WO2023092307A1

WO2023092307A1 - Communication method, model training method, and device

Info

Publication number: WO2023092307A1
Application number: PCT/CN2021/132584
Authority: WO
Inventors: 田文强
Original assignee: Oppo广东移动通信有限公司
Priority date: 2021-11-23
Filing date: 2021-11-23
Publication date: 2023-06-01
Also published as: CN117813801A

Abstract

Embodiments of the present application relates to a communication method, a model training method, and a device. The communication method comprises: a terminal device receives a first signal, the signal being generated by a first model; and the terminal device processes the first signal by using a second model, to obtain first information, the first information comprising channel information, wherein the first model and the second model are obtained by means of joint training. The embodiments of the present application can improve the overall performance of the network.

Description

Communication method, model training method and device

technical field

The present application relates to the field of communication, and more particularly, to a communication method, a model training method and a device.

Background technique

There are uplink and downlink reference signals in the wireless communication system, and these reference signals are used to achieve different purposes such as channel estimation. However, when these reference signals are designed, it is not considered to apply these reference signals to wireless communication solutions based on artificial intelligence (AI, Artificial Intelligence) or neural network methods, so the existing reference signals are different from those based on AI or neural network methods. Wireless communication solutions are difficult to achieve the best matching results. It can be seen that how to complete the AI-based wireless communication solution and the adapted reference signal design as an overall solution to improve the overall advantages of the reference signal design and wireless communication solution has become a problem that needs to be solved.

Contents of the invention

Embodiments of the present application provide a communication method, a model training method, and equipment, which can improve the overall advantages of reference signal design and wireless communication solutions.

The embodiment of this application proposes a communication method, including:

The terminal device receives a first signal, where the first signal is generated by a first model;

The terminal device processes the first signal by using the second model to obtain first information, where the first information includes channel information;

Wherein, the first model and the second model are obtained through joint training.

The embodiment of the present application also proposes a communication method, including:

The network device sends a first signal, where the first signal is generated by a first model; the first signal is used for processing by a second model to obtain first information, where the first information includes channel information;

The embodiment of the present application also proposes a model training method, including:

The input information and/or the first channel simulation module are used to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.

The embodiment of the present application also proposes a terminal device, including:

a first receiving module, configured to receive a first signal, the first signal is generated by a first model;

The first processing module is configured to process the first signal by using the second model to obtain first information, where the first information includes channel information;

The embodiment of this application also proposes a network device, including:

The sixth sending module is configured to send a first signal, the first signal is generated by the first model; the first signal is used for processing by the second model to obtain first information, and the first information includes channel information;

The embodiment of the present application also proposes a model training device, including:

The joint training module is configured to use the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.

The embodiment of the present application also proposes a terminal device, including: a processor, a memory, and a transceiver, the memory is used to store computer programs, the processor is used to call and run the computer programs stored in the memory, and control the The transceiver, performing the method described in any one of the above first communication methods.

The embodiment of the present application also proposes a network device, including: a processor, a memory, and a transceiver, the memory is used to store computer programs, the processor is used to call and run the computer programs stored in the memory, and control the The transceiver, performing the method described in any one of the above second communication methods.

The embodiment of the present application also proposes a model training device, including: a processor and a memory, the memory is used to store a computer program, the processor is used to call and run the computer program stored in the memory, and execute the above-mentioned model training method any one of the methods described.

The embodiment of the present application also proposes a chip, including: a processor, configured to call and run a computer program from the memory, so that the device installed with the chip executes the method described in any one of the above-mentioned first communication methods .

The embodiment of the present application also proposes a chip, including: a processor, configured to call and run a computer program from the memory, so that the device installed with the chip executes the method described in any one of the above-mentioned second communication methods .

The embodiment of the present application also proposes a chip, including: a processor, configured to call and run a computer program from a memory, so that a device installed with the chip executes the method described in any one of the above-mentioned model training methods.

The embodiment of the present application also provides a computer-readable storage medium for storing a computer program, and the computer program causes a computer to execute the method described in any one of the above-mentioned first communication methods.

The embodiment of the present application also provides a computer-readable storage medium for storing a computer program, and the computer program causes a computer to execute the method described in any one of the above-mentioned second communication methods.

The embodiment of the present application also provides a computer-readable storage medium for storing a computer program, and the computer program enables the computer to execute the method described in any one of the above-mentioned model training methods.

The embodiment of the present application also provides a computer program product, including computer program instructions, where the computer program instructions cause a computer to execute the method described in any one of the above first communication methods.

An embodiment of the present application also provides a computer program product, including computer program instructions, where the computer program instructions cause a computer to execute the method described in any one of the above-mentioned second communication methods.

The embodiment of the present application also provides a computer program product, including computer program instructions, the computer program instructions cause a computer to execute the method described in any one of the above model training methods.

The embodiment of the present application also proposes a computer program, the computer program causes a computer to execute the method described in any one of the above-mentioned first communication methods.

The embodiment of the present application also provides a computer program, the computer program causes a computer to execute the method described in any one of the above-mentioned second communication methods.

The embodiment of the present application also proposes a computer program, the computer program causes a computer to execute the method described in any one of the above model training methods.

Using the embodiment of this application, the terminal device uses the second model to process the received first signal, the first signal is generated by the first model, and the first model and the second model are jointly trained, because the first The model and the second model are obtained through joint training, so the performance requirements of the entire signal generation and signal processing can be taken into account, and the overall performance of the network can be improved.

Description of drawings

FIG. 1 is a schematic diagram of an application scenario of an embodiment of the present application.

Fig. 2 is a schematic diagram of a neural network structure.

FIG. 3A is a schematic diagram of a CSI feedback manner.

Fig. 3B is a schematic diagram of a manner of performing channel estimation.

Fig. 3C is a schematic diagram of a positioning method.

Fig. 4 is a schematic flowchart of a communication method 400 according to an embodiment of the present application.

Fig. 5 is a schematic flowchart of another communication method 500 according to an embodiment of the present application.

Fig. 6 is a schematic diagram of a model sending manner according to an embodiment of the present application.

Fig. 7 is a schematic structural diagram of a model according to the present application.

Fig. 8 is a schematic diagram of another model sending manner according to an embodiment of the present application.

FIG. 9 is a schematic diagram of an AI-based multi-user reference signal and channel estimation integrated design scheme for a wireless communication system according to an embodiment of the present application.

Fig. 10 is a schematic diagram of a channel information structure according to the present application.

Fig. 11 is a schematic diagram of another channel information structure according to the present application.

Fig. 12 is a schematic structural diagram of channel feature vector information according to the present application.

Fig. 13 is a schematic diagram of a neural network structure according to the present application.

Fig. 14 is a schematic diagram of model structure and information transmission in an AI-based wireless communication system multi-user reference signal and channel estimation integrated design scheme according to an embodiment of the present application.

Fig. 15 is a schematic diagram of an AI-based integrated design scheme of multi-user reference signal, channel estimation, and channel information feedback in a wireless communication system according to an embodiment of the present application.

Fig. 16 is a schematic diagram of another neural network structure according to the present application.

Fig. 17 is a schematic diagram of model structure and information transmission in an AI-based wireless communication system multi-user reference signal, channel estimation, and channel information feedback integrated design scheme according to an embodiment of the present application.

Fig. 18 is a schematic flowchart of another communication method 1800 according to an embodiment of the present application.

Fig. 19 is a schematic flowchart of another communication method 1900 according to an embodiment of the present application.

Fig. 20 is a schematic flowchart of another model training method 2000 according to an embodiment of the present application.

Fig. 21 is a schematic structural diagram of a terminal device 2100 according to an embodiment of the present application.

Fig. 22 is a schematic structural diagram of a network device 2200 according to an embodiment of the present application.

Fig. 23 is a schematic structural diagram of a model training device 2300 according to an embodiment of the present application.

FIG. 24 is a schematic structural diagram of a communication device or a model training device 700 according to an embodiment of the present application;

FIG. 25 is a schematic structural diagram of a chip 800 according to an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.

It should be noted that the terms "first" and "second" in the description and claims of the embodiments of the present application and the above drawings are used to distinguish similar objects, and not necessarily used to describe a specific order or sequence order. The objects described by "first" and "second" described at the same time may be the same or different.

The technical solution of the embodiment of the present application can be applied to various communication systems, such as: Global System of Mobile communication (Global System of Mobile communication, GSM) system, code division multiple access (Code Division Multiple Access, CDMA) system, broadband code division multiple access (Wideband Code Division Multiple Access, WCDMA) system, General Packet Radio Service (GPRS), Long Term Evolution (LTE) system, Advanced long term evolution (LTE-A) system , New Radio (NR) system, evolution system of NR system, LTE (LTE-based access to unlicensed spectrum, LTE-U) system on unlicensed spectrum, NR (NR-based access to unlicensed spectrum) on unlicensed spectrum unlicensed spectrum (NR-U) system, Universal Mobile Telecommunication System (UMTS), Wireless Local Area Networks (WLAN), Wireless Fidelity (WiFi), next-generation communications (5th-Generation , 5G) system or other communication systems, etc.

Generally speaking, the number of connections supported by traditional communication systems is limited and easy to implement. However, with the development of communication technology, mobile communication systems will not only support traditional communication, but also support, for example, Device to Device (Device to Device, D2D) communication, machine to machine (Machine to Machine, M2M) communication, machine type communication (Machine Type Communication, MTC), and vehicle to vehicle (Vehicle to Vehicle, V2V) communication, etc., the embodiment of this application can also be applied to these communications system.

Optionally, the communication system in the embodiment of the present application may be applied to a carrier aggregation (Carrier Aggregation, CA) scenario, may also be applied to a dual connectivity (Dual Connectivity, DC) scenario, and may also be applied to an independent (Standalone, SA) deployment Web scene.

The embodiment of the present application does not limit the applied frequency spectrum. For example, the embodiments of the present application may be applied to licensed spectrum, and may also be applied to unlicensed spectrum.

Embodiments of the present application describe various embodiments in conjunction with network equipment and terminal equipment, wherein: terminal equipment may also be referred to as user equipment (User Equipment, UE), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent or user device, etc. The terminal device can be a station (STAION, ST) in the WLAN, a cellular phone, a cordless phone, a Session Initiation Protocol (Session Initiation Protocol, SIP) phone, a wireless local loop (Wireless Local Loop, WLL) station, a personal digital processing (Personal Digital Assistant, PDA) devices, handheld devices with wireless communication functions, computing devices or other processing devices connected to wireless modems, vehicle-mounted devices, wearable devices, and next-generation communication systems, such as terminal devices in NR networks or Terminal equipment in the future evolution of the Public Land Mobile Network (PLMN) network.

As an example but not a limitation, in this embodiment of the present application, the terminal device may also be a wearable device. Wearable devices can also be called wearable smart devices, which is a general term for the application of wearable technology to intelligently design daily wear and develop wearable devices, such as glasses, gloves, watches, clothing and shoes. A wearable device is a portable device that is worn directly on the body or integrated into the user's clothing or accessories. Wearable devices are not only a hardware device, but also achieve powerful functions through software support, data interaction, and cloud interaction. Generalized wearable smart devices include full-featured, large-sized, complete or partial functions without relying on smart phones, such as smart watches or smart glasses, etc., and only focus on a certain type of application functions, and need to cooperate with other devices such as smart phones Use, such as various smart bracelets and smart jewelry for physical sign monitoring.

The network device can be a device used to communicate with mobile devices, and the network device can be an access point (Access Point, AP) in WLAN, a base station (Base Transceiver Station, BTS) in GSM or CDMA, or a base station (BTS) in WCDMA. A base station (NodeB, NB), can also be an evolved base station (Evolutional Node B, eNB or eNodeB) in LTE, or a relay station or access point, or a vehicle device, a wearable device, and a network device (gNB) in an NR network Or a network device in a future evolved PLMN network, etc.

In this embodiment of the present application, the network device provides services for the cell, and the terminal device communicates with the network device through the transmission resources (for example, frequency domain resources, or spectrum resources) used by the cell. The cell may be a network device (for example, The cell corresponding to the base station) can belong to the macro base station or the base station corresponding to the small cell (Small cell). The small cell here can include: Metro cell, Micro cell, Pico cell cell), Femto cell, etc. These small cells have the characteristics of small coverage and low transmission power, and are suitable for providing high-speed data transmission services.

Fig. 1 exemplarily shows one network device 110 and two terminal devices 120, optionally, the wireless communication system 100 may include multiple network devices 110, and the coverage of each network device 110 may include other numbers The terminal device 120, which is not limited in this embodiment of the present application. The embodiment of the present application may be applied to a terminal device 120 and a network device 110 , and may also be applied to a terminal device 120 and another terminal device 120 .

Optionally, the wireless communication system 100 may also include other network entities such as a mobility management entity (Mobility Management Entity, MME), an access and mobility management function (Access and Mobility Management Function, AMF). This is not limited.

It should be understood that the terms "system" and "network" are often used interchangeably herein. The term "and/or" in this article is just an association relationship describing associated objects, which means that there can be three relationships, for example, A and/or B can mean: A exists alone, A and B exist simultaneously, and there exists alone B these three situations. In addition, the character "/" in this article generally indicates that the contextual objects are an "or" relationship.

It should be understood that the "indication" mentioned in the embodiments of the present application may be a direct indication, may also be an indirect indication, and may also mean that there is an association relationship. For example, A indicates B, which can mean that A directly indicates B, for example, B can be obtained through A; it can also indicate that A indirectly indicates B, for example, A indicates C, and B can be obtained through C; it can also indicate that there is an association between A and B relation.

In the description of the embodiments of the present application, the term "corresponding" may indicate that there is a direct or indirect correspondence between the two, or that there is an association between the two, or that it indicates and is indicated, configuration and is configuration etc.

In order to facilitate the understanding of the technical solutions of the embodiments of the present application, the related technologies of the embodiments of the present application are described below. The following related technologies can be combined with the technical solutions of the embodiments of the present application as optional solutions, and all of them belong to the embodiments of the present application. protected range. It should be understood that the basic processes and basic concepts described below do not limit the embodiments of the present application.

In the current wireless communication system, there are many designs about reference signals, such as downlink reference signals, including downlink demodulation reference signal (DMRS, Demodulation Reference Signal), channel state information reference signal (CSI-RS, Channel State Information Reference Signal ), downlink phase tracking reference signal (PT-RS, Phase Tracking Reference Signal), positioning reference signal (PRS, Positioning Reference Signal), etc., for uplink reference signal, including sounding reference signal (SRS, Sounding Reference Signal), uplink DMRS, Uplink PT-RS, etc. The design of these reference signals is mainly used to complete different tasks, such as channel estimation, phase tracking, positioning and so on.

In recent years, artificial intelligence research represented by neural networks has achieved great results in many fields, and it will also play an important role in people's production and life for a long time to come.

The basic structure of a simple neural network includes: input layer, hidden layer and output layer, as shown in Figure 2. The input layer is responsible for receiving data, the hidden layer processes the data, and the final result is generated in the output layer. Among them, each node represents a processing unit, which can be regarded as simulating a neuron. Multiple neurons form a layer of neural network, and multiple layers of information transmission and processing construct an overall neural network.

With the continuous development of neural network research, neural network deep learning algorithms have been proposed in recent years, more hidden layers have been introduced, and feature learning is performed through layer-by-layer training of neural networks with multiple hidden layers, which greatly improves the learning of neural networks. And processing capabilities, and are widely used in pattern recognition, signal processing, optimization combination, anomaly detection, etc.

The combination of neural network and wireless communication system is a current research direction. At present, there are many works on the application of neural network to wireless communication problems such as channel estimation, phase tracking, positioning, and beam management. However, the source of these works relies on The information is also the existing various reference signals.

For example, in the CSI feedback problem, as shown in FIG. 3A , the CSI information acquired and processed through the current CSI-RS reference signal will be encoded and decoded by AI to recover the CSI information at the base station side. In the channel estimation problem, as shown in FIG. 3B , based on the current CSI-RS or DMRS, the UE can realize high-performance estimation of a given channel through an AI channel estimator. In the positioning problem, as shown in Figure 3C, the UE can obtain the corresponding channel information through the current PRS, and then rely on the positioning channel information through the AI-based positioning algorithm to obtain high-precision positioning results.

For wireless communication systems, current reference signals are not designed with future use in wireless communication solutions based on AI, neural network approaches. When these existing reference signals are directly applied to various wireless communication solutions based on AI and neural network methods, these existing reference signals are only borrowed, and the wireless communication based on the existing reference signals is given. system solutions.

It is doubtful whether the above-mentioned existing reference signals are the best matching relationship with the wireless communication solutions based on AI and neural networks, because it is difficult to achieve the reference signal design and wireless communication solutions here when they are independently designed. Best matching result.

In addition, from another point of view, the design of the existing reference signal considers the reference signal design of common scenarios, rather than specific scenarios. For wireless communication solutions based on AI and neural networks, considering the dependence on environment and scene-related data in the process of technology construction, such solutions are often based on scene optimization. In this case, for the problem of scene optimization and scene adaptability, the original intention of the design of the existing reference signal is inconsistent with the original intention of the design of the AI-based wireless communication solution.

In summary, this solution believes that it is necessary to build reference signal design and function matching in wireless communication systems based on AI technology, and the above reference signal design needs to take into account the optimization of AI performance and the requirements of multi-user applications.

The embodiment of the present application proposes a communication method. FIG. 4 is a schematic flowchart of a communication method 400 according to the embodiment of the present application. The method can optionally be applied to the system shown in FIG. 1 , but is not limited thereto . The method includes at least some of the following.

S410: The terminal device receives a first signal, where the first signal is generated by a first model;

S420: The terminal device processes the first signal by using the second model to obtain first information, where the first information includes channel information;

Optionally, the foregoing first signal may include a reference signal. The aforementioned reference signal is an uplink reference signal or a downlink reference signal, for example, a downlink DMRS, CSI-RS, downlink PT-RS or PRS signal.

In step S410, the terminal device may receive a first signal from the network device, and the first signal may be generated by the network device using the first model.

Optionally, the above-mentioned second model may include a channel estimation sub-model;

The channel estimation sub-model is used to perform channel estimation based on a first signal (such as a reference signal) to obtain channel information.

Wherein, the above channel information may be used to characterize channel quality, channel state or channel estimation result obtained by performing channel estimation based on the first information.

