WO2024051594A1

WO2024051594A1 - Information transmission method and apparatus, ai network model training method and apparatus, and communication device

Info

Publication number: WO2024051594A1
Application number: PCT/CN2023/116504
Authority: WO
Inventors: 任千尧
Original assignee: 维沃移动通信有限公司
Priority date: 2022-09-07
Filing date: 2023-09-01
Publication date: 2024-03-14
Also published as: CN117715072A

Abstract

The present application relates to the technical field of communications, and discloses an information transmission method and apparatus, an AI network model training method and apparatus, and a communication device. The information transmission method of embodiments of the present application comprises: a terminal obtains first information, wherein the first information comprises first channel information of all sub-bands of a target downlink channel; the terminal performs first processing on the first information to obtain AI training data, wherein the first processing comprises at least one of: screening processing, quantization processing, delay domain conversion processing, codebook conversion processing, and orthogonalization processing, the AI training data is used for training a first AI network model and/or a second AI network model, the first AI network model is used for processing second channel information into first channel feature information, the second AI network model is used for recovering the first channel feature information into the second channel information, and the second channel information is channel information of the target downlink channel; and the terminal sends second information to a first device, wherein the second information comprises the AI training data.

Description

Information transmission method, AI network model training method, device and communication equipment

Cross-references to related applications

This application claims priority from Chinese Patent Application No. 202211091643.2 filed in China on September 7, 2022, the entire content of which is incorporated herein by reference.

Technical field

This application belongs to the field of communication technology, and specifically relates to an information transmission method, artificial intelligence (Artificial Intelligence, AI) network model training method, device and communication equipment.

Background technique

In related technologies, methods for transmitting channel characteristic information using AI network models have been studied.

The AI network model may include an encoding part (i.e., encoding AI network model) and a decoding part (i.e., decoding AI network model). The encoding AI network model is used to encode channel information into channel feature information, and the decoding AI network model is used to convert the encoding AI network model into The channel characteristic information output by the network model is restored to channel information. In this way, in order to match the encoding AI network model and the decoding AI network model, the encoding AI network model and the decoding AI network model are usually jointly trained in the same device, and then jointly trained The obtained encoded AI network model is transmitted to the terminal, and the decoded AI network model obtained through joint training is transmitted to the base station.

In related technologies, the base station trains the encoding AI network model and the decoding AI network model based on pre-acquired training data. Since the pre-acquired training data is inconsistent with the actual channel, the trained encoding AI network model and decoding AI network model will be The matching degree with the actual channel is relatively low, thus reducing the accuracy of the encoding AI network model and the decoding AI network model in processing actual channel information.

Contents of the invention

Embodiments of the present application provide an information transmission method, AI network model training method, device and communication equipment, enabling the terminal to report actual estimated channel information to the base station, so that the base station can train a coding AI network with higher accuracy based on the channel information. Model and decode AI network models.

In the first aspect, an information transmission method is provided, which method includes:

The terminal acquires first information, where the first information includes first channel information of all subbands of the target downlink channel;

The terminal performs a first process on the first information to obtain AI training data, where the first process includes at least one of the following: screening processing, quantization processing, delay domain conversion processing, codebook conversion processing, and normalization processing. Cross-processing, the AI training data is used to train the first AI network model and/or the second AI network model, the first AI network model is used to process the second channel information into the first channel feature information, the The second AI network model is used to restore the first channel characteristic information to the second channel information, where the second channel information is the channel information of the target downlink channel;

The terminal sends second information to the first device, where the second information includes the AI training data.

In a second aspect, an information transmission device is provided, applied to a terminal, and the device includes:

An acquisition module, configured to acquire first information, where the first information includes first channel information of all subbands of the target downlink channel;

A first processing module, configured to perform first processing on the first information to obtain AI training data, where the first processing includes at least one of the following: screening processing, quantization processing, delay domain conversion processing, codebook Conversion processing and orthogonalization processing, the AI training data is used to train the first AI network model and/or the second AI network model, the first AI network model is used to process the second channel information into the first channel characteristics Information, the second AI network model is used to restore the first channel characteristic information to the second channel information, and the second channel information is the channel information of the target downlink channel;

A first sending module, configured to send second information to the first device, where the second information includes the AI training data.

In the third aspect, an AI network model training method is provided, including:

The first device receives second information from the terminal, wherein the second information includes AI training data, the AI training data is obtained based on first processing of the first information, and the first information includes all of the target downlink channels. The first channel information of the subband;

The first device trains a first AI network model and/or a second AI network model according to the AI training data. The first AI network model is used to process the second channel information into the first channel characteristic information. The second AI network model is used to restore the first channel characteristic information to the second channel information, where the second channel information is the channel information of the target downlink channel.

In the fourth aspect, an AI network model training device is provided, which is applied to network-side equipment. The device includes:

A first receiving module, configured to receive second information from the terminal, where the second information includes AI training data, the AI training data is obtained based on first processing of the first information, and the first information includes a target The first channel information of all subbands of the downlink channel;

A training module, configured to train a first AI network model and/or a second AI network model according to the AI training data, the first AI network model being used to process the second channel information into the first channel feature information, the The second AI network model is used to restore the first channel characteristic information to the second channel information, where the second channel information is the channel information of the target downlink channel.

In a fifth aspect, a communication device is provided. The communication device includes a processor and a memory. The memory stores a program or instructions that can be run on the processor. The program or instructions are implemented when executed by the processor. The steps of the method described in the first aspect or the third aspect.

In a sixth aspect, a communication device is provided, including a processor and a communication interface, wherein the communication interface is used to obtain first information, where the first information includes first channel information of all subbands of a target downlink channel; The processor is configured to perform first processing on the first information to obtain AI training data, wherein the first processing includes at least one of the following: screening processing, quantization processing, delay domain conversion processing, and codebook conversion processing and orthogonalization processing, the AI training data is used to train the first AI network model and/or the second AI network model, the first AI network model is used to train The second channel information is processed into first channel characteristic information. The second AI network model is used to restore the first channel characteristic information into the second channel information. The second channel information is the target downlink channel. channel information; the communication interface is also used to send second information to the first device, where the second information includes the AI training data;

or,

The communication interface is used to receive second information from the terminal, wherein the second information includes AI training data, the AI training data is obtained based on first processing of the first information, and the first information includes target downlink First channel information of all subbands of the channel; the processor is used to train a first AI network model and/or a second AI network model according to the AI training data, and the first AI network model is used to convert the second channel The information is processed into first channel characteristic information, and the second AI network model is used to restore the first channel characteristic information into the second channel information. The second channel information is the channel information of the target downlink channel. .

In a seventh aspect, a communication system is provided, including: a terminal and a first device. The terminal can be used to perform the steps of the information transmission method as described in the first aspect. The first device can be used to perform the steps of the information transmission method as described in the third aspect. The steps of the AI network model training method.

In an eighth aspect, a readable storage medium is provided. Programs or instructions are stored on the readable storage medium. When the programs or instructions are executed by a processor, the steps of the method described in the first aspect are implemented, or the steps of the method are implemented as described in the first aspect. The steps of the method described in the third aspect.

In a ninth aspect, a chip is provided. The chip includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the method described in the first aspect. , or implement the method as described in the third aspect.

In a tenth aspect, a computer program/program product is provided, the computer program/program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to implement the method as described in the first aspect The steps of the information transmission method, or the computer program/program product is executed by at least one processor to implement the steps of the AI network model training method as described in the third aspect.

In this embodiment of the present application, the terminal obtains first information, which includes first channel information of all subbands of the target downlink channel; the terminal performs first processing on the first information to obtain AI training data , wherein the first processing includes at least one of the following: screening processing, quantization processing, delay domain conversion processing, codebook conversion processing and orthogonalization processing, and the AI training data is used to train the first AI network model and / Or a second AI network model, the first AI network model is used to process the second channel information into the first channel characteristic information, and the second AI network model is used to restore the first channel characteristic information to the first channel characteristic information. The second channel information is the channel information of the target downlink channel; the terminal sends the second information to the first device, and the second information includes the AI training data. After estimating and obtaining the first channel information of each subband of the target downlink channel, the terminal performs at least one of screening processing, quantization processing, delay domain conversion processing, codebook conversion processing, and orthogonalization processing on the channel information. The item can reduce the data amount of the obtained AI training data, and when the terminal sends the AI training data to the first device, the amount of resources occupied by transmitting the AI training data can be reduced. In addition, since the AI training data is obtained based on the first processing of channel information of all subbands of the target downlink channel, the AI training data can reflect the channel status of the target downlink channel, In the process of the first device training the encoding AI network model (i.e., the first AI network model) and/or the decoding AI network model (i.e., the second AI network model) based on the AI training data, the trained encoding AI network model can be improved And/or the degree of matching between the decoding AI network model and the target downlink channel. In the process of using the first AI network model to compress the channel information of the target downlink channel into channel feature information, and/or using the second AI network model to restore the channel feature information of the target downlink channel into channel information, the first step can be improved. The compression coding accuracy of the AI network model and/or the decoding accuracy of the second AI network model can be improved and/or the matching degree between the first AI network model and the second AI network model can be improved, thereby reducing the occupancy of the channel information reporting process. resources, and improve the accuracy of the reporting process.

Description of the drawings

Figure 1 is a schematic structural diagram of a wireless communication system to which embodiments of the present application can be applied;

Figure 2 is a flow chart of an information transmission method provided by an embodiment of the present application;

Figure 3 is a flow chart of an AI network model training method provided by an embodiment of the present application;

Figure 4 is a schematic structural diagram of an information transmission device provided by an embodiment of the present application;

Figure 5 is a schematic structural diagram of an AI network model training device provided by an embodiment of the present application;

Figure 6 is a schematic structural diagram of a communication device provided by an embodiment of the present application;

Figure 7 is a schematic diagram of the hardware structure of a terminal provided by an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art fall within the scope of protection of this application.

The terms "first", "second", etc. in the description and claims of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. It is to be understood that the terms so used are interchangeable under appropriate circumstances so that the embodiments of the present application can be practiced in sequences other than those illustrated or described herein, and that "first" and "second" are distinguished objects It is usually one type, and the number of objects is not limited. For example, the first object can be one or multiple. In addition, "and/or" in the description and claims indicates at least one of the connected objects, and the character "/" generally indicates that the related objects are in an "or" relationship.

