WO2022217502A1 - Information processing method and apparatus, communication device, and storage medium - Google Patents

Abstract

The present application relates to the field of wireless communications, and discloses an information processing method and apparatus, a communication device, and a storage medium. The method comprises: obtaining n pieces of first channel information corresponding to n feedback cycles, the feedback cycles being feedback cycles of channel information, and n being a positive integer greater than 1; performing m times of different scales of splicing on the n pieces of first channel information to obtain m pieces of spliced channel information, the different scales of splicing being used for indicating that the number of the spliced first channel information in the m pieces of spliced channel information is different from each other, and m being a positive integer; and inputting the m pieces of spliced channel information into a neural network model for processing to obtain second channel information, wherein the neural network model is one of an encoding model and a decoding model. According to embodiments of the present application, the encoding performance or decoding performance of the channel information can be enhanced.

Description

Information processing method, device, communication device and storage medium technical field

The present application relates to the field of wireless communication, and in particular, to an information processing method, apparatus, communication device and storage medium.

Background technique

In a New Radio (New Radio, NR) system, the channel information feedback is a codebook-based feedback scheme.

In the codebook-based feedback scheme, the terminal device selects the optimal feedback matrix from the codebook according to the estimated channel, and the codebook itself has a finite nature, that is, the mapping from the estimated channel to the feedback matrix in the codebook The process is quantization lossy, which reduces the accuracy of the feedback channel information, which in turn reduces the performance of precoding.

In order to solve the problem of low accuracy caused by the codebook-based feedback scheme, the introduction of the feedback scheme based on neural network is discussed. The end decodes and recovers the channel information.

SUMMARY OF THE INVENTION

Embodiments of the present application provide an information processing method, apparatus, communication device, and storage medium, which can enhance the encoding performance or decoding performance of channel information. The technical solution is as follows:

According to one aspect of the present application, an information processing method is provided, the method comprising:

acquiring n pieces of first channel information corresponding to n feedback periods, where the feedback period is a feedback period of the channel information, and the n is a positive integer greater than 1;

The n pieces of first channel information are spliced with different scales m times to obtain m pieces of spliced channel information, and the splicing of different scales is used to indicate the number of pieces of the first channel information spliced in the m pieces of spliced channel information. The numbers are different from each other, and the m is a positive integer;

Inputting the m pieces of splicing channel information into a neural network model for processing to obtain second channel information;

Wherein, the neural network model is one of an encoding model and a decoding model.

According to an aspect of the present application, an information processing device is provided, the device comprising: an information acquisition module, an information splicing module, and an information processing module;

the information acquisition module, configured to acquire n pieces of first channel information corresponding to n feedback periods, where the feedback period is a feedback period of the channel information, and the n is a positive integer greater than 1;

The information splicing module is configured to perform m times of splicing of the n pieces of first channel information with different scales, to obtain m pieces of spliced channel information, and the splicing of different scales is used to indicate which pieces of the m pieces of spliced channel information are. The number of spliced first channel information is different from each other, and the m is a positive integer;

The information processing module is configured to input the m pieces of spliced channel information into a neural network model for processing to obtain second channel information;

Wherein, the neural network model is one of an encoding model and a decoding model.

According to an aspect of the present application, a terminal device is provided, the terminal device comprising: a processor; wherein,

the processor, configured to acquire n pieces of first channel information corresponding to n feedback periods, where the feedback period is a feedback period of the channel information, and the n is a positive integer greater than 1;

The processor is configured to perform m times of splicing of the n pieces of first channel information with different scales to obtain m pieces of spliced channel information, where the splicing of different scales is used to indicate the pieces of spliced channel information in the m pieces of splicing channel information The number of the first channel information is different from each other, and the m is a positive integer;

the processor, configured to input the m pieces of spliced channel information into a neural network model for processing to obtain second channel information;

Wherein, the neural network model is an encoding model.

According to an aspect of the present application, a network device is provided, the network device comprising: a processor and a transceiver connected to the processor; wherein,

the transceiver, configured to acquire n pieces of first channel information corresponding to n feedback periods, the feedback periods being the feedback periods of the channel information, and the n being a positive integer greater than 1;

The processor is configured to perform m times of splicing of the n pieces of first channel information with different scales to obtain m pieces of spliced channel information, where the splicing of different scales is used to indicate the pieces of spliced channel information in the m pieces of splicing channel information The number of the first channel information is different from each other, and the m is a positive integer;

the processor, configured to input the m pieces of spliced channel information into a neural network model for processing to obtain second channel information;

Wherein, the neural network model is a decoding model.

According to one aspect of the present application, a computer-readable storage medium is provided, and executable instructions are stored in the readable storage medium, and the executable instructions are loaded and executed by a processor to realize the information processing described in the above-mentioned aspects. method.

According to an aspect of the embodiments of the present application, a chip is provided, the chip includes a programmable logic circuit and/or program instructions, and when the chip runs on a computer device, it is used to implement the information processing described in the above aspect method.

According to one aspect of the present application, there is provided a computer program product, the computer program product or computer program comprising computer instructions stored in a computer-readable storage medium from which a processor The computer instructions are read and executed to implement the information processing method described in the above aspects.

The technical solutions provided by the embodiments of the present application include at least the following beneficial effects:

The n pieces of first channel information corresponding to n feedback cycles are spliced at different scales to obtain m pieces of spliced channel information, and then the m pieces of spliced channel information are processed by the neural network model to obtain the second channel information. The neural network model It is one of an encoding model and a decoding model, thereby realizing multi-scale utilization of the first channel information of different feedback periods, and enhancing the encoding performance or decoding performance corresponding to the channel information.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present application more clearly, the following briefly introduces the drawings that are used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

FIG. 1 is a schematic diagram of a network architecture provided by an exemplary embodiment of the present application;

FIG. 2 is a schematic diagram of a neural network provided by an exemplary embodiment of the present application;

3 is a schematic diagram of a convolutional neural network provided by an exemplary embodiment of the present application;

4 is a schematic diagram of a long short-term memory network provided by an exemplary embodiment of the present application;

5 is a schematic diagram of channel information feedback based on artificial intelligence provided by an exemplary embodiment of the present application;

6 is a schematic diagram of a channel information feedback system provided by an exemplary embodiment of the present application;

7 is a schematic diagram of channel recovery using historical feedback information provided by an exemplary embodiment of the present application;

8 is a flowchart of an information processing method provided by an exemplary embodiment of the present application;

FIG. 9 is a flowchart of an information processing method provided by an exemplary embodiment of the present application;

10 is a schematic diagram of encoding performance enhancement based on multi-scale information at the transmitting end provided by an exemplary embodiment of the present application;

11 is a schematic diagram of decoding performance enhancement based on multi-scale information at the receiving end provided by an exemplary embodiment of the present application;

12 is a schematic diagram of decoding performance enhancement based on multi-scale information at the receiving end provided by an exemplary embodiment of the present application;

13 is a schematic diagram of decoding performance enhancement based on multi-scale information at the receiving end provided by an exemplary embodiment of the present application;

14 is a structural block diagram of an information processing apparatus provided by an exemplary embodiment of the present application;

FIG. 15 is a schematic structural diagram of a communication device provided by an exemplary embodiment of the present application.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

The network architecture and service scenarios described in the embodiments of the present application are for the purpose of illustrating the technical solutions of the embodiments of the present application more clearly, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. The evolution of new business scenarios and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

Please refer to FIG. 1 , which shows a schematic diagram of a network architecture 100 provided by an embodiment of the present application. The network architecture 100 may include: a terminal device 10 , an access network device 20 and a core network device 30 .