Wherein, the above-mentioned network device may be an access network device serving the terminal device (such as a base station, eNB or gNB), or may be an access network device communicating with the terminal device such as a base station, eNB or gNB).

The embodiment of the present application proposes another communication method. FIG. 5 is a schematic flowchart of another communication method 500 according to the embodiment of the present application. This method can optionally be applied to the system shown in FIG. 1 , but not only limited to this. As shown in Figure 5, after the above step S420, it may further include:

S530: The terminal device processes the first information by using a third model to obtain second information;

S540: The terminal device sends the second information, where the second information is used for processing by the fourth model to obtain third information;

Wherein, the above-mentioned first model, second model, third model and fourth model are obtained through joint training.

Optionally, the above-mentioned third model includes a compressible sub-model; the compression sub-model is used to compress the first information (such as channel information) to obtain the compressed information of the first information; correspondingly, the above-mentioned second information includes the first information compressed information.

Correspondingly, the above-mentioned fourth model may include a restoration sub-model; the restoration sub-model is used to restore the compressed information of the above-mentioned first information to obtain the restoration information of the first information; correspondingly, the above-mentioned third information includes the Recovery information.

Using the above model, in step S530, the terminal device can use the compressed sub-model to compress the first information (such as channel information) to obtain second information (such as compressed information of channel information); in step S540, the terminal device can compress the The compressed information of the channel information is sent to the network device, where the network device may be the network device that sent the first signal (such as a reference signal) before; after that, the network device may restore the compressed information of the channel information by using the recovery submodel, Get the third information. The third model and the fourth model above constitute a channel information feedback model.

Alternatively, the above-mentioned third model includes a generation sub-model and a compression sub-model; wherein, the generation sub-model is used to perform feature transformation on the first information (such as channel information) to obtain a first feature vector corresponding to the first information; the compression sub-model is used to The first feature vector is compressed to obtain compressed information of the first feature vector; the second information includes the compressed information of the first feature vector.

Correspondingly, the above-mentioned fourth model includes a restoration sub-model; the restoration sub-model is used to restore the compressed information of the above-mentioned first feature vector to obtain the restoration information of the first feature vector; the above-mentioned third information includes the restoration information of the first feature vector .

Using the above model, in step S530, the terminal device can use the generation sub-model in the third model to generate the first feature vector of the first information, and then use the compression sub-model in the third model to compress the first feature vector , to obtain the second information (such as the compressed information of the first feature vector); in step S540, the terminal device can send the compressed information of the first feature vector to the network device, wherein the network device can send the first signal ( Such as the network device of the reference signal); afterward, the network device can recover the compressed information of the first feature vector by using the restoration sub-model to obtain the third information. The third model and the fourth model above constitute a channel information feedback model.

The first model can also be called a signal generation sub-model, which is used to generate multi-user reference signals, such as generating a reference signal set including multiple reference signals, and the reference signals in the reference signal set can be used by multiple terminal devices (such as belonging to terminal equipment in the same cell, or terminal equipment served by the same access equipment).

The second model may also be called a channel estimation sub-model, and is used to perform channel estimation based on a reference signal to obtain channel information.

The generation sub-model can also be called the channel information generation sub-model, which is used to process the received channel information by means of data transformation, etc., and obtain the channel feature vector of the channel information, for example, by using SVD (Singular Value Decomposition) method. Channel information feature vector.

The compression sub-model may also be called a channel information compression sub-model, which is used to compress the received channel information or channel information feature vector. The restoration sub-model may also be referred to as a channel information restoration sub-model, and is used to restore the received compressed information.

The above-mentioned signal generation sub-model, channel estimation sub-model, channel information generation sub-model, channel information compression sub-model or recovery sub-model can be composed of one of the fully connected network, convolutional neural network, residual network, and self-attention mechanism network. or multiple forms.

It can be seen that when the signal generation sub-model and channel estimation sub-model are jointly trained, the integrated design of multi-user reference signal and channel estimation in the wireless communication system based on AI is realized; when the signal generation sub-model, channel estimation sub-model model, channel information generation sub-model, channel information compression sub-model and restoration sub-model for joint training, or when the signal generation sub-model, channel estimation sub-model, channel information compression sub-model and restoration sub-model are jointly trained, the The integrated design of multi-user reference signal, channel estimation and channel information feedback in AI's wireless communication system.

The above training of the model can be completed by the terminal device or by the network device. Correspondingly, there are at least the following two modes of model training and transmission:

method one:

The training process is completed by a network device (such as a base station). The network device can generate all or part of the trained signal generation sub-model, channel estimation sub-model, channel information generation sub-model, channel information compression sub-model, and channel information recovery sub-model. sent to the terminal device. In addition, the network device can send the first channel model module and/or the second channel simulation module to the terminal device, the first channel model module and the second channel simulation module are used to simulate the influence of the downlink transmission channel and the uplink transmission channel on the signal respectively .

Alternatively, the network device may use the trained channel estimation sub-model, channel information generation sub-model and channel information compression sub-model as an overall channel estimation and channel information feedback module, such as sending it to the terminal device as a first coding model; or, the network The device may send the trained channel estimation sub-model and channel information compression sub-model to the terminal device as an overall channel estimation and channel information feedback module, such as a second coding model. Alternatively, the network device may further send the trained channel information recovery sub-model to the UE as a decoding module. Figure 6 shows the above model sending method.

The above-mentioned models, sub-models, and modules transmitted by the network device can be completed in an independent transmission, or in a non-independent transmission (for example, all the above-mentioned information is transmitted through one signaling or message).

Since one network device can serve multiple terminal devices, the network device can send the above-mentioned model, sub-model and/or module to all terminal devices (or at least some terminal devices) it serves. Taking network equipment as a base station and terminal equipment as a mobile terminal as an example, for example, base station A serves mobile terminal 1, mobile terminal 2, mobile terminal 3, and mobile terminal 4, and base station A can provide mobile terminal 1, mobile terminal 2, and mobile terminal The terminal 3 and the mobile terminal 4 transmit the above-mentioned models, sub-models and/or modules.

Fig. 7 is a structural schematic diagram of two models according to the present application. Referring to FIG. 7, the terminal device may receive the channel estimation sub-model, and use the channel estimation sub-model for channel estimation based on the received reference signal. Alternatively, the terminal device may receive the channel estimation sub-model and the channel information compression sub-model respectively, and use the sub-models for channel estimation and channel information compression respectively. Alternatively, the terminal device may receive the channel estimation sub-model, the channel information generation sub-model, and the channel information compression sub-model respectively, and use the sub-models for channel estimation, channel information feature vector generation, and channel information compression respectively.

Alternatively, the terminal device may respectively receive the signal generation sub-model, the first channel simulation module, and the channel estimation sub-model, and use these sub-models/modules to evaluate the performance of the signal generation sub-model and the channel estimation sub-model. For example, the terminal device uses the signal generation sub-model to generate a reference signal, uses the first channel simulation module to process the reference signal to simulate the reference signal received by the terminal device through the downlink channel; then uses the channel estimation sub-model to simulate The signal processed by the module performs channel estimation to obtain a channel estimation result; the channel estimation result is compared with the parameters of the first channel simulation module, based on the comparison result, and the reference signal quality and/or the first channel simulation module generated by the signal generation sub-model The quality of the reference signal obtained after processing by a channel simulation module evaluates the performance of the signal generation sub-model and the channel estimation sub-model. After passing the evaluation, the terminal device can use the channel estimation sub-model to perform channel estimation based on the actually received reference signal. In addition, the above-mentioned first channel simulation module may be pre-stored in the terminal device. In this case, the terminal device does not need to receive the first channel simulation module from the network device. Moreover, the above-mentioned first channel analog module can be simplified as a unit matrix. In this case, after the reference signal is processed by the first channel analog module, the obtained signal is the same as the reference signal; that is, the simulated reference signal is transmitted in the downlink There is no change during the process, and the reference signal received by the terminal device through the downlink channel is the same as the reference signal sent by the network device.

Or, the terminal device can respectively receive the signal generation submodel, the first channel simulation module, the channel estimation submodel, the channel information generation submodel, the channel information compression submodel, the second channel simulation module and the channel information restoration submodel, and use these submodels Model/module, the terminal device can evaluate the performance of the signal generation sub-model, channel estimation sub-model, channel information generation sub-model, channel information compression sub-model and channel information recovery sub-model. For example, the terminal device uses the signal generation sub-model to generate a reference signal, uses the first channel simulation module to process the reference signal to simulate the reference signal received by the terminal device through the downlink channel; then uses the channel estimation sub-model to simulate The signal processed by the module performs channel estimation to obtain channel estimation results (such as channel information); the channel information is processed by the channel information generation sub-model to obtain the vector characteristics of the channel information; the channel information compression sub-model is used to extract the channel information Compress the vector features to obtain the compressed channel information; use the second channel simulation module to process the compressed channel information to simulate the compressed channel information received by the network equipment through the uplink channel; use the channel information to restore the sub-model The processed result of the second channel simulation module is processed to simulate the channel information obtained after the network equipment recovers the channel information. After that, continue to compare the information output by the channel information restoration sub-model with the information input by the channel information compression sub-model, based on the comparison result, and the quality of the reference signal generated by the signal generation sub-model and/or obtained after processing by the first channel simulation module The quality of the reference signal (further based on the comparison result of the channel estimation result and the parameters of the first channel simulation module), the signal generation sub-model, the channel estimation sub-model, the channel information generation sub-model, the channel information compression sub-model and the channel The performance of the information recovery sub-model is evaluated. After the evaluation is qualified, the terminal device can use the channel estimation sub-model to perform channel evaluation based on the real received reference signal, and use the channel information generation sub-model and the channel information compression sub-model to compress the information obtained after evaluation, and compress the The channel information is sent to the network device. In addition, the above-mentioned first channel simulation module and/or the second channel simulation module may be pre-stored in the terminal device. In this case, the terminal device does not need to receive the first channel simulation module and/or the second channel simulation module from the network device. Moreover, the above-mentioned first channel simulation module can be simplified as a unit matrix. In this case, after the reference signal is processed by the first channel simulation module, the obtained signal is the same as the reference signal; that is, the reference signal is simulated in the downlink transmission process. There is no change in , and the reference signal received by the terminal device through the downlink channel is the same as the reference signal sent by the network device. Similarly, the above-mentioned second channel simulation module can also be simplified as a unit matrix. In this case, after the compressed channel information is processed by the second channel simulation module, the obtained result is the same as the compressed channel information; It is simulated that the compressed channel information does not change during the uplink transmission, and the compressed channel information received by the network device through the uplink channel is the same as the compressed channel information sent by the terminal device.

In addition, the above-mentioned channel information generation sub-model is an optional model. If there is no channel information generation sub-model, the channel information compression sub-model directly compresses the channel information generated by the channel estimation sub-model to obtain compressed channel information. In addition, the processing methods of other models/modules in the process of receiving, evaluating, and using are consistent with the above content, and will not be repeated here.

If the overall evaluation result of the terminal device on the received model/sub-model is poor (for example, the quality of the reference signal is low, or the accuracy of channel estimation is low, etc.), the received model/sub-model may not be used, but The terminal device can re-train the models/sub-models jointly to update the model parameters of these models/sub-models, or the terminal device can train itself to obtain a new model/sub-model. After the terminal device obtains a new model/sub-model after joint training or updating, it can also synchronize its model/sub-model to the network device; correspondingly, after the network device receives the new model/sub-model, it can Replace the model/sub-model originally trained by itself, and also synchronize the new model/sub-model to other terminal devices. Regarding other subsequent related processes, this embodiment will not list them one by one. Through the above processing, it can be ensured that the model/sub-model with the best performance is used in the entire communication system, thereby further improving the overall performance of the entire system.

The method for the terminal device to receive the sub-model/module may be through one or more of the following methods: downlink control signaling, media access control (MAC, Medica Access Control) control element (CE, Control Element) message, wireless Resource control (RRC, Radio Resource Control) messages, broadcast, downlink data transmission, downlink data transmission for artificial intelligence services or neural network transmission requirements.

In some implementation manners, the present application further includes: the terminal device receives the foregoing second model.

Optionally, the terminal device may also receive the foregoing first model.

In some implementation manners, the present application further includes: the terminal device receives the foregoing second model and the third model.

Optionally, the terminal device may also receive the foregoing first model and fourth model.

The first model, the second model, the third model, the fourth model, the sub-model in the first model, the sub-model in the second model, the sub-model in the third model or the first model The sub-models in the four models are carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, and downlink data transmission for artificial intelligence business transmission requirements.

Optionally, the foregoing method may further include: the terminal device receiving a first coding model, where the first coding model includes a channel estimation sub-model, a generation sub-model, and a compression sub-model; wherein,

The channel estimation sub-model constitutes the above-mentioned second model;

The generation sub-model and the compression sub-model constitute the third model described above.

Correspondingly, the above terminal device uses the second model to process the first signal to obtain the first information, and uses the third model to process the first information to obtain the second information. The process may be combined into one step, including: the terminal The device processes the first signal by using the first coding model to obtain the second information.

In some embodiments, the first coding model is carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, downlink data transmission for artificial intelligence service transmission requirements.

Optionally, the foregoing method may further include: the terminal device receiving a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

The channel estimation sub-model constitutes the above-mentioned second model;

The compressed sub-models constitute the third model described above.

Correspondingly, the above terminal device uses the second model to process the first signal to obtain the first information, and uses the third model to process the first information to obtain the second information. The process may be combined into one step, including: the terminal The device uses the second encoding model to process the first signal to obtain second information.

In some embodiments, the second coding model is carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, downlink data transmission for artificial intelligence service transmission requirements.

Method 2:

The training process is completed by the terminal device (such as UE), and the terminal device can use all or part of the trained signal generation sub-model, channel estimation sub-model, channel information generation sub-model, channel information compression sub-model, and channel information recovery sub-model sent to network devices. In addition, the terminal device may send the first channel model module and/or the second channel simulation module to the network device, and the first channel model module and the second channel simulation module are respectively used to simulate the influence of the downlink transmission channel and the uplink transmission channel on the signal .

Alternatively, the terminal device can use the trained channel estimation sub-model, channel information generation sub-model and channel information compression sub-model as an overall channel estimation and channel information feedback module, such as sending it to the network device as the first coding model, and the channel information recovery The sub-model can be used as the first decoding model; or, the terminal device can use the trained channel estimation sub-model and channel information compression sub-model as an overall channel estimation and channel information feedback module, such as sending it to the network device as a second coding model , the channel information recovery sub-model can be used as the second decoding model. Figure 8 shows the above model sending method.

The above-mentioned models, sub-models, and modules transmitted by the terminal device can be completed in independent transmission, or in non-independent transmission (for example, all the above-mentioned information is transmitted through one signaling or message).

The terminal device may send a signal generation sub-model for the network device to use the signal generation sub-model to generate a reference signal. Alternatively, the terminal device may send a signal generation sub-model and a channel information recovery sub-model for the network device to use the signal generation sub-model to generate a reference signal, and use the channel information recovery sub-model to process the compressed channel information sent by the terminal device to recover.

Alternatively, the terminal device may send the signal generation sub-model, the first channel simulation module, and the channel estimation sub-model respectively, so that the network device may use these sub-models/modules to evaluate the performance of the signal generation sub-model and the channel estimation sub-model. The method of evaluation is the same as the evaluation method of the above-mentioned terminal equipment, and will not be repeated here. After passing the evaluation, the network device can use the signal generation sub-model to generate a reference signal, and send the reference signal to the terminal device for the terminal device to perform channel estimation based on the received reference signal. In addition, the above-mentioned first channel simulation module may be pre-stored in the network device. In this case, the network device does not need to receive the first channel simulation module from the terminal device. Moreover, the above-mentioned first channel analog module can be simplified as a unit matrix. In this case, after the reference signal is processed by the first channel analog module, the obtained signal is the same as the reference signal; that is, the simulated reference signal is transmitted in the downlink There is no change during the process, and the reference signal received by the terminal device through the downlink channel is the same as the reference signal sent by the network device.

Alternatively, the terminal device may respectively send the signal generation submodel, the first channel simulation module, the channel estimation submodel, the channel information generation submodel, the channel information compression submodel, the second channel simulation module, and the channel information restoration submodel for the The network devices use these submodels/modules to evaluate the performance of the signal generation submodel, channel estimation submodel, channel information generation submodel, channel information compression submodel and channel information recovery submodel. The evaluation method is the same as the evaluation method of the above-mentioned terminal equipment, and will not be repeated here. After the evaluation is qualified, the network device can use the signal generation sub-model to generate a reference signal, and send the reference signal to the terminal device for the terminal device to perform channel estimation based on the received reference signal; and use the channel information recovery sub-model to The received compressed channel information is recovered. In addition, the first channel simulation module and/or the second channel simulation module may be stored in the network device in advance, and in this case, the network device does not need to receive the first channel simulation module and/or the second channel simulation module from the terminal device. Moreover, the above-mentioned first channel simulation module can be simplified as a unit matrix. In this case, after the reference signal is processed by the first channel simulation module, the obtained signal is the same as the reference signal; that is, the reference signal is simulated in the downlink transmission process. There is no change in , and the reference signal received by the terminal device through the downlink channel is the same as the reference signal sent by the network device. Similarly, the above-mentioned second channel simulation module can also be simplified as a unit matrix. In this case, after the compressed channel information is processed by the second channel simulation module, the obtained result is the same as the compressed channel information; It is simulated that the compressed channel information does not change during the uplink transmission, and the compressed channel information received by the network device through the uplink channel is the same as the compressed channel information sent by the terminal device.

The method for the above-mentioned terminal equipment to send the sub-model/module can be through one or more of the following methods: downlink control signaling, MAC CE message, RRC message, broadcast, downlink data transmission, for artificial intelligence services or neural network services Downlink data transmission for transmission requirements.

Since one network device can serve multiple terminal devices, the network device can receive the above-mentioned model, sub-model and/or module sent by one of the terminal devices it serves, and use the received model, sub-model and/or module Reference signal generation and channel feedback process for all or part of the terminal equipment it serves. Taking network equipment as a base station and terminal equipment as a mobile terminal as an example, for example, base station A serves mobile terminal 1, mobile terminal 2, mobile terminal 3 and mobile terminal 4, and mobile terminal 1 sends the above model, submodel and/or or modules, the base station A uses the above models, sub-models and/or modules for all or part of the mobile terminals it serves (such as one or more of the mobile terminal 1, the mobile terminal 2, the mobile terminal 3 and the mobile terminal 4) Reference signal generation and channel feedback process.