It is worth pointing out that the technology described in the embodiments of this application is not limited to Long Term Evolution (LTE)/LTE Evolution (LTE-Advanced, LTE-A) systems, and can also be used in other wireless communication systems, such as code Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access, OFDMA), Single-carrier Frequency Division Multiple Access (SC-FDMA) and other systems. The terms "system" and "network" in the embodiments of this application are often used interchangeably, and the described technology can be used not only for the above-mentioned systems and radio technologies, but also for other systems and radio technologies. The following description describes a New Radio (NR) system for example purposes, And NR terminology is used in most of the following descriptions, but these technologies can also be applied to applications other than NR system applications, such as 6th Generation ( ^6th Generation, 6G) communication systems.

Figure 1 shows a block diagram of a wireless communication system to which embodiments of the present application are applicable. The wireless communication system includes a terminal 11 and a network side device 12. Among them, the terminal 11 can be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer), or a notebook computer, a personal digital assistant (Personal Digital Assistant, PDA), a handheld computer, a netbook, or a super mobile personal computer. (ultra-mobile personal computer, UMPC), mobile Internet device (Mobile Internet Device, MID), augmented reality (augmented reality, AR)/virtual reality (VR) equipment, robots, wearable devices (Wearable Device) , vehicle user equipment (VUE), pedestrian terminal (Pedestrian User Equipment, PUE), smart home (home equipment with wireless communication functions, such as refrigerators, TVs, washing machines or furniture, etc.), game consoles, personal computers (personal computer, PC), teller machine or self-service machine and other terminal-side devices. Wearable devices include: smart watches, smart bracelets, smart headphones, smart glasses, smart jewelry (smart bracelets, smart bracelets, smart rings, smart necklaces, smart anklets) bracelets, smart anklets, etc.), smart wristbands, smart clothing, etc. It should be noted that the embodiment of the present application does not limit the specific type of the terminal 11. The network side equipment 12 may include access network equipment or core network equipment, where the access network equipment may also be called wireless access network equipment, radio access network (Radio Access Network, RAN), radio access network function or wireless access network unit. Access network equipment can include base stations, Wireless Local Area Network (WLAN) access points or WiFi nodes, etc. The base station can be called Node B, Evolved Node B (eNB), access point, base transceiver station ( Base Transceiver Station (BTS), radio base station, radio transceiver, Basic Service Set (BSS), Extended Service Set (ESS), home B-node, home evolved B-node, sending and receiving point ( Transmitting Receiving Point (TRP) or some other suitable term in the field, as long as the same technical effect is achieved, the base station is not limited to specific technical terms. It should be noted that in the embodiment of this application, only the NR system is used The base station is introduced as an example, and the specific type of base station is not limited.

Artificial intelligence is currently widely used in various fields. There are many ways to implement AI network models, such as neural networks, decision trees, support vector machines, Bayesian classifiers, etc. This application takes a neural network as an example for explanation, but does not limit the specific type of AI network model.

In related technologies, the encoding AI network model is used on the terminal side to compress and encode the channel information, and the compressed and encoded channel characteristic information is reported. The decoding AI network model can be used on the network side to decode the received channel characteristic information. processing to recover channel information. Among them, the encoding AI network model needs to match the decoding AI network model. In related technologies, the encoding AI network model and the decoding AI network model are usually jointly trained on the network side to achieve an appropriate matching degree between the two.

Of course, in the embodiment of the present application, it is also possible that the terminal already has a decoding AI network model. In this case, the decoding AI network model can be trained on the network side so that the trained decoding AI network model matches the decoding AI network model of the terminal. Here The process of training the encoding AI network model and/or the decoding AI network model in the embodiments of the present application is not specifically limited.

In related technologies, AI training data for training encoding AI network models and decoding AI network models are usually pre- The acquired data is either channel information collected by a terminal or network-side device dedicated to collecting training data. After training the encoding AI network model and decoding AI network model based on the AI training data, the network-side device will then encode the AI The network model is delivered to the terminal, so that the terminal encodes channel information based on the coding AI network model delivered by the network side device. However, the encoding AI network model and decoding AI network model trained by the network side device based on pre-acquired data or channel information collected by a terminal or network side device dedicated to collecting training data may be different from the actual channel information of the terminal. In the case of matching, at this time, it may happen that the encoding AI network model has a low degree of compression of the actual channel information of the terminal, causing the compressed channel characteristic information to occupy a large amount of resources, or the encoding AI network model and the decoding AI network model have a negative impact on the terminal. The processing of actual channel information is not accurate enough, which reduces the accuracy of channel information reporting.

In the embodiment of this application, the terminal will perform filtering processing, quantization processing, delay domain conversion processing, codebook conversion processing and orthogonalization processing on the estimated channel information of all subbands of the actual channel, so that the processed channel The information can reduce the number of bits as much as possible while accurately describing the channel status of the target downlink channel. In this way, the terminal reports the processed channel information as AI training data to the training encoding AI network model (ie, the first AI network model in the embodiment of the present application) and/or the decoding AI network model (ie, the first AI network model in the embodiment of the present application). second AI network model), the cost of transmitting the AI training data can be reduced, and the encoding AI network model and/or the decoding AI network model trained by the first device based on the AI training data can be consistent with the terminal's actual The channel status matches, thereby reducing the resources occupied by the channel information reporting process and improving the accuracy of the channel information reporting process.

The following is a detailed description of the information transmission method, AI network model training method, information transmission device, AI network model training device, communication equipment, etc. provided by the embodiments of the present application through some embodiments and application scenarios with reference to the accompanying drawings.

Referring to Figure 2, an information transmission method provided by an embodiment of the present application is executed by a terminal, such as various types of terminals 11 listed in Figure 1. As shown in Figure 2, the information transmission method performed by the terminal may include the following steps:

Step 201: The terminal acquires first information, where the first information includes first channel information of all subbands of the target downlink channel.

The first channel information may include at least one of a channel matrix and a precoding matrix. For convenience of explanation, in the embodiment of the present application, the first channel information is a precoding matrix as an example for illustration. This does not constitute a detailed description. limited.

Step 202: The terminal performs a first process on the first information to obtain AI training data, where the first process includes at least one of the following: screening processing, quantization processing, delay domain conversion processing, and codebook conversion. Processing and orthogonalization processing, the AI training data is used to train the first AI network model and/or the second AI network model, the first AI network model is used to process the second channel information into the first channel characteristic information , the second AI network model is used to restore the first channel characteristic information to the second channel information, where the second channel information is the channel information of the target downlink channel.

Step 203: The terminal sends second information to the first device, where the second information includes the AI training data.

Among them, the above-mentioned first processing is mainly used to reduce the number of bits of the AI training data, thereby reducing the amount of resources occupied when transmitting the AI training data.

In a possible implementation, the terminal performs filtering processing on the first information. The terminal may filter out non-zero elements or elements whose amplitude is greater than a preset value (such as 0.5) in the precoding matrix of each subband, and Use the amplitude, phase, position and other information of the filtered elements as AI training data.

It is worth mentioning that when there are no zero elements in the precoding matrix, that is, when all elements are non-zero elements, the position information of each element does not need to be reported.

In a possible implementation, the terminal performs quantization processing on the first information. The terminal may perform quantization processing on the amplitude and phase of the non-zero elements in the precoding matrix of each subband, where the quantization processing may include: normalization Unified processing, elements with amplitudes smaller than the minimum value of amplitude quantization are treated as zero elements. For high-rank (Rank) target downlink channels, based on the principle of orthogonality between precoding matrices of each layer (layer), the actual reported The number of elements of the precoding matrix. Among them, for the high-rank target downlink channel, the precoding matrix of each subband has multiple columns, each column represents a layer, and each layer has N elements. By reducing the number of elements of the actually reported precoding matrix , it can be that the number of elements corresponding to the same layer in the precoding matrix can be less than or equal to (N-1).

In a possible implementation, the terminal performs delay domain conversion processing on the first information, which may be that the terminal converts the precoding matrix of each subband into the delay domain, and selects at least one precoding matrix with a larger amplitude in the delay domain. Prepare matrix for reporting.

It is worth noting that after converting the precoding matrix of each subband into the delay domain, at least one of the above screening processing and quantization processing can be performed on the precoding matrix in the delay domain, and the processed result can be reported. precoding matrix.

In a possible implementation manner, the terminal performs codebook conversion processing on the first information. The terminal may use a high-precision codebook to encode the precoding matrix of each subband, and report the encoded codebook information.

The high-precision codebook may represent a codebook with higher accuracy than the R15 codebook or R16 codebook in related technologies. For example, the AI training data includes a codebook for each subband based on the first codebook. The codebook data obtained by encoding the precoding matrix, at least one of the number of ports, the number of delay paths, the number of beams, and the non-zero coefficient ratio of the first codebook is greater than that of the R15 codebook or the R16 codebook. The corresponding number of ports, number of delay paths, number of beams and non-zero coefficient ratio.

For example: the number of ports, the number of delay paths, the number of beams, and the proportion of non-zero coefficients in the R16 codebook are as shown in Table 1 below:

Table 1

Among them, Pv is used to calculate the number of delay paths as shown in Table 1 above, and the R16 codebook supports up to 32 ports.

In this embodiment, at least one of the number of ports, the number of delay paths, the number of beams, and the non-zero coefficient ratio of the first codebook may be greater than any combination of the number of ports and the number of delay paths in Table 1 above. , the number of beams, and the proportion of non-zero coefficients. For example: the first parameter of the first codebook in this implementation may be a parameter combination as shown in Table 2 below:

Table 2

That is to say, in this embodiment, a codebook with higher accuracy than the R15 codebook and R16 codebook can be used to report AI training data. At this time, the AI training data can more accurately describe the channel status of the target downlink channel. , so that the AI network model trained based on the AI training data better matches the actual channel status of the target downlink channel.

In a possible implementation, the terminal performs orthogonalization processing on the first information, which may be that the terminal performs orthogonalization processing on a vector composed of elements in any two columns of the precoding matrix. At this time, the terminal reports the precoding matrix. During the process of encoding the matrix, only the amplitude and phase of some elements can be reported. For the remaining elements whose amplitude and phase are not reported, the network side device can be based on vectors composed of elements in any two columns of the precoding matrix that are orthogonal to each other. The principle is calculated. That is to say, by making the element vectors in the precoding matrix orthogonal, the number of elements reported by the terminal can be reduced.

