NMS determines a second model and a fourth model based on at least one first model.

[0160] Specifically, the second model is a first one-sided model, and the fourth model is a second one-sided model. For example, if the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; conversely, if the first one-sided model is used for data decoding, the second one-sided model is used for data encoding. For example, the first device performs simulation tests on each of the at least one first model to determine the accuracy of each first model for original channel recovery, and selects the first model with the best original channel recovery accuracy. Based on this first model, a second model and a fourth model are trained.

[0161] Optionally, NMS can also determine a second model and a fourth model based on at least one first model and the local bilateral model.

[0162] S318, the NMS sends first information to OTT#1, which indicates the second model. Accordingly, OTT#1 receives the first information.

[0163] For example, the first information may include at least one of the following:

[0164] The second model includes its model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the second model can be obtained.

[0165] For example, NMS can indicate a second model to OTT#1 in the following four ways:

[0166] 1) Send the model file for the second model. For example, send model files in formats such as .bin, .pt, .json, and .onnx.

[0167] 2) Send the training dataset, model structure, loss function type, and training method for the second model. For example, the loss function type can be the function expression of the loss function formula, parameter values, etc., and the training method can be the type of optimizer used for training (Adam, SGD, etc.).

[0168] 3) Send the model structure and parameter indication information of the second model. For example, the parameter indication information can be a list of parameter matrices.

[0169] 4) Sending the second model can obtain the device address.

[0170] S320, OTT#1 deploys the second model to the vendor's terminal devices.

[0171] If the NMS instructs the second model to the OTT#1 through method 1) or method 4) shown in S318 above, the OTT#1 can directly obtain the second model; if the NMS instructs the second model to the OTT#1 through method 2) or method 3) shown in S318 above, the OTT#1 needs to train the model based on the training dataset, model structure, loss function type, training method, parameter indication information, etc. of the second model to obtain the second model. The OTT#1 can also optimize, accelerate, and adapt the second model according to the hardware of this manufacturer (for example, customize application programming interfaces (APIs)) so that the second model can run on the chip and operating system of the terminal device of this manufacturer.

[0172] For example, OTT#1 can deploy the second model to the vendor's terminal devices via over-the-air (OTA) technology.

[0173] S322, the NMS sends a second message to EMS#1, which indicates the fourth model. Accordingly, EMS#1 receives the second message.

[0174] For example, the second information may include at least one of the following:

[0175] The fourth model includes its model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the fourth model can be obtained.

[0176] For example, NMS can indicate the fourth model to EMS#1 in the following four ways:

[0177] 1) Send the model file for the fourth model. For example, send model files in formats such as .bin, .pt, .json, and .onnx.

[0178] 2) Send the training dataset, model structure, loss function type, and training method for the fourth model. For example, the loss function type can be the function expression of the loss function formula, parameter values, etc., and the training method can be the type of optimizer used for training (Adam, SGD, etc.).

[0179] 3) Send the model structure and parameter indication information of the fourth model. For example, the parameter indication information can be a list of parameter matrices.

[0180] 4) Sending the fourth model can obtain the device address.

[0181] S324, EMS#1 deploys the fourth model to the managed network devices.

[0182] If the NMS instructs the fourth model to EMS#1 through method 1) or method 4) shown in S322 above, EMS#1 can directly obtain the fourth model; if the NMS instructs the fourth model to EMS#1 through method 2) or method 3) shown in S322 above, EMS#1 needs to train the model based on the training dataset, model structure, loss function type, training method, parameter indication information, etc. of the fourth model to obtain the fourth model. EMS#1 can also optimize, accelerate, and adapt the fourth model according to the hardware of the network device manufacturer (e.g., customize the API) so that model #2 can run on the chip and operating system of the managed network device.

[0183] Through the aforementioned communication method 300, NMS can determine a set of AI models for data encoding and decoding for the entire network, avoiding the need for terminal devices and network devices to store a large number of AI models and saving storage costs.

[0184] The communication method 400 shown below is an example of a technical solution applicable to situation 2 above. Figure 4 is an exemplary flowchart of the communication method 400.

[0185] Communication method 400 may include the following steps:

[0186] S410, OTT#1 is trained on the first dataset to obtain the first model #1.

[0187] Specifically, the first model #1 is the first one-sided model, which can be used for data encoding or data decoding.

[0188] Specifically, the data included in the first dataset corresponds to the function of the first model #1, that is, the data in the first dataset is collected according to the function of the first model #1 to be trained. For example, if the first model #1 is a model used for CSI compression or CSI decompression, the first dataset may include raw channel data collected from terminal devices and network devices, such as the raw channel time-frequency resource matrix.