The terminal device 10 may refer to a UE (User Equipment, user equipment), an access terminal, a subscriber unit, a subscriber station, a mobile station, a mobile station, a remote station, a remote terminal, a mobile device, a wireless communication device, a user agent or a user equipment. Optionally, the terminal device 10 may also be a cellular phone, a cordless phone, a SIP (Session Initiation Protocol, session initiation protocol) phone, a WLL (Wireless Local Loop, wireless local loop) station, a PDA (Personal Digital Assistant, personal digital processing ), handheld devices with wireless communication functions, computing devices or other processing devices connected to wireless modems, in-vehicle devices, wearable devices, terminal devices in 5GS (5th Generation System, fifth-generation mobile communication system) or future evolved Terminal equipment in a PLMN (Pub1ic Land Mobi1e Network, public land mobile communication network), etc., are not limited in this embodiment of the present application. For the convenience of description, the devices mentioned above are collectively referred to as terminal devices. The number of terminal devices 10 is usually multiple, and one or more terminal devices 10 may be distributed in a cell managed by each access network device 20 .

The access network device 20 is a device deployed in the access network to provide the terminal device 10 with a wireless communication function. The access network device 20 may include various forms of macro base stations, micro base stations, relay stations, access points, and the like. In systems using different wireless access technologies, the names of devices with access network device functions may be different, for example, in 5G NR systems, they are called gNodeBs or gNBs. As communication technology evolves, the name "Access Network Equipment" may change. For convenience of description, in the embodiments of the present application, the above-mentioned apparatuses for providing a wireless communication function for the terminal device 10 are collectively referred to as access network devices. Optionally, through the access network device 20, a communication relationship can be established between the terminal device 10 and the core network device 30. Exemplarily, in an LTE (Long Term Evolution, Long Term Evolution) system, the access network device 20 may be EUTRAN (Evolved Universal Terrestrial Radio Access Network, Evolved Universal Terrestrial Radio Access Network) or one or more eNodeBs in EUTRAN; In the 5G NR system, the access network device 20 may be a RAN (Radio Access Network, radio access network) or one or more gNBs in the RAN. In the embodiments of the present application, the network device refers to an access network device 20, such as a base station, unless otherwise specified.

The core network device 30 is a device deployed in the core network. The functions of the core network device 30 are mainly to provide user connection, user management and service bearer, and serve as an interface for the bearer network to provide an external network. For example, the core network equipment in the 5G NR system may include AMF (Access and Mobility Management Function) entity, UPF (User Plane Function, user plane function) entity and SMF (Session Management Function, session management function) entity function) entity and other equipment.

In an example, the access network device 20 and the core network device 30 communicate with each other through a certain air interface technology, such as the NG interface in the 5G NR system. The access network device 20 and the terminal device 10 communicate with each other through a certain air interface technology, such as a Uu interface.

The "5G NR system" in the embodiments of this application may also be referred to as a 5G system or an NR system, but those skilled in the art can understand its meaning. The technical solutions described in the embodiments of this application can be applied to LTE systems, 5G NR systems, and subsequent evolution systems of 5G NR systems, and can also be applied to systems such as NB-IoT (Narrow Band Internet of Things, narrowband other communication systems such as the Internet of Things) system, which is not limited in this application.

Before introducing the technical solutions of the present application, some background technical knowledge involved in the present application is introduced and explained.

For the 5G NR system, in the current Channel State Information (CSI) feedback design, the codebook-based scheme is mainly used to achieve channel feature extraction and feedback. That is, after the channel estimation is performed at the transmitting end, the precoding matrix that best matches the current channel is selected from the preset precoding codebook according to the result of the channel estimation according to a certain optimization criterion, and the matrix is converted through the feedback link of the air interface. The index information is fed back to the receiving end for the receiving end to implement precoding.

In recent years, artificial intelligence research represented by neural networks has achieved great results in many fields, and it will also have an important impact on people's production and life for a long time in the future.

Please refer to FIG. 2 , which shows a schematic diagram of a neural network provided by an embodiment of the present application. As shown in Figure 2, the basic structure of a simple neural network includes: input layer, hidden layer and output layer. Among them, the input layer is responsible for receiving data, the hidden layer is responsible for processing data, and the final result is generated in the output layer. As shown in Figure 1, each node represents a processing unit, which can also be considered to simulate a neuron. Multiple neurons form a layer of neural network, and multiple layers of information transmission and processing construct an overall neural network.

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

At the same time, with the development of deep learning, convolutional neural networks have also been further studied. Please refer to FIG. 3 , which shows a schematic diagram of a convolutional neural network provided by an embodiment of the present application. As shown in Figure 3, in a convolutional neural network, its basic structure includes: an input layer, multiple convolutional layers, multiple pooling layers, a fully connected layer and an output layer. The introduction of the convolutional layer and the pooling layer effectively controls the sharp increase of network parameters, limits the number of parameters and exploits the characteristics of local structures, and improves the robustness of the algorithm.

Recurrent Neural Network (RNN) is a type of recurrent neural network that takes sequence data as input, performs recursion in the evolution direction of the sequence, and connects all nodes (recurrent units) in a chain. As the most commonly used and most traditional deep learning model in Natural Language Processing (NLP), the RNN network reads sequence data step by step in order for processing, which is similar to the way humans understand text. , to understand word by word.

Long Short-Term Memory (LSTM) as a variant model of RNN is shown in Figure 4. Its essence lies in the introduction of the concept of cell state. Unlike RNN, which only considers the most recent state, the cell of LSTM The state determines which states should be left behind and which states should be forgotten, solving the shortcomings of traditional RNNs in long-term memory.