Alternatively, the network device may receive the above-mentioned model, sub-model and/or module sent by each terminal device it serves, and use the received model, sub-model and/or module for reference signal generation and channel feedback of the corresponding terminal device process. Take the network device as a base station and the terminal device as a mobile terminal as an example. For example, base station A serves mobile terminal 1, mobile terminal 2, mobile terminal 3 and mobile terminal 4, and mobile terminal 1, mobile terminal 2, mobile terminal 3 and mobile terminal 4. Send the above-mentioned models, sub-models and/or modules trained by itself to the base station A respectively, and the base station uses the models, sub-models and/or modules received from each mobile terminal to perform the reference signal generation and channel feedback process of the corresponding terminal equipment.

Alternatively, the network device may receive the above-mentioned models, sub-models and/or modules sent by at least two terminal devices served by it, and combine or optimize the received models, sub-models and/or modules respectively to obtain a new model , sub-model and/or module, and use the new model, sub-model and/or module for the reference signal generation and channel feedback process of all or part of the terminal equipment served by it. Take the network device as a base station and the terminal device as a mobile terminal as an example. For example, base station A serves mobile terminal 1, mobile terminal 2, mobile terminal 3, and mobile terminal 4. Mobile terminal 1 and mobile terminal 2 send their own training to base station A respectively. The above models, sub-models and/or modules, the base station combines or optimizes the received models, sub-models and/or modules to obtain new models, sub-models and/or modules, and combines the new models, sub-models and The /or module is used for the reference signal generation and channel feedback process of all or part of the mobile terminals served by it (such as one or more of the mobile terminal 1, the mobile terminal 2, the mobile terminal 3 and the mobile terminal 4).

In some implementation manners, the present application further includes: the terminal device sends the first model. Optionally, the terminal device may also send the second model.

In some implementation manners, the present application further includes: the terminal device sends the first model and the fourth model. Optionally, the terminal device may also send the second model and/or the third model.

Optionally, the above-mentioned first model, the second model, the third model, the fourth model, the sub-model in the first model, the sub-model in the second model, the second model The sub-models in the three models or the sub-models in the fourth model are carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, uplink for artificial intelligence service transmission requirements data transmission.

Optionally, the foregoing method may further include: the terminal device sending a first coding model, where the first coding model includes a channel estimation sub-model, a generation sub-model, and a compression sub-model; wherein,

the channel estimation sub-model constitutes a second model;

The generating sub-model and the compressing sub-model constitute a third model.

In some embodiments, the first coding model is carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, and uplink data transmission for artificial intelligence service transmission requirements.

Optionally, the foregoing method may further include: the terminal device sending a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

the channel estimation sub-model constitutes a second model;

The compressed sub-models constitute the third model.

In some embodiments, the second coding model is carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, and uplink data transmission for artificial intelligence service transmission requirements.

The above introduces the two training subjects of the model, as well as the transmission, evaluation and usage of the model/sub-model in the case of different subjects for model training. In this embodiment of the present application, other devices can also be used for model training, and the trained models are sent to the terminal device and the network device respectively. The above-mentioned models may be transmitted through a wired connection or a wireless connection. For example, the second first model (or the model parameters of the second model) is transmitted to the terminal device through a wired connection with the terminal device, or the second first model is transmitted to the terminal device through other wireless connections with the terminal device. The model (or the model parameters of the second model) is transmitted to the terminal device. Wherein, the above-mentioned wireless connection method may be bluetooth or wireless fidelity (Wi-Fi, Wireless Fidelity) connection method or the like.

The above-mentioned embodiment is introduced by taking the channel estimation of the downlink channel as an example. The embodiment of the present application is also applicable to generating the uplink reference signal and evaluating the uplink channel. The specific method corresponds to the above-mentioned embodiment and will not be repeated here. repeat.

In the embodiment of the present application, there are at least two specific ways for the terminal device to perform model training:

Example 1, an AI-based multi-user reference signal and channel estimation integrated design scheme for a wireless communication system:

Example 2: An AI-based integrated design scheme for multi-user reference signals, channel estimation, and channel information feedback in a wireless communication system.

The above two examples are introduced respectively below.

Example one:

Optionally, the terminal device may use the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the second model. Wherein, the first model can be obtained after the first initial model is jointly trained, and the second model can be obtained after the second initial model is jointly trained.

Specifically, the terminal device may input the input information into the first initial model to obtain a first set output by the first initial model, where the first set includes a plurality of first reference signals;

inputting any one of the first reference signals in the first set to the first channel simulation module to obtain a second reference signal;

inputting the second reference signal into the second initial model to obtain channel information;

determining a first loss function based on at least one of the first set, the second reference signal, the channel information, and parameters in the first channel simulation module;

updating the first initial model and the second initial model according to the first loss function.

In some implementations, determining the first loss function may include:

Determine the channel information and the parameters of the first channel simulation module based on at least one of the first set, the second reference signal, the channel information and the parameters of the first channel simulation module degree of variance and/or reference signal quality;

The first loss function is determined based on the degree of difference between the channel information and the parameters of the first channel simulation module and/or the quality of a reference signal.

The above manner of determining the quality of the reference signal will be described in detail in subsequent implementation manners.

FIG. 9 is a schematic diagram of an AI-based multi-user reference signal and channel estimation integrated design scheme for a wireless communication system according to an embodiment of the present application. In Fig. 9, the above-mentioned first model (before the training is completed, the first model is the first initial model) is specifically a signal generation sub-model, and the above-mentioned second model (before the training is completed, the second model is the second initial model) Specifically, it is a channel estimation sub-model. The first channel simulation module may not participate in the training. For example, the parameters of the first channel simulation module are fixed, and are used to simulate the reference signal received by the terminal device after the reference signal is transmitted through the channel. The first channel simulation module can be stored in the terminal device and the network device respectively in advance, or sent to the terminal device by the network device or other devices before the joint training.

In Figure 9, although the signal generation sub-model and the channel estimation sub-model are also distinguished, the above-mentioned models are not independently generated, but are realized through a jointly designed and trained neural network. Supervision in an all-in-one solution. Specifically, firstly, this scheme will give the input, output, loss function, and model structure information of the multi-user reference signal design. At the same time, this scheme will give the input, output, and Loss function, model structure information.

The above input information includes at least one of the following: no input, noise, random numbers, sequences in the preset sequence set, channel type indication information, channel data sample information, wireless channel or scene related information.

Specifically, the following types of input:

(a) No input: There is no independent input information.

(b) Noise: The input can be noise, which can come from the real environment or artificially generated.

(c) Random number: The input can be a sequence of random numbers or a sequence of pseudo-random numbers.

(d) Sequence: The input can be a sequence in a given sequence set, where the sequence set can be one or more of sequence sets such as m sequence set, golden sequence set, zc sequence set, etc.

(e) Channel type indication information: The input of the above joint scheme may also include channel type indication information, such as frequency information, environment information, and scene information corresponding to the indicated channel, such as: high frequency, low frequency, indoor, outdoor, densely populated cells, Open field, IoT scene, industrial scene, etc.

(f) Channel data sample information: The input of the above joint scheme may also include channel data samples.

For (b) (c) and (d) when using noise, pseudo-random numbers or predefined sequences as the input of the multi-user signal joint construction scheme, the format of the above-mentioned noise, random numbers, and predefined sequences can be One-dimensional vectors, or two-dimensional matrices, or high-dimensional noise, random numbers, or predefined sequence sets. The format of the above noise, pseudo-random number, and predefined sequence can be agreed in advance through agreement or signaling. The format of the noise, the pseudo-random number, and the predefined sequence may be consistent with the expected generated reference signal sequence format.

For (b) (c) and (d) when using noise, pseudo-random numbers or predefined sequences as the input of the multi-user signal joint construction scheme, different noises, pseudo-random numbers or predefined sequence sequences are input It constitutes the basic variables for constructing different reference signals. The diversity of these different input sequences will form the diversity of the output of the signal generation sub-model, and the reference signals that are diverse and satisfy the subsequent loss function constraints constitute the scenario- and task-oriented A collection of multi-user reference signals.

For the channel type indication information and channel data sample information described in (e) and (f), it can be directly used as the input of the signal generation sub-model, or can be used as one or more items in the first channel simulation module and the channel estimation sub-model input of.

In addition, the input can also include other information related to the wireless channel or scene, such as channel signal-to-noise ratio, signal-to-interference-noise ratio, channel type, bandwidth information, delay information, etc., which can be used as one or more of the above sub-models item input.

During the joint training process, whether to input one or more types of the above information may be determined according to actual conditions or actual scenarios, and is not limited here.

The output of the proposed multi-user reference signal design for this example includes the following:

The first item, the set of reference signals:

For the signal generation sub-model, its input is the input described in the previous section, and its output is a reference signal set, which includes multiple reference signals. For example, the input of the signal generation sub-model can be a set of random numbers or a set of given sequences, and the output of the signal generation sub-model can be a set of corresponding sequences output by the set of random numbers or sequences after the signal generation sub-model, This group of output sequences is the output reference signal set, and the reference signal set may include multiple reference signals. Each reference signal can be applied to different UEs.

The second item, channel information:

The output of the channel estimation sub-model described above may include channel information. The channel information may be complete channel information, such as time-domain channel information or frequency-domain channel information.

Optionally, channel information may be distributed in the first dimension and/or the second dimension.

Or, the channel information may be distributed in at least one of the first dimension, the second dimension and the third dimension.

Specifically, a single sample of channel information can be composed of a matrix of size M*N, which has M first granularities in the first dimension and N second granularities in the second dimension, and M and N can be equal They can also be unequal, and the specific numerical indications in the matrix represent the channel quality. Wherein, the channel quality may be represented by a signal strength value; the unit of the signal strength value may be dBm; or, the signal strength value may have no unit, but be represented by a numerical value obtained after normalization. In addition, the two-dimensional data of M*N can also be synthesized into one-dimensional data of size 1*(M*N) or (M*N)*1. The specific transformation can be the first dimension and then the second dimension. It can also be the second dimension first and then the first dimension. This transformation is the difference in the form of expression.

In some implementations, the above-mentioned first dimension may be a frequency-domain dimension, and the channel information includes data distributed on M1 (M1 is a value of the above-mentioned M) frequency-domain granularities of the frequency-domain dimension; wherein M1 is a positive integer.

Optionally, the foregoing frequency domain granularity includes a RBs and/or b subcarriers, where a or b is a positive integer.

Specifically, a single sample of channel information can be distributed on a first dimension with M1 granularities (for example, denoted as m), the first dimension can be a frequency domain dimension, and when the first dimension is a frequency domain dimension, the granularity m can be a RB (a is greater than or equal to 1, such as 2RB, 4RB, 8RB), or b subcarriers (b is greater than 1, such as 4 subcarriers, 6 subcarriers, or 18 subcarriers). When the first dimension is the frequency domain dimension, the frequency domain range indicated by a single sample of channel information is the frequency domain range of M1*m; for example, if the granularity m is 4RB, the frequency domain range indicated by a single sample of channel information Then it is M1*4RB.

In some implementations, the above-mentioned first dimension may be a time-domain dimension, and the channel information includes M2 (M2 is a value of the above-mentioned M, and M2 may be the same as or different from the above-mentioned M1) in the time-domain dimension. Data distributed on the delay granularity; where M2 is a positive integer.

Optionally, the above delay granularity includes at least one of the following: p1 microseconds, p2 symbol length, p3 symbol number of sampling points, where p1, p2 or p3 is a positive integer. Wherein, the above symbols may include Orthogonal Frequency Division Multiplexing (OFDM, Orthogonal Frequency Division Multiplexing) symbols.

Specifically, a single sample of channel information may be distributed on a first dimension with M2 granularities (for example, denoted as p), the first dimension may be a time-domain dimension, and when the first dimension is a time-domain dimension, the granularity p may be Delay granularity, for example, a delay granularity is the number of sampling points of p1 microseconds, or p2 symbol length, or p3 symbols. When the first dimension is the time-domain dimension, the time-domain range (or delay range) indicated by a single sample of the training set is the time-domain range of M2*p; for example, if the granularity p is 8 symbols in length, the channel The time domain range indicated by a single sample of information is M2*8 symbol lengths.

In some embodiments, the above-mentioned second dimension may be a spatial domain dimension.

Optionally, the foregoing spatial domain dimension may be an antenna dimension, and the channel information includes data distributed on N1 first granularities of the antenna dimension, where N1 is a positive integer.

Wherein, the foregoing first granularity may include a pair of transmitting and receiving antennas.

Optionally, the foregoing space domain dimension may be an angle domain dimension, and the channel information includes data distributed on N2 second granularities of the angle domain dimension, where N2 is a positive integer.

Wherein, the above-mentioned second granularity includes angular intervals. Such as the angular interval of the transmitting and receiving antennas, and/or the receiving angular interval of the channel information.

Specifically, a single sample of channel information may be distributed on a second dimension with N1 granularities (for example, denoted as n), and the second dimension may be a space domain dimension, specifically, an antenna dimension, for example, the second dimension is composed of N1 The second granularity is a pair of transmitting and receiving antennas.

In addition, a single sample of channel information can also be distributed on a second dimension with N2 granularities (for example, denoted as q), the second dimension can be a space domain dimension, specifically an angle domain dimension, for example, the second dimension is composed of It consists of N2 angles, and the second granularity is the angle interval between the above N angles.

The data on a specific combination of dimensions in a single sample of channel information indicates the channel quality indication situation under the specific combination of dimensions. For example, FIG. 10 is a schematic diagram of a channel information structure according to the present application. FIG. 10 shows an M*N matrix structure in which the first dimension is the frequency domain dimension and the second dimension is the space dimension. The indication value X on the 6th column can be used to represent the 3rd frequency domain granularity (the frequency domain as shown in Figure 10 The channel quality situation on the granularity is 2RB). As another example, FIG. 11 is a schematic diagram of another channel information structure according to the present application. FIG. 11 shows an M*N matrix structure in which the first dimension is the time domain dimension and the second dimension is the space dimension. The indication value Y on the fifth column of the row can be used to represent the channel quality situation on the fourth delay granularity on the fifth spatial granularity (as shown in FIG. 11 , the spatial granularity is 1 angle of arrival). In FIG. 10 and FIG. 11 , K represents the number of channel information, and K is a positive integer.

In some implementation manners, the above channel information includes S groups of feature sequences with a length of U, where S or U is a positive integer.

Optionally, the above S can be 2, 4 or 8.

Optionally, the above U may be 16, 32, 48, 64, 128 or 256.

Specifically, the output information of the above-mentioned channel estimation sub-model may be the channel characteristic information obtained by mathematical transformation of the above-mentioned original channel information, for example, the channel characteristic vector information obtained by decomposing the Singular Value Decomposition (SVD, Singular Value Decomposition), can be is the channel eigenvector information decomposed into a single stream, or the channel eigenvector information decomposed into multiple streams. For example, the output information of the above-mentioned channel estimation sub-model is S stream feature vectors, and each stream is composed of a feature sequence of length U. For example, it may be 2 streams, 4 streams or 8 streams of channel feature vector information, and each stream is composed of 16, 32, 48, 64, 128 or 256 length feature sequences. FIG. 12 is a schematic structural diagram of channel feature vector information according to the present application. In the example of FIG. 12 , the output of the channel estimation sub-model is 4-stream feature vectors, and each feature vector is composed of feature sequences with a length of 32.

It should be noted that because the channel information output by the signal generation sub-model and the channel information output after being processed by the first channel simulation module can be presented in the form of complex numbers, therefore, the channel information can be additionally based on the content described above One more dimension is caused by the independent presentation of the imaginary part and real part data of the channel information output by the signal generation sub-model (or the channel information output after being processed by the first channel simulation module). For example, in addition to the first dimension and the second dimension mentioned above, there may also be a third dimension, and the third dimension comes from the real part and the imaginary part of the channel information.

That is, the third dimension includes a complex dimension, and the complex dimension includes 2 elements, which are respectively used to carry the real part and the imaginary part of the data included in the channel information.

In addition, it should be noted that the output of the above-mentioned channel information may also be split and combined on the basis of the above-mentioned first dimension, second dimension, and third dimension. For example, the channel information is distributed in a T-dimensional matrix, the T-dimensional matrix is a matrix formed after splitting and/or combining at least one of the above-mentioned first dimension, second dimension and third dimension, and the T is positive integer.

For example, when the second dimension is the antenna pair dimension, it can also be split into a transmitting antenna sub-dimension and a receiving antenna sub-dimension, so as to expand the dimension of the above-mentioned virtual channel output form. , this embodiment no longer exhaustively enumerates the various possible dimensions after splitting

In the subsequent description, for the sake of simplicity, the two-dimensional channel information composed of the first dimension and the second dimension is used as an example, but it should be clarified that the dimension of the channel information is not limited to two dimensions.

The above describes the second output of the multi-user reference signal design scheme, namely channel information. The multi-user reference signal design scheme can also have the output of the entire joint scheme, as follows:

The third term, the output of the entire joint scheme:

The output of the entire joint training scheme includes a trained signal generation sub-model, and/or a channel estimation sub-model. It may also include a first channel simulation module, which may be pre-set and used to simulate changes in the reference signal passing through a wireless channel, and the first channel simulation module may not participate in model training; or, the first channel simulation module may not participate in model training; The channel simulation module can also be obtained through joint training.

The input of the channel estimation sub-model is the output of the reference signal sequence output by the signal generation sub-model after passing through the first channel simulation module. The first channel simulation module is used to simulate the changes of the reference signal passing through the wireless channel. For example, the first channel simulation module can be composed of a fully connected network, and the weight of the fully connected network is the channel matrix H of the wireless channel, that is to say, the signal generation After the reference signal S output by the module passes through the first channel simulation module, the reference signal result of simulating S passing through the channel H is obtained, such as S'=H*S.