[0189] Communication method 400 is illustrated using a single OTT provider for a terminal device as an example; other OTT providers may exist. These other OTT providers will also perform similar receiving or sending actions as OTT#1, which will not be elaborated upon here.

[0190] Specifically, OTT#1 can also assign identification information #1 to the first model #1 trained, which includes the identifier of OTT#1 and identifier #1.

[0191] S412, OTT#1 instructs EMS#1 to first model #1.

[0192] For example, OTT#1 can indicate the first model #1 to EMS#1 in the following four ways:

[0193] 1) Send the model file for the first model #1. For example, send a model file in formats such as .bin, .pt, .json, or .onnx.

[0194] 2) Send the training dataset, model structure, loss function type, and training method for the first model #1. For example, the training dataset for the first model #1 can be the first dataset mentioned above, the loss function type can be the function expression of the loss function formula, parameter values, etc., and the training method can be the type of optimizer used for training (Adam, SGD, etc.).

[0195] 3) Send the model structure and parameter indication information of the first model #1. For example, the parameter indication information can be a list of parameter matrices.

[0196] 4) Sending the first model #1 will retrieve the device address.

[0197] Communication method 400 is illustrated using an EMS for managing network devices as an example; other EMSs may exist. Other EMSs will also perform similar receiving or sending actions as EMS#1, which will not be elaborated here.

[0198] S414, EMS#1 trains a model based on at least one first model to obtain at least one third model.

[0199] Specifically, the third model is the second one-sided model. If the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data encoding.

[0200] For example, EMS#1 trains a model based on a first model#1 from OTT#1 to obtain a third model#1. EMS#1 can assign identification information#2 to this third model#1, which includes the identifier of OTT#1 and identifier#2. The identifier of OTT#1 can be considered as the unique virtual identifier of the terminal device manufacturer OTT#1, and the unique virtual identifier of OTT#1 can be assigned by NMS.

[0201] Similarly, if another OTT (e.g., OTT#2) exists, OTT#2 performs model training in S410 to obtain a first model #2, and assigns identification information #3 to this second model #2. This identification information #3 includes the identifier of OTT#2 and identifier #3. EMS#1 can perform model training based on the first model #2 from OTT#2 to obtain a third model #2. EMS#1 can assign identification information #4 to this third model #2, which includes the identifier of OTT#2 and identifier #4.

[0202] Since both the first model #1 and the third model #1 include the identifier OTT #1, it can be considered that the first model #1 and the third model #1 have a corresponding relationship, or it can be understood that the first model #1 and the third model #1 can be used as a set of models for simulation testing in S428 below. Similarly, since both the first model #2 and the third model #2 include the identifier OTT #2, it can be considered that the first model #2 and the third model #2 have a corresponding relationship, or it can be understood that the first model #2 and the third model #2 can be used as a set of models for simulation testing in S428 below.

[0203] S410 to S414 shown above is one method for obtaining at least one first model through OTT#1 and at least one third model through EMS#1. The at least one third model in this method is obtained by training based on at least one first model.

[0204] S416 to S420 shown below are two methods for obtaining at least one first model using OTT#1 and at least one third model using EMS#1. In this method, at least one first model is obtained by training based on at least one third model.

[0205] S416, EMS#1 was trained on the second dataset to obtain the third model #1.

[0206] Specifically, the third model #1 is the second one-sided model, which can be used for data encoding or data decoding.

[0207] Specifically, the data included in the second dataset corresponds to the function of the third model #1, that is, the data in the second dataset is collected according to the function of the third model #1 to be trained. For example, if the third model #1 is a model used for CSI compression or CSI decompression, the second dataset may include raw channel data collected from terminal devices or network devices, such as the raw channel time-frequency resource matrix.

[0208] Communication method 400 is illustrated using one EMS as an example; other EMSs may exist. Other EMSs will also perform similar receiving or sending actions as EMS#1, which will not be described in detail here.

[0209] Specifically, EMS#1 can also assign identification information #1 to the trained third model #1, which includes the identifier of EMS#1 and identifier #1.

[0210] S418, EMS#1 instructs OTT#1 to the third model #1.

[0211] For example, EMS#1 can indicate the third model #1 to OTT#1 in the following four ways:

[0212] 1) Send the model file for the third model #1. For example, send model files in formats such as .bin, .pt, .json, .onnx, etc.

[0213] 2) Send the training dataset, model structure, loss function type, and training method for the third model #1. For example, the training dataset for the third model #1 can be the second dataset mentioned above, the loss function type can be the function expression of the loss function formula, parameter values, etc., and the training method can be the type of optimizer used for training (Adam, SGD, etc.).

[0214] 3) Send the model structure and parameter indication information of the third model #1. For example, the parameter indication information can be a list of parameter matrices.