As an optional implementation manner, when the first channel information is a precoding matrix, the terminal performs a first process on the first information to obtain AI training data, including:

The terminal performs quantization processing on the amplitude and phase of the elements in the precoding matrix to obtain third information of the first matrix;

Wherein, the AI training data includes the third information.

In this embodiment, the amplitude and phase of the elements in the precoding matrix of all subbands of the target downlink channel are quantized respectively, and the obtained third information of the first matrix can reflect the amplitude and phase of the quantized elements, In this way, the first device can determine the amplitude and phase of the elements in the precoding matrix of all subbands of the target downlink channel based on the third information of the first matrix. Wherein, the first matrix may represent a quantized precoding matrix.

method one

As an optional implementation, the third information includes at least one of the following:

The position information of the first element and the amplitude and phase difference information of the second element relative to the first element, where the first element is the element with the largest amplitude in the first matrix, and the second element is the Elements in the first matrix other than the first element.

The third information includes the position information of the first element and the amplitude and phase difference information of the second element relative to the first element, which can mean that the terminal reports the position of the first element with the largest amplitude in the first matrix to the first device. , and calculate Calculate the amplitude and phase difference between the other second elements and the first element, for example: divide the second element by the first element to obtain the amplitude and phase difference of the second element relative to the first element. Finally, the terminal reports the quantized amplitude ratio and phase difference. In this way, the first element does not need to be reported, but only its position is reported, which can save resources occupied when reporting the first element.

Optionally, the first element and the second element are non-zero elements in the first matrix.

In this way, only the amplitude and phase difference information of the non-zero elements in the first matrix need to be reported. For the non-zero elements, the terminal does not need to report them. The first device can determine the element for which the amplitude and phase difference information has not been received. is zero element.

It should be noted that in implementation, if there are both zero elements and non-zero elements in the first matrix, the third information may also include the position information of the zero elements or the positions of the non-zero elements in the first matrix. information. Of course, in the case where all elements in the first matrix are non-zero elements, the third information may not include position information of the non-zero elements in the first matrix.

Optionally, when the rank of the target downlink channel is greater than 1, the first element and the second element are elements corresponding to the same layer in the first matrix.

Wherein, when the rank (Rank) of the target downlink channel is K, and K is greater than 1, the precoding matrix of each subband of the target downlink channel includes K columns, and each column corresponds to a layer, so When , each layer can be normalized separately, that is, the element with the largest amplitude is selected from each layer, the position information of the element is reported, and the amplitude and phase difference information of other elements and the element with the largest amplitude are reported. .

In another implementation, K layers can be treated as a whole, the element with the largest amplitude can be selected from them, the position information of the element can be reported, and the sum of the amplitudes of other elements in the K layers and the element with the largest amplitude can be reported. phase difference information.

Method 2

As an optional implementation, the third information includes:

The amplitude and phase of the third element, wherein the third element is a non-zero element in the first matrix, the non-zero element includes an element whose amplitude is greater than or equal to the first threshold, or an element whose amplitude is not equal to 0 ;

In the case where the first matrix includes zero elements and non-zero elements, the third information further includes at least one of the following:

First position information, the first position information being the position information of non-zero elements in the first matrix;

Second position information, the second position information is the position information of the zero element in the first matrix.

The third information includes the amplitude and phase of the third element. The terminal may directly quantize the amplitude and phase of the non-zero elements of the precoding matrix of each sub-band of the target downlink channel and then feed it back to the first device. In this way, When receiving the third information, the first device may determine that elements that have not received feedback are 0 elements, thereby determining the precoding matrix of each subband of the target downlink channel.

Alternatively, the terminal quantizes the amplitude and phase of elements whose amplitude is greater than or equal to the minimum value of amplitude quantization (for example: the minimum value of amplitude quantization is 0.5), and then feeds it back to the first device. For elements whose amplitude is less than the minimum value of amplitude quantization, , you don’t need to report it. In this way, when receiving the third information, the first device can regard the element that has not received feedback as a 0 element, or its amplitude as a fixed value (for example: 0.5).

It should be noted that in this embodiment, the terminal can also report the position information of the zero element or the element whose amplitude is less than the minimum value of the amplitude quantization, or report the position information of the third element. In this way, the first device can according to the position information and the received amplitude and phase of each element to determine the precoding matrix of each subband of the target downlink channel. Of course, when there is no zero element in the first matrix, the terminal may not report the position information of the third element, which is not specifically limited here.

Wherein, when the terminal reports a non-zero element and there is a zero element in the first matrix, the terminal can report the location information of the non-zero element, and the first device can determine that the unreported location information is the location of the zero element; or , the terminal can also report the location information of the zero element, and the first device can determine that the unreported location information is the location of the non-zero element.

Of course, when the terminal reports non-zero elements and there are no zero elements in the first matrix, the terminal does not need to report the location information. In this case, the first device can determine the order or position of each non-zero element reported by the terminal. to determine the meaning of the element.

Method three

As an optional implementation manner, when the rank of the target downlink channel is greater than 1, the first matrix includes at least two layers, and the third information includes the third information of each layer in the first matrix. The amplitude and phase of the four elements;

Wherein, the number of fourth elements corresponding to each of the at least two layers in the first matrix decreases layer by layer on the basis of N, where N is the number of elements corresponding to each layer in the precoding matrix, Alternatively, the number of fourth elements corresponding to each of the at least two layers is less than N.

Wherein, the number of fourth elements corresponding to each of the at least two layers in the first matrix decreases layer by layer on the basis of N, which can mean: for high-rank channel information feedback, the precoding matrix of each subband has Multiple columns, each column represents a layer, and each layer includes N elements. At this time, the first layer (such as layer 0) can report N elements, and the second layer (such as layer 1) can report (N- 1) elements, the third layer (such as layer 2) can report (N-2) elements, and so on.

It is worth noting that when receiving the above third information, the first device can determine the amplitude and phase of the elements not reported by the terminal based on the principle of orthogonality between any two layers of the target downlink channel, for example: based on Orthogonality between layer 0 and layer 1 to calculate the amplitude and phase of unreported elements in layer 1.

In addition, the above-mentioned unreported elements can be fixed-position elements, or fixed-position non-zero elements, for example: the first non-zero element counting from back to front in layer 1, and the first non-zero element in layer 2 from back to front. The first and second non-zero elements of the number, or the first non-zero element counting from back to front in layer 1 except the element with the largest amplitude, or the first non-zero element in layer 2 except the element with the largest amplitude The first and second non-zero elements counted from back to front; alternatively, the terminal can also report the position information of the unreported element. In this way, the first device can learn which element or elements are not reported by the terminal.

It should be noted that in implementation, in addition to decreasing layer by layer on the basis of N, the number of fourth elements included in the first matrix can also be other combinations, for example: assuming K is equal to 3, the third information Include any of the following:

The amplitude and phase of the N elements corresponding to layer 0 in the first matrix, the sum of The amplitude and phase of the (N-1) elements corresponding to layer 1, and the amplitude and phase of the (N-2) elements corresponding to layer 2 in the first matrix;

The amplitude and phase of the N elements corresponding to layer 0 in the first matrix, the amplitude and phase of the (N-2) elements corresponding to layer 1 in the first matrix, and the The amplitude and phase of the (N-3) elements corresponding to layer 2;

The amplitude and phase of the (N-1) elements in the first matrix corresponding to layer 0, the amplitude and phase of the (N-1) elements in the first matrix corresponding to layer 1, and the Describe the amplitude and phase of the (N-1) elements corresponding to layer 2 in the first matrix.

It should be noted that any two or three of the above methods 1 to 3 can be combined, for example:

In a possible implementation, method 1 is combined with method 3. At this time, the fourth element can be the first element and the second element, that is, the terminal can report the amplitude, phase and position information of the element with the largest amplitude, and report Normalized amplitude and phase information of other elements.

It should be noted that when the amplitude and phase of the fourth element included in the third information are the amplitude and phase after normalization, the position of the largest element of each layer needs to be reported at this time, and the fourth The minimum number of elements is: N elements corresponding to layer 0 in the first matrix, (N-1) elements corresponding to layer 1 in the first matrix, and There are (N-2) elements corresponding to layer 2, and so on.

It should be noted that when the amplitude and phase of the fourth element included in the third information are not the amplitude and phase after normalization, the position of the maximum coefficient of each layer does not need to be reported at this time, and since no For normalization processing, the modulus of the precoding matrix of each layer is 1. Based on this limitation, the number of fourth elements can be further reduced. For example: the fourth element includes: N elements corresponding to layer 0 in the first matrix, (N-2) elements corresponding to layer 1 in the first matrix, and (N-3) elements corresponding to layer 2, and so on.

In another possible implementation, method 2 is combined with method 3. At this time, the fourth element can be the third element, that is, the terminal can report the non-zero element in the first matrix or the amplitude is greater than or equal to the minimum amplitude quantization Magnitude, phase, and position information for the elements of the value.

Of course, the above method one can also be combined with method two, that is, the terminal selects the non-zero elements in the precoding matrix for normalization processing, and reports the amplitude and phase of the normalized non-zero elements, which will not be described in detail here. .

Optionally, the third information also includes at least one of the following:

Third position information, the third position information is the position information of the fourth element in the first matrix. At this time, the non-zero elements in the first matrix except the fourth element The location information can be agreed through a protocol or indicated by the first device;

Fourth position information, the fourth position information is the position information of non-zero elements in the first matrix except the fourth element. At this time, if the precoding matrix does not include zero elements, the terminal does not The position information of the fourth element in the first matrix needs to be reported;

Fifth position information, the fifth position information is the position information of the zero element and the fourth element in the first matrix At this time, the precoding matrix includes zero elements and non-zero elements, and the terminal needs to report the position information of the fourth element in the first matrix, as well as the position information of the zero element.

Through at least one of the above third location information, fourth location information and fifth location information, the first device can determine where the fourth element reported by the terminal is located in the precoding matrix, or determine that the terminal has not Which element or elements in the precoding matrix are the reported elements, or determine which element or elements need to be calculated based on the orthogonality between the layers or the principle that the modulus of the precoding matrix corresponding to each layer is equal to 1. element.