Each of the above-mentioned models, sub-models or modules may adopt a neural network structure, such as one or more network structures of a fully connected network, a convolutional neural network, a residual network, and a self-attention mechanism network. Fig. 13 is a schematic diagram of a neural network structure according to the present application. As shown in Fig. 13 , a multi-user reference signal design scheme according to an embodiment of the present application includes a signal generation sub-model, a first channel simulation module and a channel estimation sub-model, Each submodel/model includes one or more fully connected layers. The input information can be a random number or an original sequence with a length of 64. The input information input signal generation sub-model; the signal generation sub-model generates a reference signal set including multiple reference signals, and the reference signal output by the signal generation sub-model is used as the first channel simulation The input of the module; the first channel simulation module outputs the result after processing the received reference signal, and inputs the result into the channel estimation sub-model; the channel estimation sub-model performs channel estimation based on the received data to obtain the final channel estimation As a result, for example, the size of the channel estimation result can be 8192, and can be transformed into a three-dimensional matrix form of [128,32,2].

The above introduces several output contents of the multi-user reference signal design scheme proposed in this example. In the subsequent use of the model, the terminal device or network device can use the third output above, that is, the trained signal generation sub-model and The channel estimation submodel generates reference signals and performs channel estimation.

The following describes the training process of the terminal device for the model and the design scheme of the loss function in this example.

In some implementation manners, the terminal device uses the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.

Specifically, the above joint training methods may include:

The terminal device inputs the input information into the first initial model to obtain a first set output by the first initial model, where the first set includes a plurality of first reference signals;

Specifically, the above determination of the first loss function may include:

Referring to the illustrations in detail, FIG. 14 is a schematic diagram of model structure and information transmission in an integrated design scheme of multi-user reference signal and channel estimation in an AI-based wireless communication system according to an embodiment of the present application. In Fig. 14, the first model (before the training is completed, the first model is the first initial model) is specifically the signal generation sub-model, and the second model (before the training is completed, the second model is the second initial model) is specifically Channel estimation submodel. The first channel simulation module may not participate in training, for example, the parameters of the first channel simulation module are fixed.

As shown in FIG. 14, the parameters of the first channel simulation module are represented by a matrix H. During the model training process, the signal generation sub-model outputs a first set, and the first set includes a plurality of first reference signals S. After the first reference signal S passes through the first channel analog module, the second reference signal S' is output, S'=H*S, where the symbol "*" means that two matrices are multiplied; S' means that the original reference signal passes through the wireless channel , the reference signal received by the receiver. Input S' into the channel estimation sub-model, and the channel estimation sub-model performs channel estimation based on S' to obtain channel information H'. Using H' to process S, the third reference signal can be obtained, denoted as S", S"=H'*S, it can be seen that S" is the result of channel estimation (H') compared to the original reference signal (S ) can also be a scene-based reference signal.

The loss function in the embodiment of the present application can be designed based on the degree of difference between H' and H and/or the quality of the reference signal. The smaller the difference between H' and H and the higher the quality of the reference signal, the better the effect of the model. The reference signal here may refer to the original reference signal S generated by the signal generation sub-model (that is, the above-mentioned first reference signal), and/or the reference signal S' (that is, the above-mentioned second reference signal) output and processed through the first channel simulation module. , and/or the reference signal S” obtained after processing the original reference signal using the result of channel estimation (that is, the above-mentioned third reference signal). The quality can be reflected in the cross-correlation between different reference signals, and/or the reference signal The peak-to-average power ratio of the reference signal; the lower the cross-correlation between different reference signals and the lower the peak-to-average power ratio of the reference signal, the higher the quality of the reference signal.

For example, the above-mentioned reference signal quality reference signal quality may be represented by at least one of the following:

cross-correlations between different first reference signals in the first set;

cross-correlations between the first reference signal and other reference signals in the first set;

Peak-to-average power ratios of the first reference signals in the first set.

Wherein, other reference signals of the above-mentioned first reference signal may be pre-stored reference signals, such as reference signals in another reference signal set generated in current training, or another reference signal generated in previous N (N is a positive integer) training A reference signal in the reference signal set.

In another example, the quality of the above reference signal is represented by at least one of the following:

cross-correlation between different second reference signals;

cross-correlation between the second reference signal and other reference signals;

The peak-to-average power ratio of the second reference signal.

Wherein, other reference signals of the above-mentioned second reference signal may be pre-stored reference signals, such as the reference signals obtained after the reference signals in another reference signal set generated in the current training are processed by the first channel simulation module, or the reference signals obtained N times before The reference signal obtained after the reference signal in another reference signal set generated during training is processed by the first channel simulation module.

As another example, the above reference signal quality is represented by at least one of the following:

cross-correlation between different third reference signals;

cross-correlation between the third reference signal and other reference signals;

The peak-to-average power ratio of the third reference signal;

Wherein, the third reference signal is obtained by processing the first reference signal based on the channel information.

Wherein, the other reference signals of the above-mentioned third reference signal may be pre-stored reference signals, such as the reference signals obtained after the above-mentioned channel information processing of the reference signals in another reference signal set generated in the current training, or the reference signals obtained in the previous N times of training Reference signals obtained after the reference signals in another generated reference signal set are processed through the channel information.

In some embodiments, the degree of difference between the above-mentioned channel information (representing the estimated channel) and the parameters of the first channel simulation module (representing the actual channel) can be measured by a specific distance, such as mean square error (MSE, Mean Squared Error ) or normalized mean square error (NMSE); it can also be measured by similarity, such as cosine similarity, cosine similarity squared, etc.

The above-mentioned several measurement methods in the first loss function can adopt the joint measurement method, such as the joint measurement of the addition of equal weights, or the joint measurement of the addition of unequal weights (for example, the proportion of the cross-correlation of the above reference signals in the joint greater weight in the measurement, or greater weight to the accuracy of the above channel estimation results, or equal weights each accounting for 50%); or joint measurement in the form of multiplication or cross-entropy calculation.

For example, the first loss function=x*log (reference signal cross-correlation)+y*log (the degree of difference between the channel estimation result output by the channel estimation sub-model and the actual channel). Referring to Fig. 14, wherein the reference signal cross-correlation can refer to the cross-correlation between different S, different S', or different S" in Fig. 14; The degree of difference between H' and H in Figure 14 is expressed as the degree of difference between H' and H. Among them, x and y can be positive numbers, and if the cross-correlation of the reference signal is given a greater weight, then x>y can be taken; To estimate a larger weight, you can take x<y.

As another example, the first loss function=x*log (reference signal cross-correlation)+y*log (the degree of difference between the channel estimation result output by the channel estimation sub-model and the actual channel)+z*log (reference signal peak-to-average power ratio). Referring to Figure 14, the reference signal can refer to S, S' or S" in Figure 14, and the difference between the channel estimation result of the output of the channel estimation sub-model and the actual channel is expressed as the difference between H' and H in Figure 14 Degree. Among them, x, y and z can be positive numbers.

For another example, the first loss function=x*log (cross-correlation of reference signals)+z*log (peak-to-average power ratio of reference signals);

Or, the first loss function=log(reference signal cross-correlation);

Alternatively, the first loss function=log(the degree of difference between the channel estimation result output by the channel estimation sub-model and the actual channel).

Referring to Figure 14, the reference signal can refer to S, S' or S" in Figure 14, and the difference between the channel estimation result of the output of the channel estimation sub-model and the actual channel is expressed as the difference between H' and H in Figure 14 Degree. Among them, x, y and z can be positive numbers.

The above introduces the first example of model training for terminal equipment in the embodiment of the present application, that is, the integrated design scheme of multi-user reference signal and channel estimation in an AI-based wireless communication system; the following introduces another example, that is, an AI-based wireless communication system Integrated design scheme of multi-user reference signal, channel estimation, and channel information feedback.

Example two:

Optionally, the terminal device may use at least one of the input information, the first channel simulation module, and the second channel simulation module to analyze the first initial model, the second initial model, the third initial model, and the fourth initial model Joint training is performed to obtain the trained first model, the second model, the third model and the fourth model. Among them, after the first initial model is jointly trained, the first model can be obtained, after the second initial model is jointly trained, the second model can be obtained, after the third initial model is jointly trained, the third model can be obtained, and the fourth initial model is jointly trained. A fourth model can be obtained after training.

inputting the channel information into the third initial model to obtain compressed information of the channel information; wherein, the third initial model includes generating an initial submodel and compressing an initial submodel, and the input of generating an initial submodel As the input of the third initial model, the output of the generated initial sub-model is used as the output of the compressed initial sub-model, and the output of the compressed initial sub-model is used as the output of the third initial model; or, the The third initial model includes a compressed initial sub-model;

Inputting the compressed information of the channel information into the second channel simulation module to obtain equivalent received information of the compressed information of the channel information;

inputting equivalent received information of the compressed information of the channel information into the fourth initial model to obtain output information of the fourth initial model;

Based on the first set, the second reference signal, the channel information, the parameters in the first channel simulation module, the input information of the compressed initial submodel, and the output information of the fourth initial model At least one item of determines the second loss function;

Updating the first initial model, the second initial model, the third initial model and the fourth initial model according to the second loss function.

In some implementations, determining the second loss function may include:

Based on the first set, the second reference signal, the channel information, the parameters in the first channel simulation module, the input information of the compressed initial submodel, and the output information of the fourth initial model At least one of the parameters, determine the reference signal quality, the degree of difference between the channel information and the parameters of the first channel simulation module, the degree of difference between the output information of the fourth initial model and the input information of the compressed initial sub-model at least one of the

Based on at least one of the quality of the reference signal, the degree of difference between the channel information and the parameters of the first channel simulation module, and the degree of difference between the output information of the fourth initial model and the input information of the compressed initial sub-model One term, determining the second loss function.

Fig. 15 is a schematic diagram of an AI-based integrated design scheme of multi-user reference signal, channel estimation, and channel information feedback in a wireless communication system according to an embodiment of the present application. In Fig. 15, the above-mentioned first model (before the training is completed, the first model is the first initial model) is specifically a signal generation sub-model, and the above-mentioned second model (before the training is completed, the second model is the second initial model) Specifically, it is a channel estimation sub-model. The first channel simulation module may not participate in the training. For example, the parameters of the first channel simulation module are fixed, and are used to simulate the reference signal received by the terminal device after the reference signal is transmitted through the channel. The first channel simulation module can be stored in the terminal device and the network device respectively in advance, or sent to the terminal device by the network device or other devices before the joint training. The above-mentioned third model (before the training is completed, the third model is the third initial model) is specifically a channel information compression sub-model, or specifically includes a channel information generation sub-model and a channel information compression sub-model; the channel information generation sub-model in Figure 15 The block diagram of is a dotted line, indicating that the channel information generation sub-model is optional. In the case that the third model includes the channel information generation sub-model, the channel information output by the channel estimation sub-model is input to the channel information generation sub-model, and the channel information generation sub-model converts the channel information into a channel information feature vector, and converts the channel information The eigenvectors are input to the channel information compression submodel. In the case that the channel information generation sub-model is not included in the third model, the channel information output by the channel estimation sub-model is input to the channel information compression sub-model. The information output by the channel information compression sub-model is input to the second channel simulation module, and the second channel simulation module may not participate in the training. For example, the parameters of the second channel simulation module are fixed, and the compressed information output by the channel information compression sub-model is passed through Compressed information received by network devices after channel transmission. The second channel simulation module can be stored in the terminal device and the network device respectively in advance, or sent to the terminal device by the network device or other devices before the joint training. The above fourth model (before the training is completed, the fourth model is the fourth initial model) is specifically a channel information recovery sub-model, which is used to recover the received compressed information. The function of the above-mentioned channel information compression sub-model can be to compress its input information, and the function of the channel information recovery sub-model can be to decompress its input information; ideally, the channel information recovery sub-model should be able to recover the channel information The compression submodel compresses the data before.

In Figure 15, although the signal generation sub-model, channel estimation sub-model, channel information generation sub-model, channel information compression sub-model, and channel information recovery sub-model are also distinguished, the above-mentioned models are not generated independently, but through joint design It can be implemented with a trained neural network, and supervised in the above-mentioned integrated solution through a specific loss function design.

The input information in this solution may be the same as the input information in the first example above, and will not be repeated here.

The first item, the set of reference signals:

The reference signal set may be the same as the reference signal set in Example 1 above, which will not be repeated here.

The second item, channel information:

The channel information may be the same as the channel information in the first example above, which will not be repeated here.

The third term, the output of the entire joint scheme:

The output of the entire joint training scheme includes a trained signal generation sub-model, channel estimation sub-model, channel information generation sub-model, channel information compression sub-model, and/or channel information recovery sub-model. It may also include a first channel simulation module. The first channel simulation module may be preset and used to simulate the changes of the reference signal passing through the wireless channel. The first channel simulation module may not participate in model training. It may also include a second channel simulation module, which may be preset and used to simulate the change of the compressed channel information (or the compressed channel information eigenvector) through the wireless channel, the second The channel simulation module may not participate in model training.

The output of the above-mentioned channel estimation sub-model may be the result of channel estimation obtained through the reference signal, for example, the complete channel information of . The result of channel estimation (such as complete channel information) can be directly input to the channel information compression sub-model, or the result of channel estimation can be input into the channel information after being transformed by the channel information generation sub-model (such as the channel eigenvector information obtained by SVD decomposition) Compress submodels. The output of the channel information compression sub-model can be directly input into the channel information recovery sub-model, or can be input into the channel information recovery sub-model after being processed by the second channel simulation module. The function of the above-mentioned second channel simulation module is to simulate the wireless channel environment, for example, it can use the actual channel, or the channel generated by the predetermined channel scene in the protocol, or the channel obtained by channel modeling and fitting, and then the channel information The output of the compression sub-model passes through the above-mentioned channel, and the output of the channel information compression sub-model can be convolved with the above-mentioned channel or data processing equivalent to convolution (for example, converted to the frequency domain by Fourier transform, multiplied, and then Convert to the time domain by inverse Fourier transform, so as to obtain the result of time domain convolution equivalently). The channel information recovery sub-model outputs recovered channel information, which may be complete channel information, or channel feature vector information obtained after conversion (for example, by SVD decomposition) of the channel information generation sub-model.

Each of the above-mentioned models, sub-models or modules may adopt a neural network structure, such as one or more network structures of a fully connected network, a convolutional neural network, a residual network, and a self-attention mechanism network. Fig. 16 is a schematic diagram of another neural network structure according to the present application. As shown in Fig. 16, a multi-user reference signal design scheme according to the embodiment of the present application includes a signal generation sub-model, a first channel simulation module, and a channel estimation sub-model , a channel information compression sub-model, a second channel simulation module and a channel information recovery sub-model, each sub-model/model includes one or more fully connected layers. The input information can be a random number with a length of 64 or an original sequence, and the input information input signal generates a sub-model. The signal generation sub-model generates a reference signal set including a plurality of reference signals, and the reference signal output by the signal generation sub-model is used as an input of the first channel simulation module. The first channel simulation module outputs the result after processing the received reference signal, and inputs the result into the channel estimation sub-model. The channel estimation sub-model performs channel estimation based on the received data to obtain channel information. The channel information is input to the channel information compression sub-model to obtain the compressed channel information. The compressed channel information is input to the second channel simulation module, and the second channel simulation module outputs the result of processing the received compressed channel information, and the processing result is input into the channel information recovery sub-model to obtain the final restored channel information. For example, the size of the recovered channel information can be 8192, and can be transformed into a three-dimensional matrix form of [128,32,2].

The above introduces several output contents of the multi-user reference signal design scheme proposed in this example. In the subsequent use process, terminal equipment or network equipment can use the third output content above, that is, the trained signal generation sub-model, channel Estimation sub-model, channel information generation sub-model (optional), channel information compression sub-model and channel information recovery sub-model, generate reference signal, perform channel estimation and channel information feedback.

In some embodiments, the terminal device uses at least one of the input information and the first channel simulation module and the second channel simulation module to perform the first initial model, the second initial model, the third initial model, and the fourth initial model Joint training to obtain the trained first model, the second model, the third model and the fourth model.

Specifically, the above joint training methods may include:

Specifically, the above determination of the second loss function may include:

Referring to the illustrations in detail, FIG. 17 is a schematic diagram of model structure and information transmission in an AI-based wireless communication system multi-user reference signal, channel estimation, and channel information feedback integrated design scheme according to an embodiment of the present application. In Fig. 17, the first model (before the training is completed, the first model is the first initial model) is specifically the signal generation sub-model, and the second model (before the training is completed, the second model is the second initial model) is specifically Channel estimation submodel. The first channel simulation module may not participate in training, for example, the parameters of the first channel simulation module are fixed.

As shown in FIG. 17, the parameters of the first channel simulation module are represented by a matrix H. During the model training process, the signal generation sub-model outputs a first set, and the first set includes a plurality of first reference signals S. After the first reference signal S passes through the first channel analog module, the second reference signal S' is output, S'=H*S, where the symbol "*" means that two matrices are multiplied; S' means that the original reference signal passes through the wireless channel , the reference signal received by the receiver. Input S' into the channel estimation sub-model, and the channel estimation sub-model performs channel estimation based on S' to obtain channel information H'. Using H' to process S, the third reference signal can be obtained, denoted as S", S"=H'*S, it can be seen that S" is the result of channel estimation (H') compared to the original reference signal (S ) after processing, can also be referred to as a scene-based reference signal. Channel information generation, channel information compression, simulation in the channel and channel information recovery are performed on the channel information H', and the channel information restoration sub-model is obtained Output recovery information.

The loss function in the embodiment of the present application may be based on the degree of difference between H' and H, and/or the quality of the reference signal, and/or the difference between the channel information output by the channel information restoration sub-model and the input information of the channel information compression sub-model degree to design. Specifically, the degree of difference between H' and H reflects the quality of channel estimation, and the degree of difference between the channel information output by the channel information recovery sub-model and the input information of the channel information compression sub-model reflects the quality of channel information feedback. The smaller the difference between H' and H, the higher the quality of the reference signal, and the smaller the difference between the channel information output by the channel information recovery sub-model and the input information of the channel information compression sub-model, the better the effect of the model. The quality of the reference signal here is the same as the quality of the reference signal in the first example above, and will not be repeated here.