In view of the great success of artificial intelligence (AI) technology in computer vision, natural language processing, etc., the communication field has begun to try to use AI technology to seek new technical ideas to solve technical problems limited by traditional methods, such as depth study. The neural network architecture commonly used in deep learning is nonlinear and data-driven. It can perform feature extraction on the actual channel matrix data and restore the channel matrix information compressed and fed back by the terminal side as much as possible on the base station side. It provides the possibility for the terminal side to reduce the feedback overhead of channel information.

As shown in FIG. 5 , in the AI-based channel information feedback, the channel information is regarded as an image to be compressed 501 , and the channel information is compressed by a deep learning autoencoder 502 to obtain a compressed channel image 503 . The receiving end uses the deep learning self-decoder 504 to reconstruct the compressed channel image 503 to obtain restored channel information 505, which can preserve the channel information to a greater extent.

A typical channel information feedback system is shown in Figure 6. The entire feedback system is divided into an encoder part and a decoder part, which are deployed at the transmitter and receiver respectively. After the transmitter obtains the channel information through channel estimation, it compresses and encodes the channel information matrix through the neural network of the encoder, and feeds back the compressed bit stream to the receiver through the air interface feedback link. The stream recovers the channel information to obtain complete channel information. The structure shown in Figure 6 uses several fully connected layers for encoding at the encoder and a residual network structure for decoding at the decoder. Under the condition that the encoding and decoding framework remains unchanged, the network model structure inside the encoder and decoder can be flexibly designed.

The channel information feedback in the current 5G NR standard is a codebook-based feedback scheme. However, this scheme only selects the optimal feedback matrix from the codebook according to the estimated channel, and the codebook itself is limited, that is, the mapping process from the estimated channel to the channel in the codebook is quantization lossy , which reduces the accuracy of the feedback channel information, and further reduces the performance of precoding.

In order to solve the problem of low accuracy caused by the codebook-based feedback scheme, the introduction of the neural network-based feedback scheme is discussed. The channel information feedback scheme based on neural network is a scheme that directly encodes and compresses the channel information obtained after channel estimation, which can alleviate the accuracy problem of the codebook-based scheme.

The channel information fed back in different feedback periods has a certain historical correlation, and the channel recovery performance of the current feedback period can be enhanced by using the historical correlation. That is, the channel information fed back in different feedback periods constitutes historical feedback information, and the historical feedback information is an image or sequence, which is used as the input of the decoder.

Referring to FIG. 7 in conjunction, it shows a schematic diagram of channel recovery using historical feedback information. The entire feedback system includes an encoder at the transmitter and a decoder at the receiver.

The transmitting end compresses and encodes the channel information into a bit stream through the encoder in different feedback cycles. In this embodiment, the maximum historical traceability scale is set to n, that is, in n feedback cycles, the transmitting end uses the encoder to compress and encode the channel information {H_1 , . When decoding the feedback channel of the nth feedback cycle, the receiving end simultaneously uses the first n-1 feedback cycles and the feedback bit stream of the nth feedback cycle as the decoder input, and the output of the decoder network is the restored channel H. '_n.

If the channel recovery is performed using the historical feedback information as shown in FIG. 7 , the length of the input historical feedback information directly affects the performance of the network after training. However, the setting of the length of the input historical feedback information is often empirical, and it is difficult to judge whether the current input length is the optimal length. Reflecting on the channel, for different channel environments (eg, different moving speeds of the terminal), when the historical feedback information is used for performance enhancement, the optimal lengths corresponding to the historical feedback information are different.

In view of the above problems, the technical solution of the present application adopts the above-mentioned scheme of utilizing historical correlation and using multiple first channel information corresponding to multiple feedback cycles to enhance the performance of the current feedback cycle for both the sending end and the receiving end. Further, the transmitting end (or the receiving end) utilizes the first channel information of different feedback periods in multiple scales, thereby enhancing the encoding performance or decoding performance corresponding to the channel information.

Hereinafter, the technical solutions of the present application will be introduced and described through several embodiments.

FIG. 8 shows a flowchart of an information processing method provided by an exemplary embodiment of the present application. The method can be applied to the network architecture shown in FIG. 1, and the method can include the following steps (802-806):

Step 802: Acquire n pieces of first channel information corresponding to n feedback periods, where the feedback period is a feedback period of the channel information.

where n is a positive integer greater than 1. That is, the communication device acquires a plurality of first channel information corresponding to a plurality of feedback periods.

In a mobile communication system, a terminal device needs to periodically feed back channel information to a network device according to a feedback period, or periodically determine channel information with different feedback periods through channel estimation. Denote the current feedback cycle as the nth feedback cycle, and go back n-1 feedback cycles, there are a total of n feedback cycles, including: the first feedback cycle, the second feedback cycle, ..., the nth feedback cycle .

The first channel information is information related to the channel information feedback procedure.

Exemplarily, for the terminal device that is the transmitter of the channel information, it needs to perform channel estimation by measuring the reference signal to determine the channel information of different feedback periods, then the first channel information is the channel information obtained through channel estimation. . The terminal device determines n pieces of channel information corresponding to n feedback periods.

Exemplarily, for a network device serving as a receiver of channel information, it needs to receive compressed channel information from a transmitter to perform channel recovery, and the first channel information is a compressed bit stream corresponding to the channel information. The network device receives compressed bit streams corresponding to n pieces of channel information corresponding to n feedback periods.

Step 804: Perform m times of splicing of the n pieces of first channel information with different scales to obtain m pieces of spliced channel information, and the splicing of different scales is used to indicate that the number of pieces of first channel information spliced in the m pieces of spliced channel information is different from each other. same.

In order to utilize the historical correlation between the n pieces of first channel information, the n pieces of first channel information can be spliced, and the spliced channel information can be obtained after the splicing.

In the embodiment of the present application, the communication device performs m times of splicing on the n pieces of first channel information, and in each splicing process, splices the first channel information in the n pieces of first channel information, and the spliced pieces of information are spliced each time. The number of the first channel information is different, that is, m times of splicing are m times of splicing of different scales. In this embodiment of the present application, m may be a positive integer not less than 2.

Exemplarily, n is 6, there are 6 pieces of first channel information in total, including information 1 to information 6, m is 3, and 3 pieces of concatenated channel information are obtained. Splicing channel information 1 splices information 1 to information 6, splicing 6 first channel information; splicing channel information 2 splicing information 3 to information 6, splicing 3 first channel information; splicing channel information 3 splicing information 5 and information 6, the two first channel information is spliced. The number of pieces of first channel information spliced in the above three pieces of splicing channel information is different from each other, which can be understood as splicing three times with different scales.