The terminal performs delay domain conversion processing on the precoding matrix to obtain a second matrix of X delay domains, where X is a positive integer;

The terminal determines that the AI training data includes fourth information of Y second matrices with the largest amplitude among the X second matrices, and Y is a positive integer less than or equal to X.

Wherein, the above-mentioned second matrix is similar to the first matrix, except that the second matrix is a precoding matrix in the delay domain, in which the precoding matrix is subjected to delay domain conversion processing, and the Y strongest delay domain precoding matrices are selected. The process of encoding the matrix is the same as the process of performing delay domain conversion processing and selecting the strongest delay domain precoding matrix in related technologies, and will not be described again here.

Optionally, the fourth information includes:

The position information of the fifth element and the amplitude and phase difference information of the sixth element relative to the fifth element, wherein the fifth element is the element with the largest amplitude in the second matrix, and the sixth element is the Elements in the second matrix other than the fifth element.

In this embodiment, the fifth element and the sixth element are normalized. The specific process of the normalization process is the same as the process of normalizing the first element and the second element in the first mode of the above embodiment. are the same and will not be repeated here.

Optionally, the fifth element and the sixth element are non-zero elements in the second matrix.

Wherein, when the precoding matrix in the time domain includes zero elements and non-zero elements, if the terminal only reports non-zero elements, the terminal also needs to report zero elements or non-zero elements in the precoding matrix in the time domain. Positional information for zero elements. So that the first device determines the precoding matrix in the time domain accordingly.

Optionally, when the rank of the target downlink channel is greater than 1, the fifth element and the sixth element are elements corresponding to the same layer in the second matrix.

This implementation mode is the same as the process of normalizing the first element and the second element of each layer in the first mode of the above embodiment, and will not be described again here.

Of course, in implementation, the terminal may also treat all layers as a whole and perform unified normalization processing on the first element and the second element in all layers.

Optionally, the fourth information includes:

The amplitude and phase of the seventh element, wherein the seventh element is a non-zero element in the second matrix, the non-zero element includes an element whose amplitude is greater than or equal to the first threshold, or an element whose amplitude is not equal to 0 ;

In the case where the second matrix includes zero elements and non-zero elements, the fourth information further includes at least one of the following:

Sixth position information, the sixth position information is the position information of non-zero elements in the second matrix;

Seventh position information, the seventh position information is the position information of the zero element in the second matrix.

The meaning of the seventh element is similar to the meaning of the third element in the above embodiment. The difference is that the seventh element is an element in the precoding matrix in the delay domain. The seventh element, the sixth position information and the seventh element are The meaning of the location information may refer to the meaning of the third element, the first location information, and the second location information in the second method of the above embodiment, and will not be described again here.

Optionally, in the case where the rank of the target downlink channel is greater than 1, the second matrix includes at least two layers, and the fourth information includes the amplitude and phase of the eighth element in the second matrix;

Wherein, the number of the eighth element in the second matrix corresponding to each of the at least two layers decreases layer by layer on the basis of L, where L is the element corresponding to each layer in the precoding matrix in the delay domain. The number of , or the number of eighth elements corresponding to each of the at least two layers is less than L.

In this implementation, the terminal does not need to report a complete precoding matrix for each layer, but a partial report. The rest can be used by the first device according to the orthogonality between the precoding matrices of each layer and the precoding matrix of each layer. The encoding matrix is calculated based on the principle that the modulus of the encoding matrix is equal to 1. For its specific description, please refer to the explanation of the fourth element in the third mode of the above embodiment, and will not be described again here.

Optionally, the eighth element includes a fifth element and a sixth element, or the eighth element includes a seventh element.

The same principle as that in the above embodiment, at least one of the first, second and third methods can be combined. In this embodiment, the channel information in the precoding matrix in the delay domain can also be normalized and filtered. At least one of the processes of zero elements, reducing reported elements, etc. will not be described again here.

Optionally, the fourth information also includes at least one of the following:

Eighth position information, the eighth position information is the position information of the eighth element in the second matrix;

Ninth position information, the ninth position information is the position information of non-zero elements in the second matrix except the eighth element;

Tenth position information, the tenth position information is the position information of the zero element and the position information of the eighth element in the second matrix.

The eighth position information, the ninth position information, and the tenth position information have similar meanings and functions to the third position information, the dead center position information, and the fifth position information in the above embodiments, and will not be described again here.

As an optional implementation manner, when the terminal has the first AI network model, the second information further includes at least one of the following:

Second channel characteristic information, the second channel characteristic information is obtained by processing the first information based on the first AI network model, and the second channel characteristic information is unquantized characteristic information;

The quantized second channel characteristic information, and the quantized information of the second channel characteristic information;

Third channel characteristic information, the third channel characteristic information is obtained by processing the first information based on the first AI network model, and the third channel characteristic information is quantized characteristic information.

In a possible implementation, when the first AI network model outputs the second channel characteristic information, the terminal can send the unquantized second channel characteristic information to the first device. At this time, the first device can The second channel characteristic information is input into the second AI network model to be trained, with the purpose of enabling the second AI network model to output the first information corresponding to the second channel characteristic information, and training to obtain the final second AI network model.

In another possible implementation, when the first AI network model outputs the second channel characteristic information, the terminal can perform quantization processing on the second channel characteristic information, and use the quantized second channel characteristic information to and reporting the quantified information of the second channel characteristic information to the first device. At this time, the first device can restore the second channel characteristic information according to the quantified information of the second channel characteristic information, and input the second channel characteristic information into the second AI network model to be trained, so that the second channel characteristic information can be The second AI network model is capable of outputting the first information corresponding to the second channel characteristic information, and is trained to obtain the final second AI network model. Compared with the above-mentioned terminal reporting unquantized second channel characteristic information, this implementation can reduce the overhead of transmitting the second channel characteristic information.

In a possible implementation, the first AI network model may directly output the quantized feature information, that is, the third channel feature information. In this case, the first AI network model has a coding function and a quantization function. After the terminal reports the third channel characteristic information to the first device, the functions of the second AI network model to be trained by the first device may correspond to the first AI network model. In this way, the first device can use the third channel The characteristic information is input into the second AI network model to be trained, with the purpose of enabling the second AI network model to output the first information corresponding to the third channel characteristic information, and the final second AI network model is trained; or, the first The device may first process the third channel characteristic information into unquantized channel characteristic information according to preset rules, and then input the unquantized channel characteristic information into the second AI network model to be trained, so that the second AI network model For the purpose of being able to output the first information corresponding to the third channel characteristic information, the final second AI network model is trained, in which the preset rules correspond to the quantification rules of the channel characteristic information of the first AI network model. Based on the preset The rules can restore the channel characteristic information after quantization of the first AI network model to the channel characteristic information before quantization of the first AI network model.

In this embodiment, when the terminal has a first AI network model, the terminal can input the first information into the first AI network model and obtain the channel characteristic information output by the first AI network model. At this time, the terminal The channel characteristic information may be quantized characteristic information, that is, the third channel characteristic information, or the channel characteristic information may be unquantized characteristic information, that is, the second channel characteristic information. At this time, the terminal reports the channel characteristic information and the corresponding channel information to the first device, so that the first device jointly trains the second AI network model based on the channel characteristic information and the corresponding channel information. At this time, based on the channel characteristic information The second AI network model trained with the corresponding channel information matches the first AI network model of the terminal.

As an optional implementation manner, when the first channel information is a precoding matrix, and the first information includes a K-layer precoding matrix, K is a positive integer, and the AI training data includes at least the following: One item:

The strongest M-layer precoding matrix among the K-layer precoding matrices, M is a positive integer less than or equal to K;

The K layer precoding matrix;

Among the K-layer precoding matrices, the precoding matrix indicated by the first indication information, and the first indication information comes from the first device;

In the K-layer precoding matrix, the precoding matrix indicated by the second indication information, where the second indication information is the indication information sent by the terminal to the first device in advance;

Among the K-layer precoding matrices, a precoding matrix that satisfies a preset condition.

When the rank K of the target downlink channel is greater than 1, at this time, the precoding matrix of the target downlink channel includes K layers.

Option 1: The stronger layer in the K-layer precoding matrix can more effectively reflect the channel status of the target downlink channel. In this way, the strongest (such as the largest amplitude) M layer is selected from the K-layer precoding matrix as AI training data. , for example: the terminal reports the precoding matrix of the strongest layer.

It is worth mentioning that when the terminal reuses the conventional Channel State Information (CSI) report as AI training data, it can use the precoding matrix corresponding to layer 0 in the CSI report as AI training data.

Compared with selecting the entire K-layer precoding matrix, it can reduce the amount of AI training data and retain as much effective information in the K-layer precoding matrix as possible.

Option two, the terminal can also use the K-layer precoding matrix as AI training data, so that it can accurately reflect the real channel status of the target downlink channel.

Option 3: The terminal can receive the above-mentioned first indication information before sending the second information. The first indication information can indicate the precoding matrix that the terminal needs to report. In this way, the terminal only needs to report the precoding matrix of the specified layer according to the instructions of the first device. Encoding matrix is enough.

Option four: the terminal can choose which layers of precoding matrices to report, and report second indication information to indicate which layers of precoding matrices the terminal reports. The terminal may send the second indication information before, after, or at the same time as the second information, which is not specifically limited here.

Option five, the terminal can select the required precoding matrix based on whether each layer in the K-layer precoding matrix meets the preset conditions. Among them, the precoding matrix that meets the preset conditions can more effectively reflect the channel status of the target downlink channel.

Optionally, the preset conditions include at least one of the following:

The channel quality indicator (Channel Quality Indicator, CQI) is greater than or equal to the first threshold;

The signal to interference plus noise ratio (SINR) is greater than or equal to the second threshold;

The characteristic value is greater than or equal to the third threshold;

The singular value is greater than or equal to the fourth threshold.

In this embodiment, the terminal can filter and report which layer or layer of precoding matrices according to preset conditions. Compared with selecting the entire K-layer precoding matrix, it can reduce the amount of AI training data and minimize the amount of AI training data. It is possible to retain the valid information in the K-layer precoding matrix.