[0215] 4) Sending the third model #1 will retrieve the device address.

[0216] Communication method 400 is illustrated using an OTT as an example; other OTTs may exist. Other OTTs will also perform similar receiving or sending actions as OTT#1, which will not be elaborated here.

[0217] S420, OTT#1 trains a model based on at least one third model to obtain at least one first model.

[0218] Specifically, the first model is a first one-sided model. If the second one-sided model is used for data encoding, the first one-sided model is used for data decoding; if the second one-sided model is used for data decoding, the first one-sided model is used for data encoding.

[0219] For example, OTT#1 trains a model based on a third model #1 from EMS#1 to obtain a first model #1. OTT#1 can assign identification information #2 to this first model #1, which includes the identifier of EMS#1 and identifier #2. The identifier of EMS#1 can be considered as a unique virtual identifier (Vendor virtual identifier) ​​of EMS#1 used to manage a single network device, and the unique virtual identifier (Vendor virtual identifier) ​​of EMS#1 can be assigned by NMS.

[0220] Similarly, if another EMS exists (e.g., EMS#2), EMS#2 is trained in S416 above to obtain a third model #2, and identification information #3 is assigned to this third model #2, which includes the identifier of EMS#2 and identifier #3. OTT#1 can train a model based on the third model #2 from EMS#2 to obtain a first model #2. OTT#1 can assign identification information #4 to this first model #2, which includes the identifier of EMS#2 and identifier #4.

[0221] Since both the third model #1 and the first model #1 include the identifier EMS #1, it can be assumed that the third model #1 and the first model #1 have a corresponding relationship. It can also be understood that the third model #1 and the first model #1 can be used as a set of models for simulation testing in S428 below. Similarly, since both the third model #2 and the first model #2 include the identifier EMS #2, it can be assumed that the third model #2 and the first model #2 have a corresponding relationship. It can also be understood that the third model #2 and the first model #2 can be used as a set of models for simulation testing in S428 below.

[0222] S422, OTT#1 indicates at least one first model to NMS.

[0223] For example, in the first method described above, OTT#1 obtains the first model #1 and instructs the first model #1 to the NMS; in the second method described above, OTT#1 obtains the first model #1 and the first model #2 and instructs the first model #1 and the first model #2 to the NMS.

[0224] For example, the method by which OTT#1 indicates at least one first model to NMS can refer to the four methods of OTT#1 indicating first model #1 to EMS#1 in S412 above, which will not be repeated here.

[0225] S424, EMS#1 indicates at least one third model to NMS.

[0226] For example, in the first method described above, EMS#1 obtains the third model #1 and the third model #2, and sends the third model #1 and the third model #2 to the NMS; in the second method described above, EMS#1 obtains the third model #1 and sends the third model #1 to the NMS.

[0227] For example, the method by which EMS#1 indicates at least one third model to NMS can refer to the four methods of EMS#1 indicating third model #1 to OTT#1 in S418 above, which will not be repeated here.

[0228] The process by which each OTT indicates a model to the NMS in the embodiments of this application can be considered as the process by which each OTT registers a model with the NMS, and the process by which each EMS indicates a model to the NMS can be considered as the process by which each EMS registers a model with the NMS.

[0229] S426, NMS acquires at least one first model and at least one third model.

[0230] S428, NMS determines a second model and a fourth model based on at least one first model and at least one third model.

[0231] Specifically, at least one first model and at least one third model have a corresponding relationship. Specifically, one of the first models in at least one first model has a corresponding relationship with multiple third models in at least one third model, or one of the third models in at least one third model has a corresponding relationship with multiple first models in at least one first model.

[0232] At least one first model and at least one third model can form multiple sets of models according to their correspondence. Taking at least one first model and at least one third model obtained by the above method one as an example, the first model #1 trained by OTT#1 and the third model #1 trained by EMS#1 based on the first model #1 are the first set of models, the first model #2 trained by OTT#2 and the third model #2 trained by EMS#1 based on the first model #2 are the second set of models, and so on.

[0233] NMS performs simulation tests on each of the multiple model sets to determine at least one of several performance metrics for each model set, such as compression ratio, decompression accuracy, time required for a single computation, and computational efficiency. It then selects the model set with the best overall performance. For example, if NMS determines that the first model set has the best overall performance, it designates model #1 from the first model set as the second model and model #1 from the third model set as the fourth model. That is, the second model belongs to at least one first model, and the fourth model belongs to at least one third model. The overall performance evaluation metric can be one or more of compression ratio, decompression accuracy, time required for a single computation, and computational efficiency. If the overall performance evaluation metric includes multiple metrics such as compression ratio, decompression accuracy, time required for a single computation, and computational efficiency, a weighted summation method can be used to determine the overall performance.