It can be understood that the splicing channel information may also include one piece of first channel information. Exemplarily, the splicing channel information includes: the nth first channel information corresponding to the nth feedback period.

Step 806: Input the m pieces of spliced channel information into the neural network model for processing to obtain second channel information.

On the one hand, since the splicing channel information splices the first channel information of different feedback periods, the first channel information of different feedback periods has a certain historical correlation, and the performance can be enhanced by using the historical correlation. On the other hand, since the number of the first channel information spliced by the m pieces of splicing channel information is different from each other, and the lengths of the m pieces of splicing channel information input to the neural network model are different, the first channel information can be multi-scaled. use.

It can be understood that, the embodiment of the present application does not limit the model structure of the neural network. Exemplarily, the model structure of the neural network includes, but is not limited to, a fully connected neural network, a convolutional neural network, a recurrent neural network, and a long short-term memory network.

Among them, the neural network model is one of an encoding model and a decoding model. In the feedback scheme based on neural network, the encoding model of the sender and the decoding model of the receiver are two models that match each other. In the embodiment of the present application, the encoding model refers to a model used to encode channel information to generate a compressed compressed bit stream; the decoding model refers to a model used to decode the received compressed bit stream to restore A model of channel information. It can be understood that the encoding model and the decoding model can also be understood as: a channel state information encoding model and a channel state information decoding model; a channel encoding model and a channel decoding model; a modulation model and a demodulation model, etc. No restrictions apply.

The second channel information is the information output after the neural network model processes the concatenated channel information of different scales.

Exemplarily, for a terminal device serving as a sender of channel information, it needs to compress and encode the channel information by using an encoding model, and the second channel information is a compressed bit stream corresponding to the channel information. The terminal device obtains a compressed bit stream corresponding to the channel information by inputting the m pieces of spliced channel information into the coding model for processing.

Exemplarily, for the network device serving as the receiving end of the channel information, it needs to decode the received compressed bit stream to restore the channel information, and the second channel information is the restored channel information. The network device obtains the restored channel information by inputting the m pieces of spliced channel information into the decoding model for processing.

To sum up, in the technical solution of the present application, n pieces of first channel information corresponding to n feedback cycles are spliced with different scales to obtain m pieces of spliced channel information, and then the m pieces of spliced channel information are processed by a neural network model , obtain the second channel information, the neural network model is one of the encoding model and the decoding model, so as to realize the multi-scale utilization of the first channel information of different feedback periods, and enhance the encoding performance or decoding performance corresponding to the channel information.

In an exemplary embodiment, the communication device performs multi-scale utilization of the first channel information of different feedback periods based on the granularity information.

FIG. 9 shows a flowchart of an information processing method provided by an exemplary embodiment of the present application. The method can be applied to the network architecture shown in FIG. 1, and the method can include the following steps (902-908):

Step 902: Acquire n pieces of first channel information corresponding to n feedback periods, where the feedback period is a feedback period of the channel information.

where n is a positive integer greater than 1.

Optionally, the n pieces of first channel information form a sequence or a feature map.

In a possible implementation manner, the communication device acquires n pieces of first channel information corresponding to n feedback cycles, and obtains n pieces of first channel information represented by a sequence.

Exemplarily, n is 6, and there are 6 pieces of first channel information in total, including information 1 to information 6, forming a sequence {information 1, information 2, information 3, information 4, information 5, information 6}.

In another possible implementation manner, the communication device acquires n pieces of first channel information corresponding to n feedback cycles, processes the n pieces of first channel information through the first neural network layer, and obtains n pieces of information represented by the feature map first channel information.

The first neural network layer is a neural network structure that supports the use of feature maps to represent the first channel information. Optionally, the first neural network layer amplifies the dimension of the first channel information and converts it into the dimension of the channel information matrix. Optionally, the first neural network layer includes a fully connected layer.

Exemplarily, n is 6, and there are 6 pieces of first channel information, including information 1 to 6, which are processed by the neural network layer to generate feature maps {information 1', information 2', information 3', information 4', Information 5', Information 6'}.

Step 904: Obtain granularity information, where the granularity information is used to indicate the granularity s.

where s is a positive integer.

Optionally, s is a preset fixed value.

Optionally, s is a value adjusted according to different channel conditions. Exemplarily, if the current channel situation requires as much splicing channel information as possible, the granularity s corresponds to a smaller value; if the current channel situation does not require using as much splicing channel information as possible, the granularity s corresponds to a larger value.

Step 906, based on the granularity information, perform m times of splicing with different scales on the n pieces of first channel information to obtain m pieces of spliced channel information.

Wherein, the difference between the number of pieces of first channel information spliced in any two pieces of spliced channel information is an integer multiple of s.

Since the granularity s is indicated in the granularity information, according to the granularity s, the next spliced channel information by the communication device is s less first channel information than the previous spliced channel information.

Optionally, the feedback periods corresponding to the first channel information spliced in the spliced channel information are continuous in the time dimension. That is, the communication device splices the x pieces of first channel information corresponding to consecutive x feedback cycles to obtain a spliced channel information, where x is a positive integer.

Optionally, the m pieces of spliced channel information respectively include the nth first channel information corresponding to the nth feedback period. It can be understood that the nth feedback cycle is the current feedback cycle, and the spliced channel information includes the nth first channel information corresponding to the nth feedback cycle, so that the spliced channel information can better reflect the current channel state. .

Optionally, in the first channel information spliced by the previous splicing channel information, remove the first s pieces of first channel information in the time dimension, and splicing the remaining first channel information to obtain the next splicing channel information.

In a possible implementation manner, step 906 is implemented as: splicing the (a-1)*s+1 th first channel information to the n th first channel information in the n pieces of first channel information, to obtain The a-th concatenated channel information, a is a positive integer that increases by one from 1, and (a-1)*s+1 is less than n.

Exemplarily, n is 10, there are 10 pieces of first channel information in total, including information 1 to 10, and s is 3, then the spliced channel information includes: {information 1, information 2, ..., information 10}, {information 4, info5,...,info10}, {info7, info8, info9, info10}, {info10}.

Step 908: Input the m pieces of spliced channel information into the neural network model for processing to obtain second channel information.

In a possible implementation manner, the neural network model performs optimal joint utilization of m spliced channel information. That is, step 908 is replaced by: inputting m pieces of splicing channel information into m second neural network layers respectively for processing to obtain m channel features; performing weighted splicing on m channel features to obtain splicing channel features, and then splicing channel features. The features are input into the third neural network layer to obtain the second channel information.

In another possible implementation manner, the neural network model performs adaptive optimal selection and utilization of m pieces of spliced channel information. That is, step 908 is replaced by: inputting m pieces of spliced channel information into m second neural network layers for processing, to obtain m channel features; selecting target channel features from m channel features, and inputting the target channel features into the first Four neural network layers to obtain the second channel information.