In this embodiment of the present application, the terminal obtains first information, which includes first channel information of all subbands of the target downlink channel; the terminal performs first processing on the first information to obtain AI training data , wherein the first processing includes at least one of the following: screening processing, quantization processing, delay domain conversion processing, codebook conversion processing and Orthogonalization processing, the AI training data is used to train the first AI network model and/or the second AI network model, the first AI network model is used to process the second channel information into the first channel feature information, so The second AI network model is used to restore the first channel characteristic information to the second channel information, and the second channel information is the channel information of the target downlink channel; the terminal sends the first device to the first device. Two information, the second information includes the AI training data. After estimating and obtaining the first channel information of each subband of the target downlink channel, the terminal performs at least one of screening processing, quantization processing, delay domain conversion processing, codebook conversion processing, and orthogonalization processing on the channel information. The item can reduce the data amount of the obtained AI training data, and when the terminal sends the AI training data to the first device, the amount of resources occupied by transmitting the AI training data can be reduced. In addition, since the AI training data is obtained based on the first processing of channel information of all subbands of the target downlink channel, the AI training data can reflect the channel status of the target downlink channel, and the first device trains the coding AI network based on the AI training data. In the process of modeling (i.e., the first AI network model) and/or the decoding AI network model (i.e., the second AI network model), the matching between the trained encoding AI network model and/or the decoding AI network model and the target downlink channel can be improved. degree. In the process of using the first AI network model to compress the channel information of the target downlink channel into channel feature information, and/or using the second AI network model to restore the channel feature information of the target downlink channel into channel information, the first step can be improved. The compression coding accuracy of the AI network model and/or the decoding accuracy of the second AI network model can be improved and/or the matching degree between the first AI network model and the second AI network model can be improved, thereby reducing the occupancy of the channel information reporting process. resources, and improve the accuracy of the reporting process.

Please refer to Figure 3. An embodiment of the present application provides an AI network model training method, the execution subject of which is the first device. The first device may be a network-side device, such as the network-side device 12 listed in the embodiment as shown in Figure 1 or a core network device. For ease of explanation, in the embodiment of this application, the first device is a base station as an example. for example.

As shown in Figure 3, the AI network model training method executed by the first device may include the following steps:

Step 301: The first device receives second information from the terminal, where the second information includes AI training data, the AI training data is obtained based on the first processing of the first information, and the first information includes target downlink The first channel information of all subbands of the channel.

Step 302: The first device trains a first AI network model and/or a second AI network model according to the AI training data. The first AI network model is used to process the second channel information into the first channel characteristic information. , the second AI network model is used to restore the first channel characteristic information to the second channel information, where the second channel information is the channel information of the target downlink channel.

In implementation, the base station can train the encoding AI network model based on the AI training data, or train the decoding AI network model based on the AI training data, or jointly train the encoding AI network model and the decoding AI network model based on the AI training data.

In the embodiment of the present application, the base station can obtain the channel information obtained by performing channel estimation on the target downlink channel from the terminal to train the encoding and/or decoding AI network model suitable for the target downlink channel, and can improve the first AI network model obtained by training. And/or the degree of matching between the second AI network model and the target downlink channel.

As an optional implementation, after the first device trains the first AI network model according to the AI training data, the method further includes:

The first device sends relevant information of the first AI network model to the terminal.

In this implementation, after completing the training of the coding AI network model, the base station sends the trained coding AI network model to the terminal. In this way, the terminal can report the subsequent channel state information (CSI). The encoding AI network model is used to compress and encode the estimated channel information, and the compressed and encoded channel characteristic information is reported. After that, the base station can use the decoding AI network model that matches the encoding AI network model to restore the channel characteristic information. into the original channel information, or the base station can also use a non-AI method to restore the channel characteristic information into the original channel information, for example, using a certain algorithm to restore the channel characteristic information into the original channel information.

Third channel characteristic information, the third channel characteristic information is obtained by processing the first information based on the first AI network model, and the third channel characteristic information is quantized characteristic information;

The first device trains the first AI network model and/or the second AI network model according to the AI training data, including:

The first device trains a second AI network model that matches the first AI network model based on the second channel characteristic information or the third channel characteristic information and the AI training data.

In this embodiment, the first device trains a second AI network model that matches the first AI network model based on the second channel characteristic information or the third channel characteristic information and the AI training data. , it may be that: the AI training data includes original channel information corresponding to the second channel characteristic information or the third channel characteristic information, and the first device converts the second channel characteristic information or the third channel characteristic information. The three-channel characteristic information is input into the second AI network model to be trained, and the second AI network model outputs the original channel information corresponding to the second channel characteristic information or the third channel characteristic information as the goal to train the second AI. network model, in this way, the second AI network model finally trained can match the first AI network model of the terminal, that is: the terminal inputs channel information to the first AI network model and obtains the output channel characteristics of the first AI network model information, and reports the channel characteristic information to the base station as CSI. The base station then inputs the channel characteristic information into the trained second AI network model and obtains the channel information output by the second AI network model, that is, the terminal implements Compression coding of channel information realizes the recovery of compressed and coded channel characteristic information at the base station.

As an optional implementation manner, the first channel information is at least one of a precoding matrix and a channel matrix.

The K layer precoding matrix;

Optionally, the preset conditions include at least one of the following:

The channel quality indicator CQI is greater than or equal to the first threshold;

The signal to interference plus noise ratio SINR is greater than or equal to the second threshold;

The characteristic value is greater than or equal to the third threshold;

The singular value is greater than or equal to the fourth threshold.

As an optional implementation manner, when the first channel information is a precoding matrix, the AI training data includes:

The third information of the first matrix is obtained based on quantization processing of the amplitude and phase of the non-zero elements in the precoding matrix.

The above-mentioned third information has the same meaning and function as the third information in the method embodiment shown in Figure 2, and will not be described again here.

As an optional implementation, the third information includes:

The above-mentioned first element and second element have the same meaning as the first element and the second element in the method embodiment shown in Figure 2, and will not be described again here.

As an optional implementation, the third information includes:

Among them, the above-mentioned third element, first position information and second position information have the same meaning as the third element, first position information and second position information in the method embodiment as shown in Figure 2, and will not be described again here. .

As an optional implementation manner, when the rank of the target downlink channel is greater than 1, the first matrix includes at least two layers;

The number of fourth elements in the first matrix corresponding to each of the at least two layers decreases layer by layer based on N, where N is the number of elements corresponding to each layer in the precoding matrix, or, The number of fourth elements corresponding to each of the at least two layers is less than N.

The above-mentioned fourth element has the same meaning as the fourth element in the method embodiment shown in Figure 2, and will not be described again here.

Optionally, the fourth element includes a first element and a second element, or the fourth element includes a third element.

Optionally, the third information also includes at least one of the following:

Third position information, the third position information is the position information of the fourth element in the first matrix;

Fourth position information, the fourth position information is the position information of non-zero elements in the first matrix except the fourth element;

Fifth position information, the fifth position information is the position information of the zero element in the first matrix and the position information of the fourth element.

The above-mentioned third position information, fourth position information and fifth position information have the same meaning as the third position information, fourth position information and fifth position information in the method embodiment as shown in Figure 2, and are not used here. To elaborate.

The fourth information of the Y second matrices in the delay domain, wherein the Y second matrices are the Y ones with the largest amplitudes among the X second matrices, and the X second matrices are based on the precoding The matrix is obtained by performing delay domain conversion processing. X is a positive integer and Y is a positive integer less than or equal to X.

The above-mentioned fourth information and the second matrix of the delay domain have the same meaning as the fourth information and the second matrix of the delay domain in the method embodiment shown in Figure 2, and will not be described again here.

Optionally, the fourth information includes:

The position information of the fifth element and the amplitude and phase difference information of the sixth element relative to the fifth element, wherein the fifth element is the element with the largest amplitude in the second matrix, and the sixth element is the The elements in the second matrix other than the fifth element.

The fifth element and the sixth element mentioned above have the same meaning as the fifth element and the sixth element in the method embodiment shown in Figure 2, and will not be described again here.

As an optional implementation, the fourth information includes:

The seventh element, sixth position information and seventh position information mentioned above have the same meaning as the seventh element, sixth position information and seventh position information in the method embodiment shown in Figure 2 and will not be described again here.

As an optional implementation manner, when the rank of the target downlink channel is greater than 1, the second matrix includes at least two layers, and the fourth information includes the eighth element of the second matrix. amplitude and phase;

The number of eighth elements in the second matrix corresponding to each of the at least two layers decreases layer by layer on the basis of L, where L is the number of elements corresponding to each layer in the precoding matrix in the delay domain. number, or the number of eighth elements corresponding to each of the at least two layers is less than L.

Optionally, the fourth information also includes at least one of the following:

Wherein, the above-mentioned eighth element, eighth position information, ninth position information and tenth position information are the same as the eighth element, eighth position information, ninth position information and tenth position information in the method embodiment as shown in Figure 2 have the same meaning and will not be repeated here.

The AI network model training method executed by the first device provided by the embodiment of the present application corresponds to the information transmission method executed by the terminal. Both can reduce the reporting of high-precision AI training data and reduce the resource consumption of transmitting the AI training data. Function, when training the first AI network model and/or the second AI network model based on the AI training data, it can also improve the matching degree between the trained first AI network model and/or the second AI network model and the target downlink channel.

In order to facilitate the explanation of the information transmission method provided by the embodiment of the present application, taking the first device being a base station as an example to illustrate that the information transmission method and AI network model training method provided by the embodiment of the present application may include the following steps:

Step 1. The terminal uses 4 receiving ports to receive CSI Reference Signal (CSI-RS). The base station uses 32 transmitting ports to send CSI-RS. The terminal performs channel estimation. In each physical resource block (Physical Resource Block, PRB) to obtain a downlink channel matrix with a dimension of 4×32. There are 52 PRBs in total, and each subband has 4 PRBs, for a total of 13 subbands. In each subband, the terminal calculates the corresponding precoding matrix based on the four 32×4 channel matrices of the 4 PRBs, and obtains the precoding matrix of up to 4 layers, that is, a 32×4 precoding matrix. Each column represents a layer.