[0234] Optionally, NMS can also determine a second model and a fourth model based on at least one first model, at least one third model, and a local model. In this case, the second model does not necessarily belong to at least one first model; the second model may be a local model. Similarly, the fourth model does not necessarily belong to at least one third model; it may also be a local model.

[0235] S430, the NMS sends first information to OTT#1, which indicates the second model. Accordingly, OTT#1 receives the first information.

[0236] For example, the first information may include at least one of the following:

[0237] The second model includes its model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the second model can be obtained.

[0238] For example, NMS can indicate a second model to OTT#1 in the following four ways:

[0239] 1) Send the model file for the second model. For example, send model files in formats such as .bin, .pt, .json, and .onnx.

[0240] 2) Send the training dataset, model structure, loss function type, and training method for the second model. For example, the loss function type can be the function expression of the loss function formula, parameter values, etc., and the training method can be the type of optimizer used for training (Adam, SGD, etc.).

[0241] 3) Send the model structure and parameter indication information of the second model. For example, the parameter indication information can be a list of parameter matrices.

[0242] 4) Sending the second model can obtain the device address.

[0243] S432, OTT#1 deploys the second model to the vendor's terminal devices.

[0244] If the NMS instructs the second model to the OTT#1 through method 1) or method 4) shown in S430 above, the OTT#1 can directly obtain the second model; if the NMS instructs the second model to the OTT#1 through method 2) or method 3) shown in S430 above, the OTT#1 needs to train the model based on the training dataset, model structure, loss function type, training method, parameter indication information, etc. of the second model to obtain the second model. The OTT#1 can also optimize, accelerate, and adapt the second model according to the hardware of this manufacturer (e.g., customize the API) so that the second model can run on the chip and operating system of the terminal device of this manufacturer.

[0245] For example, OTT#1 can deploy the second model to the vendor's terminal devices via OTA technology.

[0246] S434, NMS sends a second message to EMS#1, which indicates the fourth model. Accordingly, EMS#1 receives the second message.

[0247] For example, the second information may include at least one of the following:

[0248] The fourth model includes its model file, training dataset, model structure, loss function type, training method, parameter matrix list, and the address of the device from which the fourth model can be obtained.

[0249] For example, NMS can indicate the fourth model to EMS#1 in the following four ways:

[0250] 1) Send the model file for the fourth model. For example, send model files in formats such as .bin, .pt, .json, and .onnx.

[0251] 2) Send the training dataset, model structure, loss function type, and training method for the fourth model. For example, the loss function type can be the function expression of the loss function formula, parameter values, etc., and the training method can be the type of optimizer used for training (Adam, SGD, etc.).

[0252] 3) Send the model structure and parameter indication information of the fourth model. For example, the parameter indication information can be a list of parameter matrices.

[0253] 4) Sending the fourth model can obtain the device address.

[0254] S436, EMS#1 deploys the fourth model to the managed network devices.

[0255] For example, if the NMS instructs the fourth model to EMS#1 through method 1) or method 4) shown in S434 above, EMS#1 can directly obtain the fourth model; if the NMS instructs the fourth model to EMS#1 through method 2) or method 3) shown in S434 above, EMS#1 needs to train the model based on the training dataset, model structure, loss function type, training method, parameter indication information, etc. of the fourth model to obtain the fourth model. EMS#1 can also optimize, accelerate, and adapt the fourth model according to the hardware of the network device manufacturer (e.g., customize the API) so that the fourth model can run on the chip and operating system of the managed network device.

[0256] Through the aforementioned communication method 400, NMS can determine a set of AI models for data encoding and decoding for the entire network, avoiding the need for terminal devices and network devices to store a large number of AI models and saving storage costs.

[0257] The communication method 500 shown below is an example of a technical solution applicable to situation 3 above. Figure 5 is an exemplary flowchart of the communication method 500.

[0258] Communication method 500 may include the following steps:

[0259] Optionally, in step S510, the NMS sends third information to OTT#1. Accordingly, OTT#1 receives the third information.

[0260] Specifically, the third information is used to request an update to the first model, which is a first one-sided model used for data encoding or data decoding.

[0261] For example, the third information can be sent periodically, that is, the NMS can periodically request OTT#1 to update the first model.

[0262] Communication method 500 is illustrated using an OTT as an example; other OTTs may exist. Other OTTs will also perform similar receiving or sending actions as OTT#1, which will not be elaborated here.

[0263] S512, OTT#1 is trained on the first dataset to obtain the first model #1.

[0264] The execution order of S510 and S512 is not important. It is possible that NMS first sends the third information to OTT#1, and then OTT#1 trains the model based on the first dataset to obtain the first model #1; or, it is possible that OTT#1 first trains the model based on the first dataset to obtain the first model #1, and then NMS sends the third information to OTT#1.