In some embodiments, the degree of difference between the channel information (representing the estimated channel) and the parameters of the first channel simulation module (representing the actual channel), and the channel information output by the channel information restoration sub-model and the input of the channel information compression sub-model The degree of difference between information can be measured by a specific distance, such as MSE or NMSE; it can also be measured by the degree of similarity, such as cosine similarity, cosine similarity squared, etc.

The above-mentioned several measurement methods in the second loss function can adopt a joint measurement method, such as a joint measurement with equal weight addition, or a joint measurement with unequal weight addition (for example, the proportion of the cross-correlation of the above reference signals in the joint More weight is given to the measurement, or the accuracy of the above channel estimation results is given more weight, or the accuracy of channel information feedback is given more weight); or joint measurement is performed in the form of multiplication or cross-entropy calculation.

For example, the second loss function=x*log (reference signal cross-correlation)+y*log (the degree of difference between the channel estimation result of the output of the channel estimation sub-model and the actual channel)+z*log (channel information recovery sub-model The degree of difference between the channel information output by the model and the input information of the channel information compression sub-model). Referring to Fig. 17, wherein the reference signal cross-correlation can refer to the cross-correlation between different S, different S', or different S" in Fig. 17; the channel estimation result of the output of the channel estimation sub-model and the actual channel The degree of difference between is expressed as the degree of difference between H' and H in Figure 17. Wherein, x, y, z can be positive numbers.

As another example, the second loss function=x*log (reference signal cross-correlation)+y*log (the degree of difference between the channel estimation result output by the channel estimation sub-model and the actual channel)+z*log (channel information recovery The degree of difference between the channel information output by the sub-model and the input information of the channel information compression sub-model)+h*log (the peak-to-average power ratio of the reference signal). Referring to Fig. 17, wherein the reference signal cross-correlation can refer to the cross-correlation between different S, different S', or different S" in Fig. 17; the channel estimation result of the output of the channel estimation sub-model and the actual channel The degree of difference between is expressed as the degree of difference between H' and H in Figure 17. Wherein, x, y, z, h can be positive numbers.

For another example, the second loss function=x*log (reference signal cross-correlation)+z*log (the degree of difference between the channel information output by the channel information recovery sub-model and the input information of the channel information compression sub-model). Referring to Fig. 17, wherein the reference signal cross-correlation can refer to the cross-correlation between different S, different S', or different S" in Fig. 17; the channel estimation result of the output of the channel estimation sub-model and the actual channel The degree of difference between is expressed as the degree of difference between H' and H in Figure 17. Wherein, x and z can be positive numbers.

As another example, the second loss function=x*log (reference signal cross-correlation)+z*log (the degree of difference between the channel information output by the channel information restoration sub-model and the input information of the channel information compression sub-model)+h* log (the peak-to-average power ratio of the reference signal). Referring to Fig. 17, wherein the reference signal cross-correlation can refer to the cross-correlation between different S, different S', or different S" in Fig. 17; the channel estimation result of the output of the channel estimation sub-model and the actual channel The degree of difference between is expressed as the degree of difference between H' and H in Figure 17. Wherein, x, z, h can be positive numbers.

The second example of model training performed by the terminal device in the embodiment of the present application is introduced above, that is, the integrated design scheme of multi-user reference signal, channel estimation, and channel information feedback in a wireless communication system based on AI.

The manner of the above-mentioned training convergence may include at least one of the following: judging whether the number of iterative training reaches a preset number, and judging whether the degree of difference is smaller than a preset threshold. Wherein, the preset number of times and the preset threshold value can be set according to actual conditions. When it is determined that the training is completed based on the above method, the first initial model, the second initial model, the third initial model and the fourth initial model after the training can be used as the first model, the second model, the third model and the fourth model respectively .

The method of determining the convergence of model training in this example is the same as the method of determining the convergence in the previous example, and will not be repeated here.

It can be seen that through the above joint training method, this solution proposes a solution based on AI for multi-user reference signals in wireless communication systems to obtain better overall advantages in reference signal design, wireless communication solution design, and scene adaptation. Specifically, the integrated design scheme of multi-user reference signal and channel estimation for AI-based wireless communication system is proposed, and the integrated design scheme of multi-user reference signal, channel estimation and channel information feedback for AI-based wireless communication system is proposed, and the corresponding In the scheme, the corresponding scheme input, output, model structure division and loss function design for the integrated design. These designs can form corresponding training schemes for reference signal generation modules and generation schemes for reference signals under different mission objectives. The diverse reference signals that satisfy the subsequent loss function constraints constitute the scenario- and task-oriented multi-user reference signals in this scheme. gather.

The joint design scheme proposed in this disclosure has at least advantages: (1) Instead of using existing reference signals to implement AI-based wireless communication solutions (including channel estimation, channel information feedback, etc.), AI-based wireless communication solutions The solution and the most suitable reference signal construction are completed as an overall solution, so that the reference signal design and the wireless communication solution can achieve the best matching effect; (2) AI-based solutions are conducive to scene adaptation and obtaining corresponding Adaptation gain, and this solution is conducive to considering scene factors when constructing reference signals, so as to obtain better reference signal design, wireless communication solution design, and overall advantages of scene adaptation.

The present application also proposes another communication method. FIG. 18 is a schematic flowchart of another communication method 1800 according to an embodiment of the present application. This method can optionally be applied to the system shown in FIG. 1 , but is not limited to this. The method includes at least some of the following.

S1810: The network device sends a first signal, where the first signal is generated by the first model; the first signal is used for processing by the second model to obtain first information, where the first information includes channel information.

The present application also proposes another communication method. FIG. 19 is a schematic flowchart of a communication method 1900 according to an embodiment of the present application. This method can optionally be applied to the system shown in FIG. 1 , but is not limited thereto. . The method includes at least some of the following.

As shown in Figure 19, the communication method also includes after S1810:

S1920: The network device receives second information, where the second information is obtained by processing the first information by a third model;

S1930: The network device processes the second information by using a fourth model to obtain third information;

Wherein, the first model, the second model, the third model and the fourth model are obtained through joint training.

Optionally, the above-mentioned first signal includes a reference signal.

Optionally, the above-mentioned second model includes a channel estimation sub-model, and the first information includes channel information;

The channel estimation sub-model is used to perform channel estimation based on the first signal to obtain channel information.

Optionally, the third model above includes a compressed sub-model;

The compression sub-model is used to compress the first information to obtain compressed information of the first information; the second information includes the compressed information of the first information.

Correspondingly, the above fourth model includes a recovery sub-model;

The restoration sub-model is used to restore the compressed information of the first information to obtain the restoration information of the first information; the third information includes the restoration information of the first information.

In another embodiment, the third model may include a generation sub-model and a compression sub-model; wherein,

The generation sub-model is used to perform feature transformation on the first information to obtain a first feature vector corresponding to the first information;

The compression sub-model is used to compress the first feature vector to obtain compressed information of the first feature vector; the second information includes the compressed information of the first feature vector.

Correspondingly, the above fourth model includes a recovery sub-model;

The restoration sub-model is used to restore the compressed information of the first feature vector to obtain the restoration information of the first feature vector; the third information includes the restoration information of the first feature vector.

In some implementation manners, the above method further includes: the network device receiving the first model.

Optionally, the foregoing method may further include: the network device receiving the second model.

In some implementation manners, the above method further includes: the network device receiving the first model and the fourth model.

Optionally, the foregoing method may further include: the network device receiving the second model and/or the third model.

Optionally, the above method may further include: the network device receiving a first coding model, where the first coding model includes a channel estimation sub-model, a generation sub-model, and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

That is to say, the terminal device packs the channel estimation sub-model, the generation sub-model and the compression sub-model into one model, that is, the first coding model, and the first coding model is transmitted and used as a whole.

Optionally, the above-mentioned first encoding model is carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, uplink data transmission for artificial intelligence business transmission requirements.

Optionally, the above method may further include: the network device receiving a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.

That is to say, the terminal device packs the channel estimation sub-model and the compression sub-model into one model, that is, the second coding model, and the second coding model is transmitted and used as a whole.

Optionally, the above-mentioned second coding model is carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, uplink data transmission for artificial intelligence business transmission requirements.

Further, the above method may also include training the model/sub-model by the network device.

In some embodiments, the above method further includes: the network device uses the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the Describe the second model.

Wherein, the joint training of the first initial model and the second initial model by the network device using the input information and/or the first channel simulation module may specifically include:

The network device inputs the input information into the first initial model to obtain a first set output by the first initial model, where the first set includes a plurality of first reference signals;

Optionally, the above-mentioned determination of the first loss function includes:

In other embodiments, the above method further includes: the network device uses at least one of the input information, the first channel simulation module, and the second channel simulation module to perform the first initial model, the second initial model, the third The initial model and the fourth initial model are jointly trained to obtain the trained first model, the second model, the third model and the fourth model.

Specific training methods can include:

Optionally, determining the second loss function may include:

In some implementation manners, the above reference signal quality is represented by at least one of the following:

cross-correlations between different first reference signals in the first set;

cross-correlations between first reference signals in the first set and other reference signals;

The peak-to-average power ratio of the first reference signals in the first set.

Or, in some implementation manners, the quality of the above-mentioned reference signal is represented by at least one of the following:

cross-correlation between different said second reference signals;

The peak-to-average power ratio of the second reference signal.

Or, in some implementation manners, the above reference signal quality is represented by at least one of the following:

cross-correlation between different said third reference signals;

a peak-to-average power ratio of the third reference signal;

Optionally, the above degree of difference is measured by distance and/or similarity.

In some embodiments, the input information includes at least one of the following: no input, noise, random numbers, sequences in a preset sequence set, channel type indication information, channel data sample information, wireless channel or scene related information.

Optionally, the preset sequence set includes at least one of the following: m sequence set, golden sequence set, and zc sequence set.

Optionally, the channel type indication information indicates at least one of the following: frequency information corresponding to the channel, environment information corresponding to the channel, and scene information corresponding to the channel.

Optionally, the wireless channel or scene-related information includes at least one of the following: channel signal-to-noise ratio, signal-to-interference-noise ratio, channel type, bandwidth information, and delay information.

Optionally, the format of the noise, the random number, or the sequence in the preset sequence set is the same as the format of the output data of the first initial model.

Optionally, the format of the noise, the random number, or the sequence in the preset sequence set includes at least one of the following formats: a one-dimensional vector, a two-dimensional matrix, and a high-dimensional matrix.

Optionally, the format of the noise, the random number, or the sequence in the preset sequence set is stipulated in a protocol or signaling.

Optionally, the above input information is used to input at least one of the following: a first initial model, a first channel simulation module, and a second initial model.

In some implementation manners, the above-mentioned channel information is distributed in the first dimension and/or the second dimension.

Or, the above channel information may be distributed in at least one of the first dimension, the second dimension and the third dimension.

Optionally, the above-mentioned first dimension is a frequency domain dimension, and the channel information includes data distributed on M1 frequency domain granularities of the frequency domain dimension; the M1 is a positive integer.

Wherein, the frequency domain granularity includes a RBs and/or b subcarriers, and a or b is a positive integer.

Optionally, the above-mentioned first dimension is a time domain dimension, and the channel information includes data distributed on M2 delay granularities of the time domain dimension; the M2 is a positive integer.

Wherein, the delay granularity includes at least one of the following: p1 microseconds, p2 symbol length, and p3 symbol sampling point numbers, and the p1, p2 or p3 are positive integers.

The above symbols may include OFDM symbols.

Optionally, the foregoing second dimension is a spatial domain dimension.

For example, the foregoing spatial domain dimension is an antenna dimension, and the channel information includes data distributed on N1 first granularities of the antenna dimension, where N1 is a positive integer.

Specifically, the foregoing first granularity may include a pair of transmitting and receiving antennas.

In another example, the foregoing space domain dimension is an angle domain dimension, and the channel information includes data distributed on N2 second granularities of the angle domain dimension, where N2 is a positive integer.

Specifically, the second granularity may include angular intervals.

In some implementation manners, the above-mentioned third dimension includes a complex dimension, and the complex dimension includes 2 elements, which are respectively used to bear the real part and the imaginary part of the data included in the channel information.

In some implementations, the above channel information is distributed in a T-dimensional matrix, and the T-dimensional matrix is a matrix formed after splitting and/or combining at least one of the first dimension, the second dimension and the third dimension, so Said T is a positive integer.

Specifically, the above S can be 2, 4 or 8.

Specifically, the above U may be 16, 32, 48, 64, 128 or 256.

In some implementation manners, the above method may further include: the network device sending the second model.

Further, the method may further include: the network device sending the first model.

In some other implementation manners, the foregoing method may further include: the network device sending the second model and the third model.

Further, the method may further include: sending the first model and/or the fourth model by the network device.

In some embodiments, the above-mentioned first model, the second model, the third model, the fourth model, the sub-model in the first model, the sub-model in the second model, the The sub-model in the third model or the sub-model in the fourth model is carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, and transmission requirements for artificial intelligence services downlink data transmission.

In some implementations, the above method may further include: the network device sending a first coding model, where the first coding model includes a channel estimation sub-model, a generation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

That is to say, the network device can package the channel estimation sub-model, generation sub-model and compression sub-model as a whole, that is, the first coding model, and the first coding model is transmitted and used as a whole.

The above-mentioned first coding model may be carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, downlink data transmission for artificial intelligence business transmission requirements.

In some other implementation manners, the above method may further include: the network device sending a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.

That is to say, the network device can package the channel estimation sub-model and the compression sub-model as a whole, that is, the second coding model, and the second coding model is transmitted and used as a whole.

The above-mentioned second coding model may be carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, downlink data transmission for artificial intelligence business transmission requirements.

The embodiment of the present application also proposes a model training method. FIG. 20 is a schematic flowchart of another communication method 2000 according to the embodiment of the present application. This method can optionally be applied to the system shown in FIG. 1 , but does not It doesn't stop there. The model training method may be executed by a terminal device, or by a network device, or by other electronic devices. The method includes at least some of the following.

S2010: Using the input information and/or the first channel simulation module, jointly train the first initial model and the second initial model to obtain the trained first model and the second model.

In some embodiments, the above-mentioned joint training includes:

inputting the input information into the first initial model to obtain a first set of outputs of the first initial model, the first set including a plurality of first reference signals;

In some embodiments, the above-mentioned joint training includes:

Using at least one of the input information, the first channel simulation module and the second channel simulation module, the first initial model, the second initial model, the third initial model, and the fourth initial model are jointly trained to obtain the trained first A model, a second model, a third model and a fourth model.

Specifically, the above joint training may include:

Optionally, the above determination of the second loss function may include:

In an implementation manner, the above-mentioned reference signal quality is represented by at least one of the following:

cross-correlations between different first reference signals in the first set;

In another implementation manner, the quality of the above-mentioned reference signal is represented by at least one of the following:

cross-correlation between different said second reference signals;

The peak-to-average power ratio of the second reference signal.

In another implementation manner, the above reference signal quality is represented by at least one of the following:

cross-correlation between different said third reference signals;

a peak-to-average power ratio of the third reference signal;

Optionally, the above-mentioned degree of difference may be measured by using distance and/or similarity.

Wherein, the preset sequence set may include at least one of the following: m sequence set, golden sequence set, and zc sequence set.

Wherein, the channel type indication information may indicate at least one of the following: frequency information corresponding to the channel, environment information corresponding to the channel, and scene information corresponding to the channel.

Wherein, the wireless channel or scene-related information may include at least one of the following: channel signal-to-noise ratio, signal-to-interference-noise ratio, channel type, bandwidth information, and delay information.

Wherein, the format of the noise, the random number or the sequence in the preset sequence set may be the same as the format of the output data of the first initial model.

Wherein, the format of the noise, the random number, or the sequence in the preset sequence set may include at least one of the following formats: a one-dimensional vector, a two-dimensional matrix, and a high-dimensional matrix.

Wherein, the format of the noise, the random number, or the sequence in the preset sequence set may be stipulated through a protocol or signaling.

Wherein, the input information may be used to input at least one of the following: the first initial model, the first channel simulation module, and the second initial model.

In some embodiments, the channel information is distributed in at least one of the first dimension, the second dimension and the third dimension.

In an implementation manner, the first dimension is a frequency domain dimension, and the channel information includes data distributed on M1 frequency domain granularities of the frequency domain dimension; the M1 is a positive integer.

For example, the frequency domain granularity includes a RBs and/or b subcarriers, where a or b is a positive integer.

In another implementation manner, the first dimension is a time domain dimension, and the channel information includes data distributed on M2 delay granularities of the time domain dimension; the M2 is a positive integer.

For example, the delay granularity includes at least one of the following: p1 microseconds, p2 symbol length, and p3 symbol sampling point numbers, where p1, p2 or p3 is a positive integer.

Optionally, the foregoing symbols include OFDM symbols.

In some embodiments, the above-mentioned second dimension is a spatial domain dimension.

Optionally, the foregoing first granularity includes a pair of transmitting and receiving antennas.

Optionally, the above-mentioned second granularity includes angular intervals.

For example, the above S may be 2, 4 or 8.

For example, the above U can be 16, 32, 48, 64, 128 or 256.

After the terminal device executes and the network device is trained by other electronic devices using the above training method, the trained model can be sent to the required device for execution and/or network device. The transmission method of the model is the same as the transmission of the model in the above communication method. The method is the same and will not be repeated here.

The embodiment of the present application also proposes a terminal device. FIG. 21 is a schematic structural diagram of a terminal device 2100 according to the embodiment of the present application, including:

The first receiving module 2110 is configured to receive a first signal, the first signal is generated by a first model;

The first processing module 2120 is configured to process the first signal by using a second model to obtain first information;

Optionally, the above-mentioned terminal equipment also includes:

a second processing module, configured to process the first information by using a third model to obtain second information;

The first sending module is configured to send the second information, and the second information is used for processing by the fourth model to obtain third information;

Optionally, the above-mentioned first signal includes a reference signal.

Optionally, the third model above includes a compressed sub-model;

Optionally, the fourth model above includes a recovery sub-model;

Optionally, the above-mentioned third model includes a generation sub-model and a compression sub-model; wherein,

Optionally, the fourth model above includes a recovery sub-model;

In some implementation manners, the terminal device above further includes: a second receiving module, configured to receive the second model.

Optionally, the above-mentioned second receiving module is also configured to receive the first model.

In some other implementation manners, the terminal device above further includes: a third receiving module, configured to receive the second model and the third model.