Step 2. The terminal can directly send the precoding matrices of the 4 layers to the base station, or, for the 32 elements of the first layer, find the element with the largest amplitude, assuming that the element with the largest amplitude in the first layer is in the 3rd , then divide the third element by the other elements to obtain the normalized precoding vector of the first layer. After quantizing the amplitude and phase of all elements except the third element, the terminal can The non-zero elements are reported to the base station, and the position of the non-zero elements is also reported. Optionally, if there are no non-zero elements, the terminal reports all elements. This There is no need to report the position. Of course, the terminal always needs to position the element with the largest amplitude in the third position.

Step 3. For the elements of the second layer, the terminal also finds the element with the largest amplitude. Assuming that the element with the largest amplitude in the second layer is the first one, the terminal divides all other elements by the strongest element to get the equivalent of the normalized precoding vector, quantizing the amplitude and phase of elements other than the first. Same as the precoding matrix of the first layer above. If there are no non-zero elements in the precoding matrix of the second layer, the second to 31st elements will be reported, that is, the 32nd element will not be reported. The method is the same as above, and so on. .

Step 4. The precoding matrix received by the base station is:

Among them, u _i,j represents the j-th element of the i-th layer.

Since the eigenvectors of each column are orthogonal, the following formula is satisfied:

The base station can solve the unknown u _2,32 .

Since the first and second columns are both orthogonal to the third column, the third column can have two unknowns. Similarly, the fourth column has three, that is, the maximum third column can underreport two, and the maximum fourth column can Three are underreported. There is no limit to the specific number of underreported.

For the information transmission method provided by the embodiments of the present application, the execution subject may be an information transmission device. In the embodiment of the present application, an information transmission device performing an information transmission method is used as an example to illustrate the information transmission device provided by the embodiment of the present application.

Please refer to Figure 4. An information transmission device provided by an embodiment of the present application can be a device in a terminal. As shown in Figure 4, the information transmission device 400 can include the following modules:

The acquisition module 401 is used to acquire first information, where the first information includes first channel information of all subbands of the target downlink channel;

The first processing module 402 is used to perform a first processing on the first information to obtain AI training data, where the first processing includes at least one of the following: screening processing, quantization processing, delay domain conversion processing, coding In this conversion process and orthogonalization process, the AI training data is used to train the first AI network model and/or the second AI network model, and the first AI network model is used to process the second channel information into the first channel. Characteristic information, the second AI network model is used to restore the first channel characteristic information to the second channel information, and the second channel information is the channel information of the target downlink channel;

The first sending module 403 is configured to send second information to the first device, where the second information includes the AI training data.

Optionally, in the case where the terminal has the first AI network model, the second information further includes at least one of the following:

Optionally, the first channel information is at least one of a precoding matrix and a channel matrix.

Optionally, when the first channel information is a precoding matrix, and the first information includes the K layer precoding matrix, K is a positive integer, and the AI training data includes at least one of the following:

The K layer precoding matrix;

Optionally, the preset conditions include at least one of the following:

The characteristic value is greater than or equal to the third threshold;

The singular value is greater than or equal to the fourth threshold.

Optionally, when the first channel information is a precoding matrix, the first processing module 402 is specifically used to:

Perform quantization processing on the amplitude and phase of the elements in the precoding matrix to obtain third information of the first matrix;

Wherein, the AI training data includes the third information.

Optionally, the third information includes:

Optionally, when the rank of the target downlink channel is greater than 1, the first matrix includes at least two layers, so The third information includes the amplitude and phase of the fourth element of each layer in the first matrix;

Optionally, the third information also includes at least one of the following:

Fifth position information, the fifth position information is the position information of the zero element and the position information of the fourth element in the first matrix.

Optionally, when the first channel information is a precoding matrix, the first processing module 402 includes:

A first processing unit configured to perform delay domain conversion processing on the precoding matrix to obtain a second matrix of X delay domains, where X is a positive integer;

A first determination unit configured to determine that the AI training data includes the fourth information of the Y second matrices with the largest amplitude among the X second matrices, where Y is a positive integer less than or equal to X.

Optionally, the fourth information includes:

Optionally, when the rank of the target downlink channel is greater than 1, the second matrix includes at least two layers, and the fourth information includes the amplitude and phase of the eighth element in the second matrix;

Optionally, the fourth information also includes at least one of the following:

The information transmission device 400 provided by the embodiment of the present application can implement various processes implemented by the terminal in the method embodiment as shown in Figure 2, and can achieve the same beneficial effects. To avoid duplication, details will not be described here.

The information transmission device in the embodiment of the present application may be an electronic device, such as an electronic device with an operating system, or may be a component in the electronic device, such as an integrated circuit or chip. The electronic device may be a terminal or other devices other than the terminal. For example, terminals may include but are not limited to the types of terminals 11 listed above, and other devices may be servers, network attached storage (Network Attached Storage, NAS), etc., which are not specifically limited in the embodiment of this application.

For the AI network model training method provided by the embodiments of the present application, the execution subject may be an AI network model training device. In the embodiment of the present application, the AI network model training device executed by the AI network model training method is used as an example to illustrate the AI network model training device provided by the embodiment of the present application.

Please refer to Figure 5. An AI network model training device provided by an embodiment of the present application can be a device in the first device. As shown in Figure 5, the AI network model training device 500 can include the following modules:

The first receiving module 501 is used to receive second information from the terminal, where the second information includes AI training data, the AI training data is obtained based on the first processing of the first information, and the first information includes The first channel information of all subbands of the target downlink channel;

The training module 502 is used to train the first AI network model and/or the second AI network model according to the AI training data. The first AI network model is used to process the second channel information into the first channel characteristic information, so The second AI network model is used to restore the first channel characteristic information to the second channel information, and the second channel information is the channel information of the target downlink channel.

Optionally, the AI network model training device 500 also includes:

The second sending module is configured to send relevant information of the first AI network model to the terminal.

The first device trains a first AI network model and/or a second AI network model according to the AI training data, including include:

The K layer precoding matrix;

Optionally, the preset conditions include at least one of the following:

The characteristic value is greater than or equal to the third threshold;

The singular value is greater than or equal to the fourth threshold.

Optionally, when the first channel information is a precoding matrix, the AI training data includes:

Optionally, the third information includes:

Optionally, in the case where the rank of the target downlink channel is greater than 1, the first matrix includes at least two layers;

The number of fourth elements in the first matrix corresponding to each of the at least two layers decreases layer by layer on the basis of N, where N is the number of elements corresponding to each layer in the precoding matrix, or, The number of fourth elements corresponding to each of the at least two layers is less than N.

Optionally, the third information also includes at least one of the following:

Optionally, the fourth information includes:

Optionally, the fourth information also includes at least one of the following:

The AI network model training device in the embodiment of the present application may be an electronic device, such as an electronic device with an operating system, or may be a component in the electronic device, such as an integrated circuit or chip. The electronic device may be a network side device. For example, the terminal may include but is not limited to the types of network side devices 12 listed above, which are not specifically limited in the embodiments of this application.

The AI network model training device 500 provided by the embodiment of the present application can implement various processes implemented by the first device in the method embodiment as shown in Figure 3, and can achieve the same beneficial effects. To avoid duplication, they will not be described again here.

Optionally, as shown in Figure 6, this embodiment of the present application also provides a communication device 600, which includes a processor 601 and a memory 602. The memory 602 stores programs or instructions that can be run on the processor 601, for example. , when the communication device 600 is a terminal, when the program or instruction is executed by the processor 601, each step of the method embodiment shown in Figure 2 is implemented, and the same technical effect can be achieved. When the communication device 600 is the first device, when the program or instruction is executed by the processor 601, each step of the method embodiment shown in Figure 3 is implemented, and the same technical effect can be achieved. To avoid duplication, the details will not be described here.

An embodiment of the present application also provides a terminal, including a processor and a communication interface. The communication interface is used to obtain first information. The first information includes first channel information of all subbands of the target downlink channel; the processor is used to obtain first information. Perform a first process on the first information to obtain AI training data, where the first process includes at least one of the following: screening processing, quantization processing, delay domain conversion processing, codebook conversion processing, and orthogonalization processing , the AI training data is used to train the first AI network model and/or the second AI network model, the first AI network model is used to process the second channel information into the first channel feature information, the second AI The network model is used to restore the first channel characteristic information to the second channel information, and the second channel information is the channel information of the target downlink channel; the communication interface is also used to send the first device to the first device. Two information, the second information includes the AI training data. Specifically, FIG. 7 is a schematic diagram of the hardware structure of a terminal that implements an embodiment of the present application.

The terminal 700 includes but is not limited to: a radio frequency unit 701, a network module 702, an audio output unit 703, an input unit 704, a sensor 705, a display unit 706, a user input unit 707, an interface unit 708, a memory 709, a processor 710, etc. At least some parts.

Those skilled in the art can understand that the terminal 700 may also include a power supply (such as a battery) that supplies power to various components. The power supply may be logically connected to the processor 710 through a power management system, thereby managing charging, discharging, and power consumption through the power management system. Management and other functions. The terminal structure shown in FIG. 7 does not constitute a limitation on the terminal. The terminal may include more or fewer components than shown in the figure, or some components may be combined or arranged differently, which will not be described again here.

It should be understood that in this embodiment of the present application, the input unit 704 may include a graphics processing unit (GPU) 7041 and a microphone 7042. The graphics processor 7041 is responsible for the image capture device (GPU) in the video capture mode or the image capture mode. Process the image data of still pictures or videos obtained by cameras (such as cameras). show The display unit 706 may include a display panel 7061, which may be configured in the form of a liquid crystal display, an organic light emitting diode, or the like. The user input unit 707 includes a touch panel 7071 and at least one of other input devices 7072 . Touch panel 7071, also called touch screen. The touch panel 7071 may include two parts: a touch detection device and a touch controller. Other input devices 7072 may include but are not limited to physical keyboards, function keys (such as volume control keys, switch keys, etc.), trackballs, mice, and joysticks, which will not be described again here.

In this embodiment of the present application, after receiving downlink data from the network side device, the radio frequency unit 701 can transmit it to the processor 710 for processing; in addition, the radio frequency unit 701 can send uplink data to the network side device. Generally, the radio frequency unit 701 includes, but is not limited to, an antenna, amplifier, transceiver, coupler, low noise amplifier, duplexer, etc.