[0265] Specifically, the first model #1 is the first one-sided model, which can be used for data encoding or data decoding.

[0266] Specifically, the data included in the first dataset corresponds to the function of the first model #1, that is, the data in the first dataset is collected according to the function of the first model #1 to be trained. For example, if the first model #1 is a model used for CSI compression or CSI decompression, the first dataset may include raw channel data collected from terminal devices and network devices, such as the raw channel time-frequency resource matrix.

[0267] S514, OTT#1 sends a fourth message to NMS, which is used to indicate the first model #1.

[0268] For example, the fourth information may include at least one of the following:

[0269] The model file, training dataset, model structure, loss function type, training method, parameter matrix list, and address of the device that can be obtained by the first model #1 are all included.

[0270] For example, OTT#1 can indicate the first model #1 to NMS in the following four ways:

[0271] 1) Send the model file for the first model #1. For example, send a model file in formats such as .bin, .pt, .json, or .onnx.

[0272] 2) Send the training dataset, model structure, loss function type, and training method for the first model #1. For example, the training dataset for the first model #1 can be the first dataset mentioned above, the loss function type can be the function expression of the loss function formula, parameter values, etc., and the training method can be the type of optimizer used for training (Adam, SGD, etc.).

[0273] 3) Send the model structure and parameter indication information of the first model #1. For example, the parameter indication information can be a list of parameter matrices.

[0274] 4) Sending the first model #1 will retrieve the device address.

[0275] S516, NMS acquires at least one first model.

[0276] S518, NMS determines a second model based on at least one first model.

[0277] Specifically, NMS trains a model based on at least one first model to obtain a second model, which is a second one-sided model. If the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data encoding.

[0278] For example, NMS updates its local list of first models based on at least one acquired first model, and trains a second model by combining all the first models. For instance, NMS generates a training dataset #1 using all the first models and the historical raw channel time-frequency resource matrix, and trains a model based on this training dataset #1 to obtain a second model.

[0279] S520, NMS sends the first message to EMS#1, which is used to indicate the second model.

[0280] For example, the method by which the NMS instructs the second model to the EMS#1 can refer to the four model instruction methods in S514, which will not be repeated here.

[0281] The above-described S518 to S520 represent one method for obtaining the second model using EMS#1, while the following-described S522 to S524 represent another method for obtaining the second model using EMS#1.

[0282] S522, NMS indicates at least one first model to EMS#1. Accordingly, EMS#1 obtains at least one first model.

[0283] For example, the method by which the NMS indicates at least one first model to the EMS#1 can refer to the three model indication methods in S514, which will not be repeated here.

[0284] S524, EMS#1 determines a second model based on at least one first model.

[0285] Specifically, EMS#1 trains a model based on at least one first model to obtain a second model, which is a second one-sided model. If the first one-sided model is used for data encoding, the second one-sided model is used for data decoding; if the first one-sided model is used for data decoding, the second one-sided model is used for data encoding.

[0286] For example, EMS#1 updates its local list of first models based on at least one acquired first model, and trains a second model by combining all the first models. For instance, EMS#1 uses all the first models and the historical raw channel time-frequency resource matrix to generate a training dataset #1, and trains a model based on this training dataset #1 to obtain a second model.

[0287] S526, EMS#1 deploys the second model to network devices.

[0288] For example, in the above-described method where EMS#1 can obtain the second model, if the NMS instructs EMS#1 to obtain the second model through method 1) or method 4 of the four model transmission methods, EMS#1 can directly obtain the second model; if the NMS instructs EMS#1 to obtain the second model through method 2) or method 3 of the four model transmission methods, EMS#1 needs to train the model based on the training dataset, model structure, loss function type, training method, parameter indication information, etc. of the second model to obtain the second model; EMS#1 can also optimize, accelerate, and adapt the second model according to the hardware of the network device manufacturer (e.g., customize the API) so that the second model can run on the chip and operating system of the network device of this manufacturer.

[0289] S528, OTT#1 deploys the first model #1 to the vendor's terminal devices.

[0290] For example, OTT#1 can deploy its trained first model #1 to the vendor's terminal device via OTA technology.

[0291] Through the aforementioned communication method 500, NMS can train a network device-side model for all network devices in the network based on the terminal device-side models provided by various terminal device manufacturers, so that all terminal devices and network devices in the network retain only one AI model for data encoding or data decoding, saving storage overhead.

[0292] It should be noted that the above-mentioned communication method 500 is that NMS trains a network device side model for all network devices based on the terminal device side model provided by each terminal device manufacturer. NMS can also train a terminal device side model for all terminal devices based on the network device side model provided by each EMS used to manage network devices. This application does not limit this.