Optionally, the above-mentioned third receiving module is further configured to receive the first model and/or the fourth model.

In some implementation manners, the above-mentioned terminal device also includes:

The fourth receiving module is configured to receive the first coding model, the first coding model includes a channel estimation sub-model, a generation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes said second model;

The generative sub-model and the compressed sub-model constitute the third model.

The fifth receiving module is configured to receive a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes said second model;

The compressed sub-models constitute the third model.

The first training module is configured to use the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.

The second training module is used to combine the first initial model, the second initial model, the third initial model, and the fourth initial model by using at least one of the input information, the first channel simulation module, and the second channel simulation module training to obtain the trained first model, the second model, the third model and the fourth model.

The specific manner of performing joint training by the first training module or the second training module is the same as the training manner in the foregoing method embodiments, and will not be repeated here.

In some implementation manners, the terminal device above further includes: a second sending module, configured to send the first model.

Optionally, the above-mentioned second sending module is further configured to send the second model.

In some implementation manners, the terminal device above further includes: a third sending module, configured to send the first model and the fourth model.

Optionally, the above-mentioned third sending module is further configured to send the second model and/or the third model.

The fourth sending module is used to send the first coding model, the first coding model includes a channel estimation sub-model, a generation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

A fifth sending module, configured to send a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.

It should be understood that the above and other operations and/or functions of the modules in the terminal device according to the embodiment of the present application are to implement the corresponding process of the terminal device in the method 400 in FIG.

The embodiment of the present application also proposes a network device. FIG. 22 is a schematic structural diagram of a network device 2200 according to the embodiment of the present application, including:

The sixth sending module 2210 is configured to send a first signal, the first signal is generated by the first model; the first signal is used for processing by the second model to obtain the first information;

Optionally, the above-mentioned first model and the second model are obtained through joint training.

In some implementation manners, the above-mentioned network equipment also includes:

A sixth receiving module, configured to receive second information, the second information is obtained by processing the first information by the third model;

a third processing module, configured to process the second information by using a fourth model to obtain third information;

Optionally, the above-mentioned first signal includes a reference signal.

In some implementations, the above-mentioned second model includes a channel estimation sub-model, and the first information includes channel information;

In some embodiments, the above-mentioned third model includes a compressed sub-model;

In some embodiments, the above-mentioned fourth model includes a recovery sub-model;

In some embodiments, the above-mentioned third model includes a generation sub-model and a compression sub-model; wherein,

Optionally, the fourth model above includes a recovery sub-model;

In some implementation manners, the foregoing network device further includes: a seventh receiving module, configured to receive the first model.

Optionally, the seventh receiving module is further configured to receive the second model.

In some implementation manners, the foregoing network device further includes: an eighth receiving module, configured to receive the first model and the fourth model.

Optionally, the eighth receiving module is further configured to receive the second model and/or the third model.

In some implementations, the network device above further includes: a ninth receiving module, configured to receive a first coding model, where the first coding model includes a channel estimation sub-model, a generation sub-model, and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

In some implementation manners, the foregoing network device further includes: a tenth receiving module, configured to receive a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.

In some embodiments, the network device above further includes: a third training module, configured to use input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain all the first model and the second model.

In some embodiments, the network device above further includes: a fourth training module, configured to use input information, at least one of the first channel simulation module and the second channel simulation module, to train the first initial model, the second initial model , the third initial model, and the fourth initial model are jointly trained to obtain the trained first model, the second model, the third model, and the fourth model.

The specific manner of performing joint training by the third training module or the fourth training module is the same as the training manner in the foregoing method embodiment, and will not be repeated here.

In some implementation manners, the foregoing network device further includes: a seventh sending module, configured to send the second model.

Optionally, the seventh sending module is further configured to send the first model.

In some implementation manners, the foregoing network device further includes: an eighth sending module, configured to send the second model and the third model.

Optionally, the eighth sending module is further configured to send the first model and/or the fourth model.

A ninth sending module, configured to send a first coding model, where the first coding model includes a channel estimation sub-model, a generation sub-model, and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

A tenth sending module, configured to send a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.

It should be understood that the above-mentioned and other operations and/or functions of the modules in the network device according to the embodiment of the present application are respectively for realizing the corresponding flow of the network device in the method 1800 in FIG.

The embodiment of the present application also proposes a model training device. FIG. 23 is a schematic structural diagram of a model training device 2300 according to the embodiment of the present application, including:

The joint training module 2310 is configured to use the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.

In some implementations, the above joint training module 2310 is used to:

Optionally, the aforementioned joint training module 2310 is used for:

Optionally, the above determination of the second loss function includes:

Optionally, the above reference signal quality is represented by at least one of the following:

cross-correlations between different first reference signals in the first set;

Optionally, the quality of the above reference signal is represented by at least one of the following:

cross-correlation between different said second reference signals;

The peak-to-average power ratio of the second reference signal.

cross-correlation between different said third reference signals;

a peak-to-average power ratio of the third reference signal;

The specific manner of the joint training of the above-mentioned joint training module is the same as the training manner in the above-mentioned method embodiment, and will not be repeated here.

It should be understood that the above and other operations and/or functions of the modules in the model training device according to the embodiment of the present application are to realize the corresponding process of the model training device in the method 2000 in FIG.

It should be noted that the functions described by the various modules (sub-models, units or components, etc.) in the terminal device 2100, the network device 2200, and the model training device 2300 in the embodiment of the present application may be composed of different modules (sub-models, units or components, etc.), or by the same module (submodel, unit or component, etc.), for example, the first sending module and the second sending module can be different modules, or the same module, both of which can Realize its corresponding function in the embodiment of the present application. In addition, the sending module and the receiving module in the embodiment of the present application may be realized by a transceiver of the device, and part or all of the other modules may be realized by a processor of the device.

Fig. 24 is a schematic structural diagram of a communication device or a model training device 700 according to an embodiment of the present application. The communication device or model training device 700 shown in FIG. 24 includes a processor 710, and the processor 710 can invoke and run a computer program from a memory, so as to implement the method in the embodiment of the present application.

Optionally, as shown in FIG. 24 , the communication device or model training device 700 may further include a memory 720 . Wherein, the processor 710 can invoke and run a computer program from the memory 720, so as to implement the method in the embodiment of the present application.

Wherein, the memory 720 may be an independent device independent of the processor 710 , or may be integrated in the processor 710 .

Optionally, as shown in FIG. 24, the communication device or model training device 700 may further include a transceiver 730, and the processor 710 may control the transceiver 730 to communicate with other devices, specifically, to send information or data to other devices , or receive messages or data from other devices.

Wherein, the transceiver 730 may include a transmitter and a receiver. The transceiver 730 may further include antennas, and the number of antennas may be one or more.

Optionally, the communication device or model training device 700 can be a terminal device in the embodiment of the present application, and the communication device or model training device 700 can implement the corresponding processes implemented by the terminal device in each method of the embodiment of the present application, in order to It is concise and will not be repeated here.

Optionally, the communication device or model training device 700 can be the network device of the embodiment of the present application, and the communication device or model training device 700 can implement the corresponding process implemented by the network device in each method of the embodiment of the present application, in order to It is concise and will not be repeated here.

FIG. 25 is a schematic structural diagram of a chip 800 according to an embodiment of the present application. The chip 800 shown in FIG. 25 includes a processor 810, and the processor 810 can call and run a computer program from a memory, so as to implement the method in the embodiment of the present application.

Optionally, as shown in FIG. 25 , the chip 800 may further include a memory 820 . Wherein, the processor 810 can call and run a computer program from the memory 820, so as to implement the method in the embodiment of the present application.

Wherein, the memory 820 may be an independent device independent of the processor 810 , or may be integrated in the processor 810 .

Optionally, the chip 800 may also include an input interface 830 . Wherein, the processor 810 may control the input interface 830 to communicate with other devices or chips, specifically, may obtain information or data sent by other devices or chips.

Optionally, the chip 800 may also include an output interface 840 . Wherein, the processor 810 can control the output interface 840 to communicate with other devices or chips, specifically, can output information or data to other devices or chips.

Optionally, the chip can be applied to the terminal device in the embodiments of the present application, and the chip can implement the corresponding processes implemented by the terminal device in the methods of the embodiments of the present application. For the sake of brevity, details are not repeated here.

Optionally, the chip can be applied to the network device in the embodiment of the present application, and the chip can implement the corresponding processes implemented by the network device in the methods of the embodiment of the present application. For the sake of brevity, details are not repeated here.

It should be understood that the chip mentioned in the embodiment of the present application may also be called a system-on-chip, a system-on-chip, a system-on-a-chip, or a system-on-a-chip.

The processor mentioned above can be a general-purpose processor, a digital signal processor (DSP), an off-the-shelf programmable gate array (FPGA), an application specific integrated circuit (ASIC) or Other programmable logic devices, transistor logic devices, discrete hardware components, etc. Wherein, the general-purpose processor mentioned above may be a microprocessor or any conventional processor or the like.

The aforementioned memories may be volatile memories or nonvolatile memories, or may include both volatile and nonvolatile memories. Among them, the non-volatile memory can be read-only memory (read-only memory, ROM), programmable read-only memory (programmable ROM, PROM), erasable programmable read-only memory (erasable PROM, EPROM), electrically programmable Erases programmable read-only memory (electrically EPROM, EEPROM) or flash memory. The volatile memory may be random access memory (RAM).

It should be understood that the above-mentioned memory is illustrative but not restrictive. For example, the memory in the embodiment of the present application may also be a static random access memory (static RAM, SRAM), a dynamic random access memory (dynamic RAM, DRAM), Synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous connection Dynamic random access memory (synch link DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM), etc. That is, the memory in the embodiments of the present application is intended to include, but not be limited to, these and any other suitable types of memory.

In the above embodiments, all or part of them may be implemented by software, hardware, firmware or any combination thereof. When implemented using software, it may be implemented in whole or in part 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 the computer, the processes or functions according to the embodiments of the present application will be generated in whole or in part. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable device. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, e.g. (such as coaxial cable, optical fiber, digital subscriber line (Digital Subscriber Line, DSL)) or wireless (such as infrared, wireless, microwave, etc.) to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device such as a server or a data center integrated with one or more available media. The available medium may be a magnetic medium (such as a floppy disk, a hard disk, or a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (Solid State Disk, SSD)), etc.

It should be understood that, in various embodiments of the present application, the sequence numbers of the above-mentioned processes do not mean the order of execution, and the execution order of the processes should be determined by their functions and internal logic, and should not be used in the embodiments of the present application. The implementation process constitutes any limitation.

Those skilled in the art can clearly understand that for the convenience and brevity of the description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

The above is only the specific implementation of the application, but the scope of protection of the application is not limited thereto. Anyone familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the application, and should covered within the scope of protection of this application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

A method of communication comprising:

The terminal device receives a first signal, where the first signal is generated by a first model;

The terminal device processes the first signal by using a second model to obtain first information; the first information includes channel information;

Wherein, the first model and the second model are obtained through joint training.
The method according to claim 1, further comprising:

The terminal device processes the first information by using a third model to obtain second information;

The terminal device sends the second information, and the second information is used for processing by a fourth model to obtain third information;

Wherein, the first model, the second model, the third model and the fourth model are obtained through joint training.
The method of claim 1 or 2, wherein the first signal comprises a reference signal.
A method according to any one of claims 1 to 3, wherein said second model comprises a channel estimation sub-model;

The channel estimation sub-model is used to perform channel estimation based on the first signal to obtain channel information.
A method according to claim 2 or 3, wherein said third model comprises a compressed sub-model;

The compression sub-model is used to compress the first information to obtain compressed information of the first information; the second information includes the compressed information of the first information.
The method of claim 5, wherein the fourth model comprises a recovery sub-model;

The restoration sub-model is used to restore the compressed information of the first information to obtain the restoration information of the first information; the third information includes the restoration information of the first information.
The method according to claim 2 or 3, wherein the third model comprises a generation sub-model and a compression sub-model; wherein,

The generation sub-model is used to perform feature transformation on the first information to obtain a first feature vector corresponding to the first information;

The compression sub-model is used to compress the first feature vector to obtain compressed information of the first feature vector; the second information includes the compressed information of the first feature vector.
The method of claim 7, wherein the fourth model comprises a restoration sub-model;

The restoration sub-model is used to restore the compressed information of the first feature vector to obtain the restoration information of the first feature vector; the third information includes the restoration information of the first feature vector.
The method according to any one of claims 1 to 8, further comprising:

The terminal device receives the second model.
The method of claim 9, further comprising:

The terminal device receives the first model.
The method according to any one of claims 1 to 8, further comprising:

The terminal device receives the second model and the third model.
The method of claim 11, further comprising:

The terminal device receives the first model and/or the fourth model.
The method of claim 12, wherein the first model, the second model, the third model, the fourth model, a submodel in the first model, the second model The submodel in the third model or the submodel in the fourth model is carried by one of the following: downlink control signaling, media access control MAC control element CE message, radio resource control RRC message , broadcast messages, downlink data transmission, and downlink data transmission for the transmission needs of artificial intelligence services.
The method of claim 2, further comprising:

The terminal device receives a first coding model, and the first coding model includes a channel estimation sub-model, a generation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes said second model;

The generative sub-model and the compressed sub-model constitute the third model.
The method according to claim 14, wherein the terminal device uses a second model to process the first signal to obtain first information, and the terminal device uses a third model to process the first information, Get second information, including:

The terminal device processes the first signal by using the first coding model to obtain the second information.
The method according to claim 14 or 15, wherein the first coding model is carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, and the transmission requirements for artificial intelligence services Downlink data transmission.
The method of claim 2, further comprising:

The terminal device receives a second coding model, and the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes said second model;

The compressed sub-models constitute the third model.
The method according to claim 17, wherein the terminal device uses a second model to process the first signal to obtain first information, and the terminal device uses a third model to process the first information, Get second information, including:

The terminal device processes the first signal by using the second coding model to obtain the second information.
The method according to claim 17 or 18, wherein the second coding model is carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, and the transmission requirements for artificial intelligence services Downlink data transmission.
The method according to any one of claims 1 to 8, further comprising:

The terminal device uses the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.
The method according to claim 20, wherein the terminal device uses the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model, including:

The terminal device inputs the input information into the first initial model to obtain a first set output by the first initial model, where the first set includes a plurality of first reference signals;

inputting any one of the first reference signals in the first set to the first channel simulation module to obtain a second reference signal;

inputting the second reference signal into the second initial model to obtain channel information;

determining a first loss function based on at least one of the first set, the second reference signal, the channel information, and parameters in the first channel simulation module;

updating the first initial model and the second initial model according to the first loss function.
The method according to claim 21, wherein said determining the first loss function comprises:

Determine the channel information and the parameters of the first channel simulation module based on at least one of the first set, the second reference signal, the channel information and the parameters of the first channel simulation module degree of variance and/or reference signal quality;

The first loss function is determined based on the degree of difference between the channel information and the parameters of the first channel simulation module and/or the quality of a reference signal.
The method according to any one of claims 1 to 8, further comprising:

The terminal device uses at least one of the input information, the first channel simulation module, and the second channel simulation module to perform joint training on the first initial model, the second initial model, the third initial model, and the fourth initial model to obtain The trained first model, the second model, the third model and the fourth model.
The method according to claim 23, wherein the terminal device uses at least one of the input information, the first channel simulation module, and the second channel simulation module to perform the first initial model, the second initial model, and the third initial model, the fourth initial model for joint training, including:

The terminal device inputs the input information into the first initial model to obtain a first set output by the first initial model, where the first set includes a plurality of first reference signals;

inputting any one of the first reference signals in the first set to the first channel simulation module to obtain a second reference signal;

inputting the second reference signal into the second initial model to obtain channel information;

inputting the channel information into the third initial model to obtain compressed information of the channel information; wherein, the third initial model includes generating an initial submodel and compressing an initial submodel, and the input of generating an initial submodel As the input of the third initial model, the output of the generated initial sub-model is used as the output of the compressed initial sub-model, and the output of the compressed initial sub-model is used as the output of the third initial model; or, the The third initial model includes a compressed initial sub-model;

Inputting the compressed information of the channel information into the second channel simulation module to obtain equivalent received information of the compressed information of the channel information;

inputting equivalent received information of the compressed information of the channel information into the fourth initial model to obtain output information of the fourth initial model;

Based on the first set, the second reference signal, the channel information, the parameters in the first channel simulation module, the input information of the compressed initial submodel, and the output information of the fourth initial model At least one item of determines the second loss function;

Updating the first initial model, the second initial model, the third initial model and the fourth initial model according to the second loss function.
The method of claim 24, wherein said determining a second loss function comprises:

Based on the first set, the second reference signal, the channel information, the parameters in the first channel simulation module, the input information of the compressed initial submodel, and the output information of the fourth initial model At least one of the parameters, determine the reference signal quality, the degree of difference between the channel information and the parameters of the first channel simulation module, the degree of difference between the output information of the fourth initial model and the input information of the compressed initial sub-model at least one of the

Based on at least one of the quality of the reference signal, the degree of difference between the channel information and the parameters of the first channel simulation module, and the degree of difference between the output information of the fourth initial model and the input information of the compressed initial sub-model One term, determining the second loss function.
The method according to claim 22 or 25, wherein the reference signal quality is represented by at least one of the following:

cross-correlations between different first reference signals in the first set;

cross-correlations between first reference signals in the first set and other reference signals;

The peak-to-average power ratio of the first reference signals in the first set.
The method according to claim 22 or 25, wherein the quality of the reference signal is represented by at least one of the following:

cross-correlation between different said second reference signals;

cross-correlation between the second reference signal and other reference signals;

The peak-to-average power ratio of the second reference signal.
The method according to claim 22 or 25, wherein the reference signal quality is represented by at least one of the following:

cross-correlation between different said third reference signals;

cross-correlation between the third reference signal and other reference signals;

a peak-to-average power ratio of the third reference signal;