The present application does not limit the specific implementations of the second neural network layer, the third neural network layer and the fourth neural network layer.

To sum up, in the technical solution of the present application, n pieces of first channel information corresponding to n feedback cycles are spliced with different scales to obtain m pieces of spliced channel information, and then the m pieces of spliced channel information are processed by a neural network model , obtain the second channel information, the neural network model is one of the encoding model and the decoding model, so as to realize the multi-scale utilization of the first channel information of different feedback periods, and enhance the encoding performance or decoding performance corresponding to the channel information.

According to the technical solution of the present application, based on the granularity information, n pieces of first channel information are spliced with different scales, because the granularity information indicates that the difference between the numbers of the first channel information spliced in any two spliced channel information is the granularity s Integer multiples of , can guarantee a reasonable number of splicing channel information.

The technical solution of the present application supports adaptive optimal selection utilization or optimal joint utilization of splicing channel information of different scales.

Based on the above embodiment, both the transmitting end and the receiving end may adopt the scheme of splicing channel information based on multi-scale as in the above embodiment for performance enhancement. Specifically, it includes the following two situations:

· In response to the neural network model being an encoding model, the first channel information includes channel information obtained through channel estimation, and the second channel information includes a compressed bit stream corresponding to the channel information.

That is, a coding model is set at the sending end corresponding to the terminal device, and the coding performance corresponding to the compression process of the current feedback cycle is enhanced by using channel information of different scales.

· In response to the neural network model being a decoding model, the first channel information includes a compressed bit stream corresponding to the channel information, and the second channel information includes restored channel information.

That is, a decoding model is set at the receiving end corresponding to the network device, and the decoding performance corresponding to the decompression process of the current feedback cycle is enhanced by using compressed bit streams of different scales.

Hereinafter, the technical solutions of the present application will be exemplarily described through the following examples.

Referring to FIG. 10 , it shows a schematic diagram of encoding performance enhancement based on multi-scale information at the transmitting end provided by an exemplary embodiment of the present application.

In this embodiment, the channel information of n feedback cycles are respectively denoted as H_1 to H_n, and the compressed bit stream B_n of the channel information corresponding to the current nth feedback cycle is output. This embodiment corresponds to a joint enhancement mechanism based on LSTM. In this embodiment, the combination of channel information of different scales is regarded as sequences of different lengths as the input of the encoder.

First, perform dimension compression on the channel information {H_1,...,H_n} of n feedback cycles through a flatten process to obtain {H'_1,...,H'_n}. Set the value of granularity information s to 1 and the maximum historical scale to n, then there are n different scales. The number of LSTM structures required by the encoder is n.

Take {H'_1,...,H'_n}, {H'_2,...,H'_n} to {H'_n} as a sequence, respectively as the 1st LSTM structure to the nth LSTM structure , while each LSTM structure only outputs the network output of the last cycle of the LSTM.

Further, the outputs {R_1, .

In this embodiment, in the design of the neural network model at the sending end, optimal joint utilization of multiple scales is performed for channel information of different feedback periods, so as to enhance the coding performance.

Referring to FIG. 11 , it shows a schematic diagram of enhancing decoding performance based on multi-scale information at the receiving end provided by an exemplary embodiment of the present application.

In this embodiment, the compressed bit streams of n feedback cycles are respectively denoted as B_1 to B_n, and the restored channel information H'_n corresponding to the current nth feedback cycle is output. This embodiment corresponds to a joint enhancement mechanism based on a convolutional neural network. In this embodiment, the combination of channel information of different scales is regarded as images of different lengths as the input of the decoder.

(a) in FIG. 11 corresponds to an encoder structure. First, the input channel information is converted into a one-dimensional vector H_n input model. The model adopts a fully connected neural network, including M layers of fully connected layers, and the last fully connected layer converts the information into a compressed bit stream B_n, where M is a positive integer. Network layers such as activation layer, normalization layer, and quantization layer can also be added between each fully connected layer.

(b) in FIG. 11 corresponds to a decoder structure. The compressed bitstream {B_1,...,B_n} of n feedback cycles is used as the input of the decoder, and each compressed bitstream firstly enlarges and converts the dimension into the dimension of the channel information matrix through the fully connected layer to generate the feature map {B' _1, ..., B'_n}. Set the value of granularity information s to 1 and the maximum historical scale to n, then there are n different scales. The number of residual block structures required by the decoder is n.

Concatenate {B'_1,...,B'_n}, {B'_2,...,B'_n} to {B'_n} according to the channel dimension, and use them as the first residual block structure respectively Input to the nth residual block structure. Further, the outputs {R_1,...,R_n} of each residual block are spliced according to the channel dimension, and combined with a 1X1 convolutional layer, and finally reconstructed and restored through the residual block to obtain the restored channel Information H'_n.

In this embodiment, in the design of the neural network model at the receiving end, multi-scale optimal joint utilization is performed for the compressed bit streams of channel information of different feedback periods, so as to enhance the decoding performance.

Referring to FIG. 12 , it shows a schematic diagram of enhancing decoding performance based on multi-scale information at the receiving end provided by an exemplary embodiment of the present application.

In this embodiment, the compressed bit streams of n feedback cycles are respectively denoted as B_1 to B_n, and the restored channel information H'_n corresponding to the current nth feedback cycle is output. This embodiment corresponds to a joint enhancement mechanism based on a recurrent neural network. In this embodiment, the combination of compressed bit streams of different scales is regarded as sequences of different lengths as the input of the decoder.

(a) in FIG. 12 corresponds to an encoder structure. The feature extractor uses a convolutional neural network, including M layers of convolutional neural network layers, and the last fully connected layer converts the information into a compressed bit stream B_n, where M is a positive integer. Network layers such as activation layer, normalization layer, and quantization layer can also be added between each neural network layer.

(b) in FIG. 12 corresponds to a decoder structure. The compressed bitstream {B_1, . . . , B_n} of n feedback cycles is taken as the decoder input. Set the value of granularity information s to 1 and the maximum historical scale to n, then there are n different scales. The number of RNN structures required by the decoder is n.

Take {B_1,...,B_n}, {B_2,...,B_n} to {B_n} as the sequence, respectively as the input of the first RNN structure to the nth RNN structure, and each RNN structure only outputs RNN The network output for the last loop of .

Further, the outputs {R_1,..., R_n} of each RNN structure are spliced according to the channel dimension, and combined with a 1X1 convolutional layer, and the dimension is enlarged and converted into the dimension of the channel information matrix through the fully connected layer, Finally, reconstruct and restore the residual block to obtain the restored channel information H'_n.