[0293] It is understood that some optional features in the embodiments of this application may not depend on other features in certain scenarios, or may be combined with other features in certain scenarios, without limitation.

[0294] It is also understood that the solutions in the embodiments of this application can be used in reasonable combinations, and the explanations or descriptions of the various terms appearing in the embodiments can be referenced or explained to each other in the various embodiments, without limitation.

[0295] It is also understood that, in the above method embodiments, the methods and operations implemented by the device (such as the first device, the second device) can also be implemented by components of the device (such as chips or circuits), without limitation.

[0296] The methods provided by the embodiments of this application have been described in detail above with reference to Figures 1 to 5. The apparatus provided by the embodiments of this application will be described in detail below with reference to Figures 6 to 8. It should be understood that the descriptions of the apparatus embodiments correspond to the descriptions of the method embodiments; therefore, any content not described in detail can be referred to the method embodiments above, and for the sake of brevity, will not be repeated here.

[0297] Referring to Figure 6, which is a schematic diagram of a communication device 600 provided in an embodiment of this application, the device 600 includes a transceiver unit 610. The transceiver unit 610 can be used to implement corresponding communication functions. The transceiver unit 610 can also be referred to as a communication interface or a communication unit.

[0298] Optionally, the device 600 further includes a processing unit 620. The processing unit 620 can be used to perform processing. The functionality of the processing unit 620 can be implemented by one or more processors. Specifically, the processor may include a modem chip, or a system-on-a-chip (SoC) or SIP chip containing a modem core.

[0299] Optionally, the device 600 may further include a storage unit for storing instructions and / or data, and the processing unit 620 may read the instructions and / or data from the storage unit to enable the device to implement the aforementioned method embodiments.

[0300] Optionally, the transceiver unit 610 may include a receiving unit and a sending unit. The receiving unit can be used to perform receiving-related operations (such as receiving data or messages), and the sending unit can be used to perform sending-related operations (such as sending data or messages).

[0301] In a first possible design, the device 600 can be the first device in the aforementioned embodiments, which can implement the steps or processes performed by the first device in the method embodiments shown in Figures 2 to 5 above. The transceiver unit 610 can be used to perform transceiver-related operations (such as sending and / or receiving data or messages) of the first device or NMS in the method embodiments shown in Figures 2 to 5 above. The processing unit 620 can be used to perform processing-related operations of the first device or NMS in the method embodiments shown in Figures 2 to 5 above, or operations other than transceiver (such as operations other than sending and / or receiving data or messages).

[0302] For example, transceiver unit 610 can be used to: acquire at least one first model; transceiver unit 610 can also be used to send a second model.

[0303] For example, processing unit 620 can be used to: determine a second model based on at least one first model.

[0304] In a second possible design, the device 600 can be the second device in the aforementioned embodiments. This device 600 can implement the steps or processes performed by the second device, OTT#1, or EMS#1 corresponding to the method embodiments shown in Figures 2 to 5 above. Specifically, the transceiver unit 610 can be used to perform transceiver-related operations (such as sending and / or receiving data or messages) of the second device, OTT#1, or EMS#1 in the method embodiments shown in Figures 2 to 5 above. The processing unit 620 can be used to perform processing-related operations of the second device, OTT#1, or EMS#1 in the method embodiments shown in Figures 2 to 5 above, or operations other than transceiver operations (such as operations other than sending and / or receiving data or messages).

[0305] For example, transceiver unit 610 can be used to: send a first model to a first device; transceiver unit 610 can also be used to receive a second model from the first device.

[0306] It should be understood that the specific process of each unit performing the above-mentioned corresponding steps has been described in detail in the above method embodiments, and will not be repeated here for the sake of brevity.

[0307] It should also be understood that the device 600 here is embodied in the form of a functional unit. The term "unit" here can refer to an application-specific integrated circuit (ASIC), electronic circuitry, a processor (e.g., a shared processor, a proprietary processor, or a group processor, etc.) and memory for executing one or more software or firmware programs, integrated logic circuitry, and / or other suitable components supporting the described functions. In an alternative example, those skilled in the art will understand that the device 600 can specifically be the communication device in the above embodiments, and can be used to execute the various processes and / or steps corresponding to the communication device in the above method embodiments; to avoid repetition, these will not be described again here.

[0308] The apparatus 600 of each of the above-described schemes has the function of implementing the corresponding steps performed by the communication device in the above-described methods. The function can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions; for example, the transceiver unit can be replaced by a transceiver (e.g., the transmitting unit in the transceiver unit can be replaced by a transmitter, and the receiving unit in the transceiver unit can be replaced by a receiver), and other units, such as processing units, can be replaced by processors, each executing the transceiver operations and related processing operations in the respective method embodiments.