Wherein, the third reference signal is obtained by processing the first reference signal based on the channel information.
The method according to claim 22 or 25, wherein the degree of difference is measured by distance and/or similarity.
The method according to any one of claims 20 to 29, wherein the input information includes at least one of the following: no input, noise, random numbers, sequences in a preset sequence set, channel type indication information, channel data samples Information, wireless channel or scene related information.
The method according to claim 30, wherein the preset sequence set includes at least one of the following: an m-sequence set, a golden sequence set, and a zc sequence set.
The method according to claim 30, wherein the channel type indication information indicates at least one of the following: frequency information corresponding to the channel, environment information corresponding to the channel, and scene information corresponding to the channel.
The method according to claim 30, wherein the wireless channel or scene-related information includes at least one of the following: channel signal-to-noise ratio, signal-to-interference-noise ratio, channel type, bandwidth information, and delay information.
The method according to claim 30, wherein the format of the noise, the random number or the sequence in the preset sequence set is the same as the format of the first initial model output data.
The method according to claim 30 or 34, wherein the formats of the noise, random numbers, or sequences in the preset sequence set include at least one of the following formats: one-dimensional vector, two-dimensional matrix, and high-dimensional matrix.
The method according to claim 30, 34 or 35, wherein the format of the noise, the random number or the sequence in the preset sequence set is stipulated through a protocol or signaling.
The method according to any one of claims 30 to 36, wherein the input information is used to input at least one of the following: a first initial model, a first channel simulation module, and a second initial model.
The method according to any one of claims 21, 22, 24 or 25, wherein the channel information is distributed in the first dimension and/or the second dimension.
The method according to any one of claims 21, 22, 24 or 25, wherein the channel information is distributed in at least one of the first dimension, the second dimension and the third dimension.
A method according to claim 38 or 39, wherein,

The first dimension is a frequency domain dimension, and the channel information includes data distributed on M1 frequency domain granularities of the frequency domain dimension; the M1 is a positive integer.
The method according to claim 40, wherein the frequency domain granularity includes a RBs and/or b subcarriers, and a or b is a positive integer.
A method according to claim 38 or 39, wherein,

The first dimension is a time domain dimension, and the channel information includes data distributed on M2 delay granularities of the time domain dimension; the M2 is a positive integer.
The method according to claim 42, wherein the delay granularity includes at least one of the following: p1 microseconds, p2 symbol lengths, p3 symbol sampling points, and the p1, p2 or p3 is positive integer.
43. The method of claim 43, wherein the symbols comprise OFDM symbols.
A method as claimed in claim 38 or 39, wherein the second dimension is a spatial domain dimension.
The method according to claim 45, wherein the spatial domain dimension is an antenna dimension, and the channel information includes data distributed on N1 first granularities of the antenna dimension, where N1 is a positive integer.
46. The method of claim 46, wherein the first granularity includes a pair of transceiving antennas.
The method according to claim 45, wherein the space domain dimension is an angle domain dimension, and the channel information includes data distributed on N2 second granularities of the angle domain dimension, where N2 is a positive integer.
48. The method of claim 48, wherein the second granularity includes angular spacing.
The method according to claim 39, wherein the third dimension includes a complex dimension, and the complex dimension includes 2 elements, which are respectively used to bear the real part and the imaginary part of the data included in the channel information.
The method according to any one of claims 39 to 50, wherein the channel information is distributed in a T-dimensional matrix, and the T-dimensional matrix is decomposed for at least one of the first dimension, the second dimension and the third dimension. The matrix formed after dividing and/or combining, said T is a positive integer.
The method according to any one of claims 21, 22, 24, 25 or 38 to 51, wherein the channel information includes S groups of characteristic sequences with a length U, and the S or U is a positive integer.
The method of claim 52, wherein S is 2, 4 or 8.
The method according to claim 52 or 53, wherein said U is 16, 32, 48, 64, 128 or 256.
A method according to any one of claims 20 to 54, further comprising:

The terminal device sends the first model.
The method of claim 55, further comprising:

The terminal device sends the second model.
A method according to any one of claims 23 to 54, further comprising:

The terminal device sends the first model and the fourth model.
The method of claim 57, further comprising:

The terminal device sends the second model and/or the third model.
The method of claim 58, wherein said first model, said second model, said third model, said fourth model, a submodel within said first model, said second model The submodel in the third model or the submodel in the fourth model is carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, for Uplink data transmission required for artificial intelligence business transmission.
A method according to any one of claims 20 to 54, further comprising:

The terminal device sends a first coding model, and the first coding model includes a channel estimation sub-model, a generation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The generating sub-model and the compressing sub-model constitute a third model.
The method according to claim 60, wherein the first coding model is carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, uplink data for artificial intelligence business transmission requirements transmission.
A method according to any one of claims 20 to 54, further comprising:

The terminal device sends a second coding model, and the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.
The method according to claim 62, wherein the second coding model is carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, uplink data for artificial intelligence business transmission requirements transmission.
A method of communication comprising:

The network device sends a first signal, the first signal is generated by the first model; the first signal is used for processing by the second model to obtain first information; the first information includes channel information;

Wherein, the first model and the second model are obtained through joint training.
The method of claim 64, further comprising:

The network device receives second information, and the second information is obtained by processing the first information by a third model;

The network device processes the second information by using a fourth model to obtain third information;

Wherein, the first model, the second model, the third model and the fourth model are obtained through joint training.
A method as claimed in claim 64 or 65, wherein the first signal comprises a reference signal.
A method according to any one of claims 64 to 66, wherein said second model comprises a channel estimation sub-model;

The channel estimation sub-model is used to perform channel estimation based on the first signal to obtain channel information.
A method according to claim 65 or 66, wherein said third model comprises a compressed sub-model;

The compression sub-model is used to compress the first information to obtain compressed information of the first information; the second information includes the compressed information of the first information.
The method of claim 68, wherein the fourth model comprises a restoration sub-model;

The restoration sub-model is used to restore the compressed information of the first information to obtain the restoration information of the first information; the third information includes the restoration information of the first information.
A method according to claim 65 or 66, wherein said third model comprises a generative sub-model and a compressed sub-model; wherein,

The generation sub-model is used to perform feature transformation on the first information to obtain a first feature vector corresponding to the first information;

The compression sub-model is used to compress the first feature vector to obtain compressed information of the first feature vector; the second information includes the compressed information of the first feature vector.
The method of claim 70, wherein the fourth model comprises a restoration sub-model;

The restoration sub-model is used to restore the compressed information of the first feature vector to obtain the restoration information of the first feature vector; the third information includes the restoration information of the first feature vector.
A method according to any one of claims 64 to 71, further comprising:

The network device receives the first model.
The method of claim 72, further comprising:

The network device receives the second model.
A method according to any one of claims 64 to 71, further comprising:

The network device receives the first model and the fourth model.
The method of claim 74, further comprising:

The network device receives the second model and/or the third model.
The method of claim 75, wherein said first model, said second model, said third model, said fourth model, a submodel within said first model, said second model The submodel in the third model or the submodel in the fourth model is carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, for Uplink data transmission required for artificial intelligence business transmission.
A method according to any one of claims 64 to 71, further comprising:

The network device receives a first coding model, and the first coding model includes a channel estimation sub-model, a generation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The generating sub-model and the compressing sub-model constitute a third model.
The method according to claim 77, wherein the first coding model is carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, uplink data for artificial intelligence business transmission requirements transmission.
A method according to any one of claims 64 to 71, further comprising:

The network device receives a second coding model, and the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.
The method according to claim 79, wherein the second coding model is carried by one of the following: uplink control signaling, MAC CE message, RRC message, broadcast message, uplink data transmission, uplink data for artificial intelligence business transmission requirements transmission.
A method according to any one of claims 64 to 71, further comprising:

The network device uses the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.
The method according to claim 81, wherein the network device uses the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model, comprising:

The network device inputs the input information into the first initial model to obtain a first set output by the first initial model, where the first set includes a plurality of first reference signals;

inputting any one of the first reference signals in the first set to the first channel simulation module to obtain a second reference signal;

inputting the second reference signal into the second initial model to obtain channel information;

determining a first loss function based on at least one of the first set, the second reference signal, the channel information, and parameters in the first channel simulation module;

updating the first initial model and the second initial model according to the first loss function.
The method of claim 82, wherein said determining a first loss function comprises:

Determine the channel information and the parameters of the first channel simulation module based on at least one of the first set, the second reference signal, the channel information and the parameters of the first channel simulation module degree of variance and/or reference signal quality;

The first loss function is determined based on the degree of difference between the channel information and the parameters of the first channel simulation module and/or the quality of a reference signal.
A method according to any one of claims 64 to 71, further comprising:

The network device uses at least one of the input information, the first channel simulation module, and the second channel simulation module to jointly train the first initial model, the second initial model, the third initial model, and the fourth initial model, and obtain The trained first model, the second model, the third model and the fourth model.
The method according to claim 84, wherein the network device uses at least one of the input information, the first channel simulation module, and the second channel simulation module to perform the first initial model, the second initial model, the third initial model, the fourth initial model for joint training, including:

The network device inputs the input information into the first initial model to obtain a first set output by the first initial model, where the first set includes a plurality of first reference signals;

inputting any one of the first reference signals in the first set to the first channel simulation module to obtain a second reference signal;

inputting the second reference signal into the second initial model to obtain channel information;

inputting the channel information into the third initial model to obtain compressed information of the channel information; wherein, the third initial model includes generating an initial submodel and compressing an initial submodel, and the input of generating an initial submodel As the input of the third initial model, the output of the generated initial sub-model is used as the output of the compressed initial sub-model, and the output of the compressed initial sub-model is used as the output of the third initial model; or, the The third initial model includes a compressed initial sub-model;

Inputting the compressed information of the channel information into the second channel simulation module to obtain equivalent received information of the compressed information of the channel information;

inputting equivalent received information of the compressed information of the channel information into the fourth initial model to obtain output information of the fourth initial model;

Based on the first set, the second reference signal, the channel information, the parameters in the first channel simulation module, the input information of the compressed initial submodel, and the output information of the fourth initial model At least one item of determines the second loss function;

Updating the first initial model, the second initial model, the third initial model and the fourth initial model according to the second loss function.
The method of claim 85, wherein said determining a second loss function comprises:

Based on the first set, the second reference signal, the channel information, the parameters in the first channel simulation module, the input information of the compressed initial submodel, and the output information of the fourth initial model At least one of the parameters, determine the reference signal quality, the degree of difference between the channel information and the parameters of the first channel simulation module, the degree of difference between the output information of the fourth initial model and the input information of the compressed initial sub-model at least one of the

Based on at least one of the quality of the reference signal, the degree of difference between the channel information and the parameters of the first channel simulation module, and the degree of difference between the output information of the fourth initial model and the input information of the compressed initial sub-model One term, determining the second loss function.
The method according to claim 83 or 86, wherein the reference signal quality is represented by at least one of the following:

cross-correlations between different first reference signals in the first set;

cross-correlations between first reference signals in the first set and other reference signals;

The peak-to-average power ratio of the first reference signals in the first set.
The method according to claim 83 or 86, wherein the quality of the reference signal is represented by at least one of the following:

cross-correlation between different said second reference signals;

cross-correlation between the second reference signal and other reference signals;

The peak-to-average power ratio of the second reference signal.
The method according to claim 83 or 86, wherein the reference signal quality is represented by at least one of the following:

cross-correlation between different said third reference signals;

cross-correlation between the third reference signal and other reference signals;

a peak-to-average power ratio of the third reference signal;

Wherein, the third reference signal is obtained by processing the first reference signal based on the channel information.
The method according to claim 83 or 86, wherein the degree of difference is measured by distance and/or similarity.
The method according to any one of claims 81 to 90, wherein the input information includes at least one of the following: no input, noise, random numbers, sequences in a preset sequence set, channel type indication information, channel data samples Information, wireless channel or scene related information.
The method according to claim 91, wherein the preset sequence set includes at least one of the following: an m-sequence set, a golden sequence set, and a zc sequence set.
The method according to claim 91, wherein the channel type indication information indicates at least one of the following: frequency information corresponding to the channel, environment information corresponding to the channel, and scene information corresponding to the channel.
The method according to claim 91, wherein the wireless channel or scene-related information includes at least one of the following: channel signal-to-noise ratio, signal-to-interference-noise ratio, channel type, bandwidth information, and delay information.
The method according to claim 91, wherein the format of the noise, random number or sequence in the preset set of sequences is the same as the format of the first initial model output data.
The method according to claim 91 or 95, wherein the formats of the noise, random numbers, or sequences in the preset sequence set include at least one of the following formats: one-dimensional vector, two-dimensional matrix, and high-dimensional matrix.
The method according to claim 91, 95 or 96, wherein the format of the noise, the random number or the sequence in the preset sequence set is stipulated through a protocol or signaling.
The method according to any one of claims 91 to 97, wherein the input information is used to input at least one of the following: a first initial model, a first channel simulation module, and a second initial model.
The method according to any one of claims 82, 83, 85 or 86, wherein the channel information is distributed in the first dimension and/or the second dimension.
The method according to any one of claims 82, 83, 85 or 86, wherein the channel information is distributed in at least one of the first dimension, the second dimension and the third dimension.
A method according to claim 99 or 100, wherein,

The first dimension is a frequency domain dimension, and the channel information includes data distributed on M1 frequency domain granularities of the frequency domain dimension; the M1 is a positive integer.
The method according to claim 101, wherein the frequency domain granularity includes a RBs and/or b subcarriers, and a or b is a positive integer.
A method according to claim 99 or 100, wherein,

The first dimension is a time domain dimension, and the channel information includes data distributed on M2 delay granularities of the time domain dimension; the M2 is a positive integer.
The method according to claim 103, wherein the delay granularity includes at least one of the following: p1 microseconds, p2 symbol length, and p3 symbol sampling points, and the p1, p2 or p3 is positive integer.
The method of claim 104, wherein the symbols comprise OFDM symbols.
The method of claim 99 or 100, wherein the second dimension is a spatial domain dimension.
The method according to claim 106, wherein the spatial domain dimension is an antenna dimension, and the channel information includes data distributed on N1 first granularities of the antenna dimension, where N1 is a positive integer.
107. The method of claim 107, wherein the first granularity includes a pair of transceiving antennas.
The method according to claim 106, wherein the space domain dimension is an angle domain dimension, and the channel information includes data distributed on N2 second granularities of the angle domain dimension, where N2 is a positive integer.
The method of claim 109, wherein the second granularity includes angular spacing.
The method according to claim 100, wherein the third dimension includes a complex dimension, and the complex dimension includes 2 elements, which are respectively used to carry a real part and an imaginary part of the data included in the channel information.
The method according to any one of claims 100 to 111, wherein the channel information is distributed in a T-dimensional matrix, and the T-dimensional matrix is decomposed for at least one of the first dimension, the second dimension and the third dimension. The matrix formed after dividing and/or combining, said T is a positive integer.
The method according to any one of claims 82, 83, 85, 86 or 99 to 112, wherein the channel information includes S groups of characteristic sequences with a length U, and the S or U is a positive integer.
The method of claim 113, wherein S is 2, 4 or 8.
The method of claim 113 or 114, wherein U is 16, 32, 48, 64, 128 or 256.
A method according to any one of claims 81 to 115, further comprising:

The network device sends the second model.
The method of claim 116, further comprising:

The network device sends the first model.
A method according to any one of claims 84 to 115, further comprising:

The network device sends the second model and the third model.
The method of claim 118, further comprising:

The network device sends the first model and/or the fourth model.
The method of claim 119, wherein said first model, said second model, said third model, said fourth model, a submodel within said first model, said second model The submodel in the third model or the submodel in the fourth model is carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, for Downlink data transmission required for artificial intelligence business transmission.
A method according to any one of claims 81 to 115, further comprising:

The network device sends a first coding model, and the first coding model includes a channel estimation sub-model, a generation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The generating sub-model and the compressing sub-model constitute a third model.
The method according to claim 121, wherein the first coding model is carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, downlink data for artificial intelligence business transmission requirements transmission.
A method according to any one of claims 81 to 115, further comprising:

The network device sends a second coding model, and the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.
The method according to claim 123, wherein the second coding model is carried by one of the following: downlink control signaling, MAC CE message, RRC message, broadcast message, downlink data transmission, downlink data for artificial intelligence business transmission requirements transmission.
A model training method, comprising:

The input information and/or the first channel simulation module are used to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.
The method according to claim 125, wherein the joint training of the first initial model and the second initial model using the input information and/or the first channel simulation module comprises:

inputting the input information into the first initial model to obtain a first set of outputs of the first initial model, the first set including a plurality of first reference signals;

inputting any one of the first reference signals in the first set to the first channel simulation module to obtain a second reference signal;

inputting the second reference signal into the second initial model to obtain channel information;

determining a first loss function based on at least one of the first set, the second reference signal, the channel information, and parameters in the first channel simulation module;

updating the first initial model and the second initial model according to the first loss function.
The method of claim 126, wherein said determining a first loss function comprises:

Determine the channel information and the parameters of the first channel simulation module based on at least one of the first set, the second reference signal, the channel information and the parameters of the first channel simulation module degree of variance and/or reference signal quality;

The first loss function is determined based on the degree of difference between the channel information and the parameters of the first channel simulation module and/or the quality of a reference signal.
The method of claim 125, wherein said joint training comprises:

Using at least one of the input information, the first channel simulation module and the second channel simulation module, the first initial model, the second initial model, the third initial model, and the fourth initial model are jointly trained to obtain the trained first A model, a second model, a third model and a fourth model.
The method according to claim 128, wherein said using at least one of the input information, the first channel simulation module and the second channel simulation module, the first initial model, the second initial model, the third initial model, The fourth initial model is jointly trained, including:

inputting the input information into the first initial model to obtain a first set of outputs of the first initial model, the first set including a plurality of first reference signals;

inputting any one of the first reference signals in the first set to the first channel simulation module to obtain a second reference signal;

inputting the second reference signal into the second initial model to obtain channel information;

inputting the channel information into the third initial model to obtain compressed information of the channel information; wherein, the third initial model includes generating an initial submodel and compressing an initial submodel, and the input of generating an initial submodel As the input of the third initial model, the output of the generated initial sub-model is used as the output of the compressed initial sub-model, and the output of the compressed initial sub-model is used as the output of the third initial model; or, the The third initial model includes a compressed initial sub-model;

Inputting the compressed information of the channel information into the second channel simulation module to obtain equivalent received information of the compressed information of the channel information;

inputting equivalent received information of the compressed information of the channel information into the fourth initial model to obtain output information of the fourth initial model;