Memory 709 may be used to store software programs or instructions as well as various data. The memory 709 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instructions required for at least one function (such as a sound playback function, Image playback function, etc.) etc. Additionally, memory 709 may include volatile memory or non-volatile memory, or memory 709 may include both volatile and non-volatile memory. Among them, 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 removable memory. Erase programmable read-only memory (Electrically EPROM, EEPROM) or flash memory. Volatile memory can be random access memory (Random Access Memory, RAM), static random access memory (Static RAM, SRAM), 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, DDRSDRAM), enhanced synchronous dynamic random access memory (Enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (Synch link DRAM) , SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DRRAM). Memory 709 in embodiments of the present application includes, but is not limited to, these and any other suitable types of memory.

The processor 710 may include one or more processing units; optionally, the processor 710 integrates an application processor and a modem processor, where the application processor mainly handles operations related to the operating system, user interface, application programs, etc., Modem processors mainly process wireless communication signals, such as baseband processors. It can be understood that the above-mentioned modem processor may not be integrated into the processor 710.

Among them, the radio frequency unit 701 is used to obtain first information, where the first information includes first channel information of all subbands of the target downlink channel;

Processor 710, configured to perform first processing on the first information to obtain AI training data, where the first processing includes at least one of the following: filtering processing, quantization processing, delay domain conversion processing, codebook conversion Processing and orthogonalization processing, the AI training data is used to train the first AI network model and/or the second AI network model, the first AI network model is used to process the second channel information into the first channel characteristic information , the second AI network model is used to restore the first channel characteristic information to the second channel information, and the second channel information is the channel information of the target downlink channel;

The radio frequency unit 701 is also configured to send second information to the first device, where the second information includes the AI training data. according to.

The K layer precoding matrix;

Optionally, the preset conditions include at least one of the following:

The characteristic value is greater than or equal to the third threshold;

The singular value is greater than or equal to the fourth threshold.

Optionally, when the first channel information is a precoding matrix, the first processing performed by the processor 710 on the first information to obtain AI training data includes:

Wherein, the AI training data includes the third information.

Optionally, the third information includes:

Optionally, when the rank of the target downlink channel is greater than 1, the first matrix includes at least two layers, and the third information includes the amplitude sum of the fourth elements of each layer in the first matrix. phase;

Optionally, the third information also includes at least one of the following:

Perform delay domain conversion processing on the precoding matrix to obtain a second matrix of X delay domains, where X is a positive integer;

It is determined that the AI training data includes the fourth information of the Y second matrices with the largest amplitude among the X second matrices, and Y is a positive integer less than or equal to X.

Optionally, the fourth information includes:

Optionally, the fourth information also includes at least one of the following:

The terminal 700 provided by the embodiment of the present application can implement various processes performed by the information transmission device as shown in Figure 4, and can achieve the same beneficial effects. To avoid duplication, details will not be described here.

Embodiments of the present application also provide a network side device. The network side device may be an access network device or a core network device. The network side device includes a communication interface and a processor, wherein the communication interface is used to receive a third signal from the terminal. Two information, wherein the second information includes AI training data, the AI training data is obtained based on first processing of the first information, and the first information includes first channel information of all subbands of the target downlink channel; The processor is used to train a first AI network model and/or a second AI network model according to the AI training data. The first AI network model is used to process the second channel information into first channel characteristic information. The second AI network model is used to restore the first channel characteristic information to the second channel information, where the second channel information is the channel information of the target downlink channel.

This network side device embodiment corresponds to the method embodiment shown in Figure 3. Each implementation process and implementation manner of the method embodiment shown in Figure 3 can be applied to this network side device embodiment, and can achieve the same technical effect.

Embodiments of the present application also provide a readable storage medium. Programs or instructions are stored on the readable storage medium. When the program or instructions are executed by a processor, each process of the method embodiment shown in Figure 2 or Figure 3 is implemented. , and can achieve the same technical effect, so to avoid repetition, they will not be described again here.

Wherein, the processor is the processor in the terminal described in the above embodiment. The readable storage medium includes computer readable storage media, such as computer read-only memory ROM, random access memory RAM, magnetic disk or optical disk, etc.

An embodiment of the present application further provides a chip. The chip includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions. The implementation is as shown in Figure 2 or Figure 3. Each process of the method embodiment is shown, and the same technical effect can be achieved. To avoid repetition, the details will not be described here.

It should be understood that the chips mentioned in the embodiments of this application may also be called system-on-chip, system-on-a-chip, system-on-chip or system-on-chip, etc.

The embodiment of the present application further provides a computer program/program product, the computer program/program product is stored In the storage medium, the computer program/program product is executed by at least one processor to implement each process of the method embodiment shown in Figure 2 or Figure 3, and can achieve the same technical effect. To avoid duplication, it will not be repeated here. Repeat.

Embodiments of the present application also provide a communication system, including: a terminal and a network side device. The terminal can be used to perform the steps of the information transmission method as shown in Figure 2. The network side device can be used to perform the steps of the information transmission method as shown in Figure 3. The steps of the AI network model training method shown below.

It should be noted that, in this document, the terms "comprising", "comprises" or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or device that includes a series of elements not only includes those elements, It also includes other elements not expressly listed or inherent in the process, method, article or apparatus. Without further limitation, an element defined by the statement "comprises a..." does not exclude the presence of additional identical elements in a process, method, article or apparatus that includes that element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved. Functions may be performed, for example, the methods described may be performed in an order different from that described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation. Based on this understanding, the technical solution of the present application can be embodied in the form of a computer software product that is essentially or contributes to the existing technology. The computer software product is stored in a storage medium (such as ROM/RAM, disk , CD), including several instructions to cause a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of this application.

The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above-mentioned specific implementations. The above-mentioned specific implementations are only illustrative and not restrictive. Those of ordinary skill in the art will Inspired by this application, many forms can be made without departing from the purpose of this application and the scope protected by the claims, all of which fall within the protection of this application.

Claims

An information transmission method including:

The terminal acquires first information, where the first information includes first channel information of all subbands of the target downlink channel;

The terminal performs a first process on the first information to obtain AI training data, where the first process includes at least one of the following: screening processing, quantization processing, delay domain conversion processing, codebook conversion processing, and normalization processing. Cross-processing, the AI training data is used to train the first AI network model and/or the second AI network model, the first AI network model is used to process the second channel information into the first channel feature information, the The second AI network model is used to restore the first channel characteristic information to the second channel information, where the second channel information is the channel information of the target downlink channel;

The terminal sends second information to the first device, where the second information includes the AI training data.
The method according to claim 1, wherein, in the case that the terminal has the first AI network model, the second information further includes at least one of the following:

Second channel characteristic information, the second channel characteristic information is obtained by processing the first information based on the first AI network model, and the second channel characteristic information is unquantized characteristic information;

The quantized second channel characteristic information, and the quantized information of the second channel characteristic information;

Third channel characteristic information, the third channel characteristic information is obtained by processing the first information based on the first AI network model, and the third channel characteristic information is quantized characteristic information.
The method according to claim 1, wherein the first channel information is at least one of a precoding matrix and a channel matrix.
The method according to claim 3, wherein when the first channel information is a precoding matrix, and the first information includes a K-layer precoding matrix, K is a positive integer, and the AI training data includes At least one of the following:

The strongest M-layer precoding matrix among the K-layer precoding matrices, M is a positive integer less than or equal to K;

The K layer precoding matrix;

Among the K-layer precoding matrices, the precoding matrix indicated by the first indication information, and the first indication information comes from the first device;

In the K-layer precoding matrix, the precoding matrix indicated by the second indication information, where the second indication information is the indication information sent by the terminal to the first device in advance;

Among the K-layer precoding matrices, a precoding matrix that satisfies a preset condition.
The method according to claim 4, wherein the preset conditions include at least one of the following:

The channel quality indicator CQI is greater than or equal to the first threshold;

The signal to interference plus noise ratio SINR is greater than or equal to the second threshold;

The characteristic value is greater than or equal to the third threshold;

The singular value is greater than or equal to the fourth threshold.
The method according to any one of claims 1 to 5, wherein when the first channel information is a precoding matrix, the terminal performs first processing on the first information to obtain AI training data ,include:

The terminal performs quantization processing on the amplitude and phase of the elements in the precoding matrix to obtain third information of the first matrix;

Wherein, the AI training data includes the third information.
The method of claim 6, wherein the third information includes:

The position information of the first element and the amplitude and phase difference information of the second element relative to the first element, where the first element is the element with the largest amplitude in the first matrix, and the second element is the Elements in the first matrix other than the first element.
The method of claim 7, wherein the first element and the second element are non-zero elements in the first matrix.
The method according to claim 7, wherein when the rank of the target downlink channel is greater than 1, the first element and the second element are elements corresponding to the same layer in the first matrix.
The method of claim 6, wherein the third information includes:

The amplitude and phase of the third element, wherein the third element is a non-zero element in the first matrix, the non-zero element includes an element whose amplitude is greater than or equal to the first threshold, or an element whose amplitude is not equal to 0 ;

In the case where the first matrix includes zero elements and non-zero elements, the third information further includes at least one of the following:

First position information, the first position information being the position information of non-zero elements in the first matrix;

Second position information, the second position information is the position information of the zero element in the first matrix.
The method according to any one of claims 6 to 10, wherein, when the rank of the target downlink channel is greater than 1, the first matrix includes at least two layers, and the third information includes the third layer. the amplitude and phase of the fourth element of each layer in a matrix;

Wherein, the number of fourth elements corresponding to each of the at least two layers in the first matrix decreases layer by layer on the basis of N, where N is the number of elements corresponding to each layer in the precoding matrix, Alternatively, the number of fourth elements corresponding to each of the at least two layers is less than N.
The method of claim 11, wherein the fourth element includes a first element and a second element, or the fourth element includes a third element.
The method of claim 11, wherein the third information further includes at least one of the following:

Third position information, the third position information is the position information of the fourth element in the first matrix;

Fourth position information, the fourth position information is the position information of non-zero elements in the first matrix except the fourth element;

Fifth position information, the fifth position information is the position information of the zero element and the position information of the fourth element in the first matrix.
The method according to any one of claims 1 to 5, wherein when the first channel information is a precoding matrix, the terminal performs first processing on the first information to obtain AI training data ,include:

The terminal performs delay domain conversion processing on the precoding matrix to obtain a second matrix of X delay domains, where X is a positive integer;