[0309] In addition, the transceiver unit 610 described above can also be a transceiver circuit (for example, it may include a receiving circuit and a transmitting circuit), and the processing unit can be a processing circuit.

[0310] It should be noted that the device in Figure 6 can be the communication device in the foregoing embodiments, or it can be a chip or chip system, such as a modem chip or a system-on-a-chip (SoC) chip or SIP chip containing a modem core. The transceiver unit can be an input / output circuit or a communication interface; the processing unit is a processor, microprocessor, or integrated circuit integrated on the chip. No limitations are imposed here.

[0311] Referring to Figure 7, which is a schematic diagram of another communication device 700 provided in an embodiment of this application, the device 700 includes a processor 710 coupled to a memory 720. The memory 720 is used to store computer programs or instructions and / or data. The processor 710 is used to execute the computer programs or instructions stored in the memory 720, or to read the data stored in the memory 720, to perform the methods in the above-described method embodiments.

[0312] Optionally, there may be one or more processors 710.

[0313] Optionally, the memory 720 may be one or more.

[0314] Alternatively, the memory 720 can be integrated with the processor 710, or it can be set separately.

[0315] Optionally, as shown in FIG7, the device 700 further includes a transceiver 730 for receiving and / or transmitting signals. For example, a processor 710 is used to control the transceiver 730 to receive and / or transmit signals. Exemplarily, the transceiver 730 may include a transmitter and / or a receiver, the transmitter being used to perform a transmission operation and the receiver being used to perform a reception operation.

[0316] As an example, processor 710 may have the functions of processing unit 620 shown in FIG. 6, memory 720 may have the functions of storage unit, and transceiver 730 may have the functions of transceiver unit 610 shown in FIG. 6.

[0317] As one option, the device 700 is used to implement the operations performed by the communication device in the various method embodiments described above.

[0318] For example, processor 710 is used to execute computer programs or instructions stored in memory 720 to implement the relevant operations of terminal devices or network devices in the various method embodiments described above.

[0319] It should be understood that the processor mentioned in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0320] It should also be understood that the memory mentioned in the embodiments of this application can be volatile memory and / or non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM). For example, RAM can be used as an external cache. By way of example and not limitation, RAM includes the following forms: static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0321] It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA, or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, the memory (storage module) can be integrated into the processor.

[0322] It should also be noted that the memory described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0323] Referring to Figure 8, Figure 8 is a schematic diagram of a chip system 800 provided in an embodiment of this application. The chip system 800 (or may also be referred to as a processing system) includes logic circuitry 810 and an input / output interface 820.

[0324] The logic circuit 810 can be a processing circuit in the chip system 800. The logic circuit 810 can be coupled to a memory unit, calling instructions from the memory unit, enabling the chip system 800 to implement the methods and functions of the embodiments of this application. The input / output interface 820 can be an input / output circuit in the chip system 800, outputting processed information from the chip system 800, or inputting data or signaling information to be processed into the chip system 800 for processing.

[0325] Optionally, the logic circuit 810 may be implemented by one or more processors, including the one or more processors or the processing portion of the one or more processors.

[0326] Optionally, the input / output interface 820 may include transceiver circuitry, a transceiver, input / output circuitry, or a communication interface.

[0327] As one approach, the chip system 800 is used to implement operations performed by a communication device (such as a first device or a second device) in the various method embodiments described above.

[0328] For example, logic circuit 810 is used to implement processing-related operations performed by a communication device (such as a first device or a second device) in the above method embodiments; input / output interface 820 is used to implement sending and / or receiving-related operations performed by a communication device (such as a first device or a second device) in the above method embodiments, with the input interface used to perform receiving operations and the output interface used to perform sending operations.

[0329] This application also provides a computer-readable storage medium storing computer instructions for implementing the methods executed by a communication device (such as a first device or a second device) in the above-described method embodiments.

[0330] For example, when the computer program is executed by a computer, it enables the computer to implement the methods performed by the communication device (such as the first device or the second device) in the various embodiments of the above methods.

[0331] This application also provides a computer program product comprising instructions which, when executed by a computer, implement the methods described above that are executed by a communication device (such as a first device or a second device).

[0332] This application also provides a communication system that includes the first device and / or the second device described in the preceding embodiments. For example, the system includes the first device and the second device shown in FIG2.

[0333] The explanations and beneficial effects of the relevant contents in any of the devices provided above can be found in the corresponding method embodiments provided above, and will not be repeated here.

[0334] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of apparatus or units may be electrical, mechanical, or other forms.