Based on the first set, the second reference signal, the channel information, the parameters in the first channel simulation module, the input information of the compressed initial submodel, and the output information of the fourth initial model At least one item of determines the second loss function;

Updating the first initial model, the second initial model, the third initial model and the fourth initial model according to the second loss function.
The method of claim 129, wherein said determining a second loss function comprises:

Based on the first set, the second reference signal, the channel information, the parameters in the first channel simulation module, the input information of the compressed initial submodel, and the output information of the fourth initial model At least one of the parameters, determine the reference signal quality, the degree of difference between the channel information and the parameters of the first channel simulation module, the degree of difference between the output information of the fourth initial model and the input information of the compressed initial sub-model at least one of the

Based on at least one of the quality of the reference signal, the degree of difference between the channel information and the parameters of the first channel simulation module, and the degree of difference between the output information of the fourth initial model and the input information of the compressed initial sub-model One term, determining the second loss function.
The method according to claim 127 or 130, wherein the reference signal quality is represented by at least one of the following:

cross-correlations between different first reference signals in the first set;

cross-correlations between first reference signals in the first set and other reference signals;

The peak-to-average power ratio of the first reference signals in the first set.
The method according to claim 127 or 130, wherein the quality of the reference signal is represented by at least one of the following:

cross-correlation between different said second reference signals;

cross-correlation between the second reference signal and other reference signals;

The peak-to-average power ratio of the second reference signal.
The method according to claim 127 or 130, wherein the reference signal quality is represented by at least one of the following:

cross-correlation between different said third reference signals;

cross-correlation between the third reference signal and other reference signals;

a peak-to-average power ratio of the third reference signal;

Wherein, the third reference signal is obtained by processing the first reference signal based on the channel information.
The method according to claim 127 or 130, wherein the degree of difference is measured by distance and/or similarity.
The method according to any one of claims 125 to 134, wherein the input information includes at least one of the following: no input, noise, random numbers, sequences in a preset sequence set, channel type indication information, channel data samples Information, wireless channel or scene related information.
The method according to claim 134, wherein the preset sequence set includes at least one of the following: m sequence set, golden sequence set, zc sequence set.
The method according to claim 134, wherein the channel type indication information indicates at least one of the following: frequency information corresponding to the channel, environment information corresponding to the channel, and scene information corresponding to the channel.
The method according to claim 134, wherein the wireless channel or scene-related information includes at least one of the following: channel signal-to-noise ratio, signal-to-interference-noise ratio, channel type, bandwidth information, and delay information.
The method of claim 134, wherein the format of the noise, random number, or sequence in the preset set of sequences is the same as the format of the first initial model output data.
The method according to claim 135 or 139, wherein the format of the noise, the random number or the sequence in the preset sequence set includes at least one of the following formats: a one-dimensional vector, a two-dimensional matrix, and a high-dimensional matrix.
The method according to claim 135, 139 or 140, wherein the format of the noise, the random number or the sequence in the preset sequence set is stipulated through a protocol or signaling.
The method according to any one of claims 125 to 141, wherein the input information is used to input at least one of the following: a first initial model, a first channel simulation module, and a second initial model.
The method of claim 126, 127, 129 or 130, wherein the channel information is distributed in the first dimension and/or the second dimension.
The method of claim 126, 127, 129 or 130, wherein the channel information is distributed in at least one of a first dimension, a second dimension and a third dimension.
A method according to claim 143 or 144, wherein,

The first dimension is a frequency domain dimension, and the channel information includes data distributed on M1 frequency domain granularities of the frequency domain dimension; the M1 is a positive integer.
The method according to claim 145, wherein the frequency domain granularity includes a RBs and/or b subcarriers, and a or b is a positive integer.
A method according to claim 143 or 144, wherein,

The first dimension is a time domain dimension, and the channel information includes data distributed on M2 delay granularities of the time domain dimension; the M2 is a positive integer.
The method according to claim 147, wherein the delay granularity includes at least one of the following: p1 microseconds, p2 symbol length, p3 symbol sampling points, and the p1, p2 or p3 is positive integer.
The method of claim 87, wherein the symbols comprise OFDM symbols.
The method of claim 143 or 144, wherein the second dimension is a spatial domain dimension.
The method according to claim 150, wherein the spatial domain dimension is an antenna dimension, and the channel information includes data distributed on N1 first granularities of the antenna dimension, where N1 is a positive integer.
151. The method of claim 151, wherein the first granularity includes a pair of transceiving antennas.
The method according to claim 150, wherein the space domain dimension is an angle domain dimension, and the channel information includes data distributed on N2 second granularities of the angle domain dimension, where N2 is a positive integer.
153. The method of claim 153, wherein the second granularity includes angular spacing.
The method according to claim 144, wherein the third dimension includes a complex dimension, and the complex dimension includes 2 elements, which are respectively used to bear the real part and the imaginary part of the data included in the channel information.
The method according to any one of claims 144 to 155, wherein the channel information is distributed in a T-dimensional matrix, and the T-dimensional matrix is decomposed for at least one of the first dimension, the second dimension and the third dimension. The matrix formed after dividing and/or combining, said T is a positive integer.
The method according to any one of claims 143 to 156, wherein the channel information includes S groups of characteristic sequences with a length of U, and the S or U is a positive integer.
The method of claim 157, wherein S is 2, 4 or 8.
The method of claim 157 or 158, wherein U is 16, 32, 48, 64, 128 or 256.
A terminal device comprising:

A first receiving module, configured to receive a first signal, the first signal is generated by a first model;

A first processing module, configured to process the first signal using a second model to obtain first information; the first information includes channel information;

Wherein, the first model and the second model are obtained through joint training.
The terminal device according to claim 160, further comprising:

a second processing module, configured to process the first information by using a third model to obtain second information;

The first sending module is configured to send the second information, and the second information is used for processing by the fourth model to obtain third information;

Wherein, the first model, the second model, the third model and the fourth model are obtained through joint training.
A terminal device according to claim 160 or 161, wherein the first signal comprises a reference signal.
The terminal device according to any one of claims 160 to 162, wherein the second model comprises a channel estimation sub-model;

The channel estimation sub-model is used to perform channel estimation based on the first signal to obtain channel information.
The terminal device according to claim 161 or 162, wherein the third model comprises a compressed sub-model;

The compression sub-model is used to compress the first information to obtain compressed information of the first information; the second information includes the compressed information of the first information.
The terminal device of claim 164, wherein the fourth model comprises a recovery sub-model;

The restoration sub-model is used to restore the compressed information of the first information to obtain the restoration information of the first information; the third information includes the restoration information of the first information.
The terminal device according to claim 161 or 162, wherein the third model includes a generation sub-model and a compression sub-model; wherein,

The generation sub-model is used to perform feature transformation on the first information to obtain a first feature vector corresponding to the first information;

The compression sub-model is used to compress the first feature vector to obtain compressed information of the first feature vector; the second information includes the compressed information of the first feature vector.
The terminal device of claim 166, wherein the fourth model comprises a recovery sub-model;

The restoration sub-model is used to restore the compressed information of the first feature vector to obtain the restoration information of the first feature vector; the third information includes the restoration information of the first feature vector.
A terminal device according to any one of claims 160 to 167, further comprising:

The second receiving module is configured to receive the second model.
The terminal device according to claim 168, wherein:

The second receiving module is further configured to receive the first model.
A terminal device according to any one of claims 160 to 167, further comprising:

A third receiving module, configured to receive the second model and the third model.
The terminal device according to claim 170, wherein,

The third receiving module is further configured to receive the first model and/or the fourth model.
The terminal device according to claim 161, further comprising:

The fourth receiving module is configured to receive the first coding model, the first coding model includes a channel estimation sub-model, a generation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes said second model;

The generative sub-model and the compressed sub-model constitute the third model.
The terminal device according to claim 161, further comprising:

The fifth receiving module is configured to receive a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes said second model;

The compressed sub-models constitute the third model.
The terminal device according to any one of claims 160 to 173, further comprising:

The first training module is configured to use the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.
The terminal device according to any one of claims 160 to 173, further comprising:

The second training module is used to combine the first initial model, the second initial model, the third initial model, and the fourth initial model by using at least one of the input information, the first channel simulation module, and the second channel simulation module training to obtain the trained first model, the second model, the third model and the fourth model.
A terminal device according to claim 174 or 175, further comprising:

A second sending module, configured to send the first model.
The terminal device according to claim 176, wherein,

The second sending module is further configured to send the second model.
The method of claim 174 or 175, further comprising:

A third sending module, configured to send the first model and the fourth model.
The terminal device according to claim 178, wherein,

The third sending module is further configured to send the second model and/or the third model.
A terminal device according to claim 174 or 175, further comprising:

The fourth sending module is used to send the first coding model, the first coding model includes a channel estimation sub-model, a generation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The generating sub-model and the compressing sub-model constitute a third model.
A terminal device according to claim 174 or 175, further comprising:

A fifth sending module, configured to send a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.
A network device comprising:

A sixth sending module, configured to send a first signal, the first signal is generated by the first model; the first signal is used for processing by the second model to obtain first information; the first information includes channel information ;

Wherein, the first model and the second model are obtained through joint training.
The network device of claim 182, further comprising:

A sixth receiving module, configured to receive second information, the second information is obtained by processing the first information by the third model;

a third processing module, configured to process the second information by using a fourth model to obtain third information;

Wherein, the first model, the second model, the third model and the fourth model are obtained through joint training.
The network device of claim 182 or 183, wherein the first signal comprises a reference signal.
The network device according to any one of claims 182 to 184, wherein the second model comprises a channel estimation sub-model;

The channel estimation sub-model is used to perform channel estimation based on the first signal to obtain channel information.
A network device according to claim 183 or 184, wherein said third model comprises a compressed sub-model;

The compression sub-model is used to compress the first information to obtain compressed information of the first information; the second information includes the compressed information of the first information.
The network device of claim 186, wherein the fourth model comprises a recovery sub-model;

The restoration sub-model is used to restore the compressed information of the first information to obtain the restoration information of the first information; the third information includes the restoration information of the first information.
The network device according to claim 183 or 184, wherein the third model comprises a generation sub-model and a compression sub-model; wherein,

The generation sub-model is used to perform feature transformation on the first information to obtain a first feature vector corresponding to the first information;

The compression sub-model is used to compress the first feature vector to obtain compressed information of the first feature vector; the second information includes the compressed information of the first feature vector.
The network device of claim 188, wherein the fourth model comprises a restoration sub-model;

The restoration sub-model is used to restore the compressed information of the first feature vector to obtain the restoration information of the first feature vector; the third information includes the restoration information of the first feature vector.
A network device according to any one of claims 182 to 189, further comprising:

A seventh receiving module, configured to receive the first model.
The network device of claim 190, wherein,

The seventh receiving module is further configured to receive the second model.
A network device according to any one of claims 182 to 189, further comprising:

An eighth receiving module, configured to receive the first model and the fourth model.
The network device of claim 192, wherein,

The eighth receiving module is further configured to receive the second model and/or the third model.
A network device according to any one of claims 182 to 189, further comprising:

A ninth receiving module, configured to receive a first coding model, where the first coding model includes a channel estimation sub-model, a generation sub-model, and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The generating sub-model and the compressing sub-model constitute a third model.
A network device according to any one of claims 182 to 189, further comprising:

A tenth receiving module, configured to receive a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.
A network device according to any one of claims 182 to 189, further comprising:

The third training module is configured to use the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.
A network device according to any one of claims 182 to 189, further comprising:

The fourth training module is used to combine the first initial model, the second initial model, the third initial model, and the fourth initial model by using at least one of the input information, the first channel simulation module, and the second channel simulation module training to obtain the trained first model, the second model, the third model and the fourth model.
A network device according to claim 196 or 197, further comprising:

A seventh sending module, configured to send the second model.
The network device of claim 198, wherein,

The seventh sending module is further configured to send the first model.
A network device according to claim 196 or 197, further comprising:

An eighth sending module, configured to send the second model and the third model.
The network device of claim 200, wherein:

The eighth sending module is further configured to send the first model and/or the fourth model.
A network device according to claim 196 or 197, further comprising:

A ninth sending module, configured to send a first coding model, where the first coding model includes a channel estimation sub-model, a generation sub-model, and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The generating sub-model and the compressing sub-model constitute a third model.
A network device according to claim 196 or 197, further comprising:

A tenth sending module, configured to send a second coding model, where the second coding model includes a channel estimation sub-model and a compression sub-model; wherein,

said channel estimation sub-model constitutes a second model;

The compressed sub-model constitutes a third model.
A model training device, comprising:

The joint training module is configured to use the input information and/or the first channel simulation module to jointly train the first initial model and the second initial model to obtain the trained first model and the second model.
The apparatus of claim 204, wherein the joint training module is configured to:

inputting the input information into the first initial model to obtain a first set of outputs of the first initial model, the first set including a plurality of first reference signals;

inputting any one of the first reference signals in the first set to the first channel simulation module to obtain a second reference signal;

inputting the second reference signal into the second initial model to obtain channel information;

determining a first loss function based on at least one of the first set, the second reference signal, the channel information, and parameters in the first channel simulation module;

updating the first initial model and the second initial model according to the first loss function.
The apparatus of claim 205, wherein said determining a first loss function comprises:

Determine the channel information and the parameters of the first channel simulation module based on at least one of the first set, the second reference signal, the channel information and the parameters of the first channel simulation module degree of variance and/or reference signal quality;

The first loss function is determined based on the degree of difference between the channel information and the parameters of the first channel simulation module and/or the quality of a reference signal.
The apparatus of claim 204, wherein the joint training module is configured to:

Using at least one of the input information, the first channel simulation module and the second channel simulation module, the first initial model, the second initial model, the third initial model, and the fourth initial model are jointly trained to obtain the trained first A model, a second model, a third model and a fourth model.
The apparatus of claim 207, wherein the joint training module is configured to:

inputting the input information into the first initial model to obtain a first set of outputs of the first initial model, the first set including a plurality of first reference signals;

inputting any one of the first reference signals in the first set to the first channel simulation module to obtain a second reference signal;

inputting the second reference signal into the second initial model to obtain channel information;

inputting the channel information into the third initial model to obtain compressed information of the channel information; wherein, the third initial model includes generating an initial submodel and compressing an initial submodel, and the input of generating an initial submodel As the input of the third initial model, the output of the generated initial sub-model is used as the output of the compressed initial sub-model, and the output of the compressed initial sub-model is used as the output of the third initial model; or, the The third initial model includes a compressed initial sub-model;

Inputting the compressed information of the channel information into the second channel simulation module to obtain equivalent received information of the compressed information of the channel information;

inputting equivalent received information of the compressed information of the channel information into the fourth initial model to obtain output information of the fourth initial model;

Based on the first set, the second reference signal, the channel information, the parameters in the first channel simulation module, the input information of the compressed initial submodel, and the output information of the fourth initial model At least one item of determines the second loss function;

Updating the first initial model, the second initial model, the third initial model and the fourth initial model according to the second loss function.
The apparatus of claim 208, wherein said determining a second loss function comprises:

Based on the first set, the second reference signal, the channel information, the parameters in the first channel simulation module, the input information of the compressed initial submodel, and the output information of the fourth initial model At least one of the parameters, determine the reference signal quality, the degree of difference between the channel information and the parameters of the first channel simulation module, the degree of difference between the output information of the fourth initial model and the input information of the compressed initial sub-model at least one of the

Based on at least one of the quality of the reference signal, the degree of difference between the channel information and the parameters of the first channel simulation module, and the degree of difference between the output information of the fourth initial model and the input information of the compressed initial sub-model One term, determining the second loss function.
The device according to claim 206 or 209, wherein the reference signal quality is represented by at least one of the following:

cross-correlations between different first reference signals in the first set;

cross-correlations between first reference signals in the first set and other reference signals;

The peak-to-average power ratio of the first reference signals in the first set.
The device according to claim 206 or 209, wherein the quality of the reference signal is represented by at least one of the following:

cross-correlation between different said second reference signals;

cross-correlation between the second reference signal and other reference signals;

The peak-to-average power ratio of the second reference signal.
The device according to claim 206 or 209, wherein the reference signal quality is represented by at least one of the following:

cross-correlation between different said third reference signals;

cross-correlation between the third reference signal and other reference signals;

a peak-to-average power ratio of the third reference signal;

Wherein, the third reference signal is obtained by processing the first reference signal based on the channel information.
A terminal device, including: a processor, a memory, and a transceiver, the memory is used to store computer programs, the processor is used to call and run the computer programs stored in the memory, and control the transceiver to execute such as The method of any one of claims 1 to 63.
A network device, comprising: a processor, a memory, and a transceiver, the memory is used to store computer programs, the processor is used to invoke and run the computer programs stored in the memory, and control the transceiver to execute such as The method of any one of claims 64 to 124.
A model training device, comprising: a processor and a memory, the memory is used to store a computer program, the processor is used to call and run the computer program stored in the memory, and execute the computer program described in any one of claims 125 to 159 described method.
A chip, comprising: a processor, configured to invoke and run a computer program from a memory, so that a device equipped with the chip executes the method as claimed in any one of claims 125 to 159.
A chip, comprising: a processor, configured to invoke and run a computer program from a memory, so that a device equipped with the chip executes the method as claimed in any one of claims 64 to 124.
A chip, comprising: a processor, configured to invoke and run a computer program from a memory, so that a device equipped with the chip executes the method as claimed in any one of claims 125 to 159.
A computer-readable storage medium for storing a computer program, the computer program causing a computer to execute the method according to any one of claims 1 to 63.
A computer-readable storage medium for storing a computer program, the computer program causing a computer to execute the method according to any one of claims 64-124.
A computer-readable storage medium for storing a computer program, the computer program causing a computer to execute the method according to any one of claims 125-159.
A computer program product comprising computer program instructions for causing a computer to perform the method as claimed in any one of claims 1 to 63.
A computer program product comprising computer program instructions for causing a computer to perform the method as claimed in any one of claims 64 to 124.
A computer program product comprising computer program instructions for causing a computer to perform the method as claimed in any one of claims 125 to 159.
A computer program that causes a computer to perform the method as claimed in any one of claims 1 to 63.
A computer program that causes a computer to perform the method as claimed in any one of claims 64 to 124.
A computer program that causes a computer to perform the method as claimed in any one of claims 125 to 159.