In this embodiment, in the design of the neural network model at the receiving end, multi-scale optimal joint utilization is performed for the compressed bit streams of channel information of different feedback periods, so as to enhance the decoding performance.

Referring to FIG. 13 , it shows a schematic diagram of enhancing decoding performance based on multi-scale information at the receiving end provided by an exemplary embodiment of the present application.

In this embodiment, the compressed bit streams of n feedback cycles are respectively denoted as B_1 to B_n, and the restored channel information H'_n corresponding to the current nth feedback cycle is output. This embodiment corresponds to an adaptive selection mechanism based on a long short-term memory network. In this embodiment, the combination of compressed bit streams of different scales is regarded as sequences of different lengths as the input of the decoder.

(a) in FIG. 13 corresponds to an encoder structure. The convolutional neural network and Inception structure are used, that is, the channel information H_n is extracted by using different convolution kernel sizes, the channel dimension is spliced to the feature map, the feature map is merged through the 1X1 convolution layer, and finally the fully connected layer is used. Convert the information into an output compressed bitstream B_n. At the same time, network layers such as activation layer, normalization layer, and quantization layer can also be added between each neural network layer.

(b) in FIG. 13 corresponds to a decoder structure. The compressed bitstream {B_1, . . . , B_n} of n feedback cycles is taken as the decoder input. Set the value of granularity information s to 1 and the maximum historical scale to n, then there are n different scales. The number of LSTM structures required by the decoder is n.

Take {B_1,...,B_n}, {B_2,...,B_n} to {B_n} as a sequence, respectively as the input of the first LSTM structure to the nth LSTM structure, and each LSTM structure only outputs LSTM The network output for the last loop of .

Further, the outputs {R_1, ..., R_n} of each LSTM structure are spliced into a tensor P by channel, channel merging is performed through a 1X1 convolutional layer, and one-hot (one-hot) is output through two fully connected layers. The selection vector of , and the selection vector is dot-multiplied with the tensor P according to the channel dimension to complete the selection operation from multi-branch to single branch. After that, the dimension is enlarged and converted into the dimension of the channel information matrix through the fully connected layer, and finally reconstructed and restored through the residual block to obtain the restored channel information H'_n.

In this embodiment, in the design of the neural network model at the receiving end, multi-scale adaptive selection and utilization are performed for the compressed bit streams of channel information of different feedback periods, so as to enhance the decoding performance.

It can be understood that the neural network model structure shown in the above examples does not constitute a limitation on the technical solutions of the present application. Different data characteristics or channel characteristics will have different influences on the selection of the above models. That is, the model selection needs to match the current data features or channel features. The neural network model can be adjusted according to different channel data, for example: modify the feature extractor of the encoder to other feature extraction networks, replace the LSTM with other forms of recurrent neural network modules, etc.

It should be noted that, the foregoing method embodiments may be implemented separately, or may be implemented in combination, which is not limited in this application.

In each of the above-mentioned embodiments, the steps performed by the terminal device can be independently implemented as an information processing method on the terminal device side, and the steps performed by the network device can be implemented independently as an information processing method on the network device side.

The following are apparatus embodiments of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

FIG. 14 shows a structural block diagram of an information processing apparatus provided by an exemplary embodiment of the present application. The apparatus can be implemented as a communication device, or be implemented as a part of a communication device. The device includes: an information acquisition module 1401, an information splicing module module 1402 and information processing module 1403;

The information acquisition module 1401 is configured to acquire n pieces of first channel information corresponding to n feedback periods, where the feedback period is the feedback period of the channel information, and the n is a positive integer greater than 1;

The information splicing module 1402 is configured to perform m times of splicing of the n pieces of first channel information with different scales to obtain m pieces of spliced channel information, and the splicing of different scales is used to indicate that among the m pieces of spliced channel information. The numbers of the spliced first channel information are different from each other, and the m is a positive integer;

The information processing module 1403 is configured to input the m pieces of spliced channel information into a neural network model for processing to obtain second channel information;

Wherein, the neural network model is one of an encoding model and a decoding model.

In an optional embodiment, the information splicing module 1402 includes: a granular information acquisition sub-module and an information splicing sub-module;

The granularity information acquisition sub-module is used to acquire granularity information, the granularity information is used to indicate the granularity s, and the s is a positive integer;

The information splicing sub-module is configured to perform m times of splicing with different scales on the n pieces of first channel information based on the granularity information, to obtain the m pieces of splicing channel information, where any two pieces of splicing channel information are spliced The difference between the numbers of the first channel information is an integer multiple of the s.

In an optional embodiment, the m pieces of the spliced channel information respectively include the nth first channel information corresponding to the nth feedback period.

In an optional embodiment, the information splicing submodule is configured to combine the (a-1)*s+1th first channel information from the nth first channel information to the nth first channel information The channel information is spliced to obtain the a-th spliced channel information, where the a is a positive integer that increases by one from 1, and the (a-1)*s+1 is less than the n.

In an optional embodiment, the information acquisition module 1401 is configured to acquire n pieces of first channel information corresponding to n feedback cycles, and obtain the n pieces of first channel information represented by a sequence;

or,

The information acquisition module 1401 is configured to acquire n pieces of first channel information corresponding to n feedback cycles, and process the n pieces of first channel information through the first neural network layer to obtain the n pieces of information represented by the feature map. first channel information.

In an optional embodiment, the information processing module 1403 is configured to: input the m pieces of spliced channel information into m second neural network layers respectively for processing to obtain m channel features; The channel features are weighted and spliced to obtain spliced channel features, and the spliced channel features are input into the third neural network layer to obtain the second channel information; or, the target channel feature is selected from the m channel features, and the The target channel feature is input into the fourth neural network layer to obtain the second channel information.

In an optional embodiment, in response to the neural network model being a coding model, the first channel information includes channel information obtained through channel estimation, and the second channel information includes compressed bits corresponding to the channel information flow.

In an optional embodiment, in response to the neural network model being a decoding model, the first channel information includes a compressed bit stream corresponding to the channel information, and the second channel information includes the restored channel information.

It should be noted that, when the device provided in the above embodiment realizes its functions, only the division of the above functional modules is used as an example for illustration. In practical applications, the above functions can be allocated to different functional modules according to actual needs. That is, the content structure of the device is divided into different functional modules to complete all or part of the functions described above.

Regarding the apparatus in the above-mentioned embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be described in detail here.

FIG. 15 shows a schematic structural diagram of a communication device (terminal device or network device) provided by an exemplary embodiment of the present application. The communication device includes: a processor 1501 , a receiver 1502 , a transmitter 1503 , a memory 1504 and a bus 1505 .