[0335] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. For example, the computer can be a personal computer, a server, or a network device, etc. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks, SSDs). For example, the aforementioned available media include, but are not limited to, USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, and other media capable of storing program code.

[0336] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A communication method characterized by comprising: The method comprises: obtaining at least one first model, the first model being a bilateral model or a first unilateral model, the bilateral model being used for data encoding and data decoding, and the first unilateral model being used for data encoding or data decoding; determining a second model based on the at least one first model, the second model being used for data encoding or data decoding; sending first information, the first information being used for indicating the second model.

2. The method of claim 1, wherein, When the first model is the first unilateral model, the determining of the second model based on the at least one first model comprises: determining a second model based on the at least one first model and at least one local model, the second model belonging to the at least one first model, and the local model being a second unilateral model; if the first unilateral model is used for data encoding, the second unilateral model is used for data decoding; or, if the first unilateral model is used for data decoding, the second unilateral model is used for data encoding.

3. The method according to claim 1 or 2, characterized in that, When the first model is the first unilateral model, the method further comprises: obtaining at least one third model, the third model being a second unilateral model; the determining of the second model based on the at least one first model comprises: determining a second model and a fourth model based on the at least one first model and the at least one third model, the second model belonging to the at least one first model, and the fourth model belonging to the at least one third model; the sending of the first information comprises: sending first information and second information, the first information being used for indicating the second model, and the second information being used for indicating the fourth model; if the first unilateral model is used for data encoding, the second unilateral model is used for data decoding; or, if the first unilateral model is used for data decoding, the second unilateral model is used for data encoding.

4. The method of claim 3, wherein, The at least one third model is trained based on the at least one first model, or the at least one first model is trained based on the at least one third model.

5. The method of claim 1, wherein, When the first model is the first unilateral model, the second model is a second unilateral model, if the first unilateral model is used for data encoding, the second unilateral model is used for data decoding; or, if the first unilateral model is used for data decoding, the second unilateral model is used for data encoding.

6. The method according to any one of claims 1 to 5, characterized in that, The method further comprises: sending third information, the third information being used for requesting to update the first model; the obtaining of the at least one first model comprises: obtaining the at least one first model based on the third information, the first model being an updated model.

7. The method according to any one of claims 1 to 6, characterized in that, The first information comprises at least one of the following: a model file of the second model, a training data set, a model structure, a loss function type, a training method, a parameter matrix list, and an address of a device from which the second model can be obtained.

8. A communication method characterized by comprising: The method comprises: transmitting fourth information, the fourth information being used for indicating a first model, the first model being a bilateral model or a first unilateral model, the bilateral model being used for data encoding and data decoding, the first unilateral model being used for data encoding or data decoding, the first model being used for a first device to determine a second model, the second model being used for data encoding or data decoding.

9. The method of claim 8, wherein, When the first model is the first unilateral model, the method further comprises: receiving first information, the first information being used for indicating the second model, the second model also being the first unilateral model.

10. The method of claim 8, wherein, When the first model is the first unilateral model, the second model is a second unilateral model, if the first unilateral model is used for data encoding, the second unilateral model is used for data decoding; or, if the first unilateral model is used for data decoding, the second unilateral model is used for data encoding.

11. The method according to any one of claims 8 to 10, characterized in that, The method further comprises: receiving third information, the third information being used for requesting to update the first model; The transmitting fourth information comprises: transmitting the fourth information based on the third information, the first model indicated by the fourth information being an updated model.

12. The method according to any one of claims 8 to 11, characterized in that, The fourth information comprises at least one of the following: a model file of the first model, a training data set, a model structure, a loss function type, a training method, a parameter matrix list, an address of a device from which the first model can be obtained.

13. An apparatus, comprising: The apparatus comprises units or modules for performing the method of any one of claims 1 to 7, or the apparatus comprises units or modules for performing the method of any one of claims 8 to 12.

14. A system, comprising: comprises: a first device and a second device, the first device being configured to perform the method of any one of claims 1 to 7, and the second device being configured to perform the method of any one of claims 8 to 12.

15. A computer-readable storage medium, characterized in that, The computer program stored in the computer readable storage medium, when executed by a computer, causes the computer to perform the method of any one of claims 1 to 7, or causes the computer to perform the method of any one of claims 8 to 12.

16. A computer program product, characterised in that, The computer program product comprises: computer program code, when the computer program code is executed on a communication apparatus, causes the apparatus to perform the method of any one of claims 1 to 7, or causes the apparatus to perform the method of any one of claims 8 to 12.

17. A chip, characterized by The chip comprises a processor and a communication interface, the processor reads instructions stored on a memory through the communication interface, and performs the method of any one of claims 1 to 7, or performs the method of any one of claims 8 to 12.

Images