The terminal determines that the AI training data includes fourth information of Y second matrices with the largest amplitude among the X second matrices, and Y is a positive integer less than or equal to X.
The method of claim 14, wherein the fourth information includes:

The position information of the fifth element and the amplitude and phase difference information of the sixth element relative to the fifth element, wherein the fifth element is the element with the largest amplitude in the second matrix, and the sixth element is the Elements in the second matrix other than the fifth element.
The method of claim 15, wherein the fifth element and the sixth element are non-zero elements in the second matrix.
The method according to claim 15, wherein when the rank of the target downlink channel is greater than 1, the fifth element and the sixth element are elements corresponding to the same layer in the second matrix.
The method of claim 14, wherein the fourth information includes:

The amplitude and phase of the seventh element, wherein the seventh element is a non-zero element in the second matrix, the non-zero element includes an element whose amplitude is greater than or equal to the first threshold, or an element whose amplitude is not equal to 0 ;

In the case where the second matrix includes zero elements and non-zero elements, the fourth information further includes at least one of the following:

Sixth position information, the sixth position information is the position information of non-zero elements in the second matrix;

Seventh position information, the seventh position information is the position information of the zero element in the second matrix.
The method according to any one of claims 14 to 18, wherein when the rank of the target downlink channel is greater than 1, the second matrix includes at least two layers, and the fourth information includes the first the amplitude and phase of the eighth element in the second matrix;

Wherein, the number of the eighth element in the second matrix corresponding to each of the at least two layers decreases layer by layer on the basis of L, where L is the element corresponding to each layer in the precoding matrix in the delay domain. The number of , or the number of eighth elements corresponding to each of the at least two layers is less than L.
The method of claim 19, wherein the eighth element includes a fifth element and a sixth element, or the eighth element includes a seventh element.
The method of claim 19, wherein the fourth information further includes at least one of the following:

Eighth position information, the eighth position information is the position information of the eighth element in the second matrix;

Ninth position information, the ninth position information is the position information of non-zero elements in the second matrix except the eighth element;

Tenth position information, the tenth position information is the position information of the zero element and the position information of the eighth element in the second matrix.
An artificial intelligence AI network model training method, including

The first device receives second information from the terminal, wherein the second information includes AI training data, the AI training data is obtained based on first processing of the first information, and the first information includes all of the target downlink channels. Subband The first channel information;

The first device trains a first AI network model and/or a second AI network model according to the AI training data. The first AI network model is used to process the second channel information into the first channel characteristic information. The second AI network model is used to restore the first channel characteristic information to the second channel information, where the second channel information is the channel information of the target downlink channel.
The method of claim 22, wherein after the first device trains the first AI network model according to the AI training data, the method further includes:

The first device sends relevant information of the first AI network model to the terminal.
The method according to claim 22, wherein, in the case that the terminal has the first AI network model, the second information further includes at least one of the following:

Second channel characteristic information, the second channel characteristic information is obtained by processing the first information based on the first AI network model, and the second channel characteristic information is unquantized characteristic information;

The quantized second channel characteristic information, and the quantized information of the second channel characteristic information;

Third channel characteristic information, the third channel characteristic information is obtained by processing the first information based on the first AI network model, and the third channel characteristic information is quantized characteristic information;

The first device trains the first AI network model and/or the second AI network model according to the AI training data, including:

The first device trains a second AI network model that matches the first AI network model based on the second channel characteristic information or the third channel characteristic information and the AI training data.
The method according to claim 22, wherein the first channel information is at least one of a precoding matrix and a channel matrix.
The method according to claim 25, wherein when the first channel information is a precoding matrix, and the first information includes a K layer precoding matrix, K is a positive integer, and the AI training data includes At least one of the following:

The strongest M-layer precoding matrix among the K-layer precoding matrices, M is a positive integer less than or equal to K;

The K layer precoding matrix;

Among the K-layer precoding matrices, the precoding matrix indicated by the first indication information, and the first indication information comes from the first device;

In the K-layer precoding matrix, the precoding matrix indicated by the second indication information, where the second indication information is the indication information sent by the terminal to the first device in advance;

Among the K-layer precoding matrices, a precoding matrix that satisfies a preset condition.
The method according to claim 26, wherein the preset conditions include at least one of the following:

The channel quality indicator CQI is greater than or equal to the first threshold;

The signal to interference plus noise ratio SINR is greater than or equal to the second threshold;

The characteristic value is greater than or equal to the third threshold;

The singular value is greater than or equal to the fourth threshold.
The method according to any one of claims 22 to 27, wherein when the first channel information is a precoding matrix, the AI training data includes:

The third information of the first matrix is obtained based on quantization processing of the amplitude and phase of the non-zero elements in the precoding matrix.
The method of claim 28, wherein the third information includes:

The position information of the first element and the amplitude and phase difference information of the second element relative to the first element, where the first element is the element with the largest amplitude in the first matrix, and the second element is the Elements in the first matrix other than the first element.
The method of claim 29, wherein the first element and the second element are non-zero elements in the first matrix.
The method according to claim 29, wherein when the rank of the target downlink channel is greater than 1, the first element and the second element are elements corresponding to the same layer in the first matrix.
The method of claim 28, wherein the third information includes:

The amplitude and phase of the third element, wherein the third element is a non-zero element in the first matrix, the non-zero element includes an element whose amplitude is greater than or equal to the first threshold, or an element whose amplitude is not equal to 0 ;

In the case where the first matrix includes zero elements and non-zero elements, the third information further includes at least one of the following:

First position information, the first position information being the position information of non-zero elements in the first matrix;

Second position information, the second position information is the position information of the zero element in the first matrix.
The method according to any one of claims 28 to 32, wherein, in the case where the rank of the target downlink channel is greater than 1, the first matrix includes at least two layers;

The number of fourth elements in the first matrix corresponding to each of the at least two layers decreases layer by layer based on N, where N is the number of elements corresponding to each layer in the precoding matrix, or, The number of fourth elements corresponding to each of the at least two layers is less than N.
The method of claim 33, wherein the fourth element includes a first element and a second element, or the fourth element includes a third element.
The method of claim 33, wherein the third information further includes at least one of the following:

Third position information, the third position information is the position information of the fourth element in the first matrix;

Fourth position information, the fourth position information is the position information of non-zero elements in the first matrix except the fourth element;

Fifth position information, the fifth position information is the position information of the zero element in the first matrix and the position information of the fourth element.
The method according to any one of claims 22 to 27, wherein when the first channel information is a precoding matrix, the AI training data includes:

The fourth information of the second matrices of Y delay domains, wherein the Y second matrices are the Y with the largest amplitude, the X second matrices are obtained based on delay domain conversion processing of the precoding matrix, X is a positive integer, and Y is a positive integer less than or equal to X.
The method of claim 36, wherein the fourth information includes:

The position information of the fifth element and the amplitude and phase difference information of the sixth element relative to the fifth element, wherein the fifth element is the element with the largest amplitude in the second matrix, and the sixth element is the Elements in the second matrix other than the fifth element.
The method of claim 37, wherein the fifth element and the sixth element are non-zero elements in the second matrix.
The method according to claim 37, wherein when the rank of the target downlink channel is greater than 1, the fifth element and the sixth element are elements corresponding to the same layer in the second matrix.
The method of claim 36, wherein the fourth information includes:

The amplitude and phase of the seventh element, wherein the seventh element is a non-zero element in the second matrix, the non-zero element includes an element whose amplitude is greater than or equal to the first threshold, or an element whose amplitude is not equal to 0 ;

In the case where the second matrix includes zero elements and non-zero elements, the fourth information further includes at least one of the following:

Sixth position information, the sixth position information is the position information of non-zero elements in the second matrix;

Seventh position information, the seventh position information is the position information of the zero element in the second matrix.
The method according to any one of claims 36 to 40, wherein when the rank of the target downlink channel is greater than 1, the second matrix includes at least two layers, and the fourth information includes the first the amplitude and phase of the eighth element in the second matrix;

The number of eighth elements in the second matrix corresponding to each of the at least two layers decreases layer by layer on the basis of L, where L is the number of elements corresponding to each layer in the precoding matrix in the delay domain. number, or the number of eighth elements corresponding to each of the at least two layers is less than L.
The method of claim 41, wherein the eighth element includes a fifth element and a sixth element, or the eighth element includes a seventh element.
The method of claim 41, wherein the fourth information further includes at least one of the following:

Eighth position information, the eighth position information is the position information of the eighth element in the second matrix;

Ninth position information, the ninth position information is the position information of non-zero elements in the second matrix except the eighth element;

Tenth position information, the tenth position information is the position information of the zero element and the position information of the eighth element in the second matrix.
An information transmission device, applied to a terminal, the device includes:

An acquisition module, configured to acquire first information, where the first information includes first channel information of all subbands of the target downlink channel;

A first processing module, configured to perform first processing on the first information to obtain AI training data, where the first processing includes at least one of the following: screening processing, quantization processing, delay domain conversion processing, codebook Conversion processing and positive Cross-processing, the AI training data is used to train the first AI network model and/or the second AI network model, the first AI network model is used to process the second channel information into the first channel feature information, the The second AI network model is used to restore the first channel characteristic information to the second channel information, where the second channel information is the channel information of the target downlink channel;

A first sending module, configured to send second information to the first device, where the second information includes the AI training data.
An artificial intelligence AI network model training device, applied to the first device, the device includes:

A first receiving module, configured to receive second information from the terminal, where the second information includes AI training data, the AI training data is obtained based on first processing of the first information, and the first information includes a target The first channel information of all subbands of the downlink channel;

A training module, configured to train a first AI network model and/or a second AI network model according to the AI training data, the first AI network model being used to process the second channel information into the first channel feature information, the The second AI network model is used to restore the first channel characteristic information to the second channel information, where the second channel information is the channel information of the target downlink channel.
A communication device, including a processor and a memory, the memory stores a program or instructions that can be run on the processor, and when the program or instructions are executed by the processor, any one of claims 1 to 21 is implemented. The steps of the information transmission method described in the item, or the steps of implementing the artificial intelligence AI network model training method as described in any one of claims 22 to 43.
A readable storage medium that stores programs or instructions that, when executed by a processor, implement the steps of the information transmission method according to any one of claims 1 to 21, or The steps of implementing the artificial intelligence AI network model training method according to any one of claims 22 to 43.