The processor 1501 includes one or more processing cores, and the processor 1501 executes various functional applications and information processing by running software programs and modules.

The receiver 1502 and the transmitter 1503 may be implemented as a communication component, which may be a communication chip.

The memory 1504 is connected to the processor 1501 through the bus 1505 .

The memory 1504 may be configured to store at least one instruction, and the processor 1501 may be configured to execute the at least one instruction to implement the various steps in the above method embodiments.

Additionally, memory 1504 may be implemented by any type or combination of volatile or non-volatile storage devices including, but not limited to, magnetic or optical disks, electrically erasable and programmable Read Only Memory (Electrically-Erasable Programmable Read Only Memory, EEPROM), Erasable Programmable Read Only Memory (EPROM), Static Random Access Memory (SRAM), Read Only Memory (Read-Only Memory, ROM), magnetic memory, flash memory, programmable read-only memory (Programmable Read-Only Memory, PROM).

Wherein, when the communication device is implemented as a terminal device, the processors and transceivers involved in the embodiments of the present application may perform the steps performed by the terminal device in any of the above-mentioned methods shown in FIG. 8 to FIG. 10 , here No longer.

In a possible implementation manner, when the communication device implements the terminal device,

the processor, configured to acquire n pieces of first channel information corresponding to n feedback periods, where the feedback period is a feedback period of the channel information, and the n is a positive integer greater than 1;

The processor is configured to perform m times of splicing of the n pieces of first channel information with different scales to obtain m pieces of spliced channel information, where the splicing of different scales is used to indicate the pieces of spliced channel information in the m pieces of splicing channel information The number of the first channel information is different from each other, and the m is a positive integer;

the processor, configured to input the m pieces of spliced channel information into a neural network model for processing to obtain second channel information;

Wherein, the neural network model is an encoding model.

Wherein, when the communication device is implemented as a network device, the processors and transceivers involved in the embodiments of the present application may execute any of the methods shown in FIG. 8 to FIG. 9 and FIG. 11 to FIG. 13 above. The steps to be performed are not repeated here.

In a possible implementation manner, when the communication device is implemented as a network device,

the transceiver, configured to acquire n pieces of first channel information corresponding to n feedback periods, the feedback periods being the feedback periods of the channel information, and the n being a positive integer greater than 1;

The processor is configured to perform m times of splicing of the n pieces of first channel information with different scales to obtain m pieces of spliced channel information, where the splicing of different scales is used to indicate the pieces of spliced channel information in the m pieces of splicing channel information The number of the first channel information is different from each other, and the m is a positive integer;

the processor, configured to input the m pieces of spliced channel information into a neural network model for processing to obtain second channel information;

Wherein, the neural network model is a decoding model.

Embodiments of the present application further provide a computer-readable storage medium, where a computer program is stored in the storage medium, and the computer program is used to be executed by a processor of a terminal device to implement the above-mentioned information processing method on the terminal device side.

Embodiments of the present application further provide a computer-readable storage medium, where a computer program is stored in the storage medium, and the computer program is configured to be executed by a processor of a network device to implement the above-mentioned information processing method on the network device side.

Optionally, the computer-readable storage medium may include: ROM (Read-Only Memory, read-only memory), RAM (Random-Access Memory, random access memory), SSD (Solid State Drives, solid-state hard disk), or an optical disk. The random access memory may include ReRAM (Resistance Random Access Memory, resistive random access memory) and DRAM (Dynamic Random Access Memory, dynamic random access memory).

Embodiments of the present application further provide a chip, where the chip includes a programmable logic circuit and/or program instructions, and when the chip runs on a terminal device, it is used to implement the above-mentioned information processing method on the terminal device side.

An embodiment of the present application further provides a chip, where the chip includes a programmable logic circuit and/or program instructions, and when the chip runs on a network device, it is used to implement the above-mentioned information processing method on the network device side.

Embodiments of the present application further provide a computer program product or computer program, where the computer program product or computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium, and the processor of the terminal device can download the computer from the computer. The readable storage medium reads and executes the computer instructions to implement the above-mentioned information processing method on the terminal device side.

Embodiments of the present application also provide a computer program product or computer program, where the computer program product or computer program includes computer instructions, where the computer instructions are stored in a computer-readable storage medium, and a processor of a network device can download from the computer The readable storage medium reads and executes the computer instructions to implement the information processing method on the network device side.

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

In the description of the embodiments of the present application, the term "corresponding" may indicate that there is a direct or indirect corresponding relationship between the two, or may indicate that there is an associated relationship between the two, or indicate and be instructed, configure and be instructed configuration, etc.

As used herein, "plurality" refers to two or more. "And/or", which describes the association relationship of the associated objects, means that there can be three kinds of relationships, for example, A and/or B, which can mean that A exists alone, A and B exist at the same time, and B exists alone. The character "/" generally indicates that the associated objects are an "or" relationship.

In addition, the numbering of the steps described in this document only exemplarily shows a possible execution sequence between the steps. In some other embodiments, the above steps may also be executed in different order, such as two different numbers. The steps are performed at the same time, or two steps with different numbers are performed in a reverse order to that shown in the figure, which is not limited in this embodiment of the present application.

Those skilled in the art should realize that, in one or more of the above examples, the functions described in the embodiments of the present application may be implemented by hardware, software, firmware, or any combination thereof. When implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage medium can be any available medium that can be accessed by a general purpose or special purpose computer.

The above are only exemplary embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection of the present application. within the range.

Claims (21)

1. An information processing method, characterized in that the method comprises:

   acquiring n pieces of first channel information corresponding to n feedback periods, where the feedback period is a feedback period of the channel information, and the n is a positive integer greater than 1;

   The n pieces of first channel information are spliced with different scales m times to obtain m pieces of spliced channel information, and the splicing of different scales is used to indicate the number of pieces of the first channel information spliced in the m pieces of spliced channel information. The numbers are different from each other, and the m is a positive integer;

   Inputting the m pieces of splicing channel information into a neural network model for processing to obtain second channel information;

   Wherein, the neural network model is one of an encoding model and a decoding model.
2. The method according to claim 1, wherein, performing m times of splicing of the n pieces of first channel information with different scales to obtain m pieces of spliced channel information, comprising:

   Obtain granularity information, where the granularity information is used to indicate the granularity s, and the s is a positive integer;

   Based on the granularity information, the n pieces of first channel information are spliced m times of different scales to obtain the m pieces of spliced channel information. The difference is an integer multiple of the s.
3. The method of claim 2, wherein:

   The m pieces of the spliced channel information respectively include the nth first channel information corresponding to the nth feedback period.
4. The method according to claim 3, wherein, based on the granularity information, the n pieces of first channel information are spliced m times of different scales to obtain the m pieces of spliced channel information, comprising:

   Splicing the (a-1)*s+1 th first channel information to the n th first channel information in the n first channel information to obtain the a th splicing channel information, where the a is from 1 starts with a positive integer incremented by one, and the (a-1)*s+1 is less than the n.
5. The method according to any one of claims 1 to 4, wherein the acquiring n pieces of first channel information corresponding to n feedback cycles comprises:

   acquiring n pieces of first channel information corresponding to n feedback cycles, and obtaining the n pieces of first channel information represented by a sequence;

   or,

   Acquire n pieces of first channel information corresponding to n feedback cycles, process the n pieces of first channel information through a first neural network layer, and obtain the n pieces of first channel information represented by a feature map.
6. The method according to any one of claims 1 to 5, wherein the inputting the m pieces of spliced channel information into a neural network model for processing to obtain the second channel information, comprising:

   The m pieces of splicing channel information are respectively input into m second neural network layers for processing to obtain m channel features;

   Weighted splicing is performed on the m channel features to obtain spliced channel features, and the spliced channel features are input into the third neural network layer to obtain the second channel information;

   or,

   A target channel feature is selected from the m channel features, and the target channel feature is input into the fourth neural network layer to obtain the second channel information.
7. The method according to any one of claims 1 to 6, wherein,

   In response to the neural network model being an encoding model, the first channel information includes channel information obtained through channel estimation, and the second channel information includes a compressed bit stream corresponding to the channel information.
8. The method according to any one of claims 1 to 6, wherein,

   In response to the neural network model being a decoding model, the first channel information includes a compressed bit stream corresponding to the channel information, and the second channel information includes the restored channel information.
9. An information processing device, characterized in that the device comprises: an information acquisition module, an information splicing module and an information processing module;

   The information acquisition module is used to acquire n first channel information corresponding to n feedback cycles, where the feedback cycles are the feedback cycles of channel information, and the n is a positive integer greater than 1;

   The information splicing module is configured to perform m times of splicing with different scales on the n pieces of first channel information, to obtain m pieces of splicing channel information, and the splicing of different scales is used to indicate all the m pieces of splicing channel information. The number of spliced first channel information is different from each other, and the m is a positive integer;

   The information processing module is configured to input the m pieces of spliced channel information into a neural network model for processing to obtain second channel information;

   Wherein, the neural network model is one of an encoding model and a decoding model.
10. The device according to claim 9, wherein the information splicing module comprises: a granularity information acquisition sub-module and an information splicing sub-module;

    The granularity information acquisition sub-module is used to acquire granularity information, the granularity information is used to indicate the granularity s, and the s is a positive integer;

    The information splicing sub-module is configured to perform m times of splicing with different scales on the n pieces of first channel information based on the granularity information, to obtain the m pieces of splicing channel information, where any two pieces of splicing channel information are spliced The difference between the numbers of the first channel information is an integer multiple of the s.
11. The device of claim 10, wherein:

    The m pieces of the spliced channel information respectively include the nth first channel information corresponding to the nth feedback period.
12. The apparatus of claim 11, wherein:

    The information splicing submodule is used for splicing the (a-1)*s+1th first channel information to the nth first channel information in the n first channel information, to obtain the ath For concatenating channel information, the a is a positive integer that increases by one from 1, and the (a-1)*s+1 is less than the n.
13. The device according to any one of claims 9 to 12, characterized in that:

    The information acquisition module is configured to acquire n pieces of first channel information corresponding to n feedback cycles, and obtain the n pieces of first channel information represented by a sequence;

    or,

    The information acquisition module is configured to acquire n pieces of first channel information corresponding to n feedback cycles, and process the n pieces of first channel information through the first neural network layer to obtain the n pieces of information represented by the feature map first channel information.
14. The device according to any one of claims 9 to 13, wherein the information processing module is configured to:

    The m pieces of splicing channel information are respectively input into m second neural network layers for processing to obtain m channel features;

    Weighted splicing is performed on the m channel features to obtain spliced channel features, and the spliced channel features are input into the third neural network layer to obtain the second channel information;

    or,

    A target channel feature is selected from the m channel features, and the target channel feature is input into the fourth neural network layer to obtain the second channel information.
15. The device according to any one of claims 9 to 14, characterized in that:

    In response to the neural network model being an encoding model, the first channel information includes channel information obtained through channel estimation, and the second channel information includes a compressed bit stream corresponding to the channel information.
16. The device according to any one of claims 9 to 14, characterized in that:

    In response to the neural network model being a decoding model, the first channel information includes a compressed bit stream corresponding to the channel information, and the second channel information includes the restored channel information.
17. A terminal device, characterized in that the terminal device comprises: a processor; wherein,

    the processor, configured to acquire n pieces of first channel information corresponding to n feedback periods, where the feedback period is a feedback period of the channel information, and the n is a positive integer greater than 1;

    The processor is configured to perform m times of splicing of the n pieces of first channel information with different scales to obtain m pieces of spliced channel information, where the splicing of different scales is used to indicate the pieces of spliced channel information in the m pieces of splicing channel information The number of the first channel information is different from each other, and the m is a positive integer;

    the processor, configured to input the m pieces of spliced channel information into a neural network model for processing to obtain second channel information;

    Wherein, the neural network model is an encoding model.
18. A network device, characterized in that the network device comprises: a processor and a transceiver connected to the processor; wherein,

    the transceiver, configured to acquire n pieces of first channel information corresponding to n feedback periods, the feedback periods being the feedback periods of the channel information, and the n being a positive integer greater than 1;

    The processor is configured to perform m times of splicing of the n pieces of first channel information with different scales to obtain m pieces of spliced channel information, where the splicing of different scales is used to indicate the pieces of spliced channel information in the m pieces of splicing channel information The number of the first channel information is different from each other, and the m is a positive integer;

    the processor, configured to input the m pieces of spliced channel information into a neural network model for processing to obtain second channel information;

    Wherein, the neural network model is a decoding model.
19. A computer-readable storage medium, wherein executable instructions are stored in the readable storage medium, and the executable instructions are loaded and executed by a processor to realize the information according to any one of claims 1 to 8 Approach.
20. A chip, characterized in that the chip includes a programmable logic circuit and/or program instructions, which are used to implement the information processing method according to any one of claims 1 to 8 when the chip runs.
21. A computer program product or computer program, characterized in that the computer program product or computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium, and a processor reads from the computer-readable storage medium And execute the computer instructions to implement the information processing method according to any one of claims 1 to 8.

Images