WO2022217506A1

WO2022217506A1 - Channel information feedback method, sending end device, and receiving end device

Info

Publication number: WO2022217506A1
Application number: PCT/CN2021/087288
Authority: WO
Inventors: 肖寒; 田文强; 刘文东
Original assignee: Oppo广东移动通信有限公司
Priority date: 2021-04-14
Filing date: 2021-04-14
Publication date: 2022-10-20
Also published as: CN116569527A

Abstract

A channel information feedback method, a sending end device, and a receiving end device, which are advantageous for feedback precision of channel information and CSI feedback overhead. The method comprises: the sending end device receiving a reference signal sent by the receiving end device; performing channel estimation according to the reference signal to obtain channel information between the sending end device and the receiving end device; performing feature decomposition on the channel information to obtain at least one first feature vector; encoding the at least one first feature vector by means of a neural network to obtain a target bit stream; and sending the target bit stream to the receiving end device.

Description

Channel information feedback method, transmitter device and receiver device

technical field

The embodiments of the present application relate to the field of communications, and in particular, to a method for feeding back channel information, a transmitter device, and a receiver device.

Background technique

In the New Radio (NR) system, in the channel state information reference signal (Channel State Information Reference Signal, CSI-RS) 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 channel estimation structure is selected from the preset precoding codebook according to the result of the channel estimation according to the specific optimization criterion, and the precoding matrix is selected through the feedback link of the air interface. The index information of the matrix is fed back to the receiving end for the receiving end to implement precoding. The mapping process of the channel information to the channel information in the precoding codebook is quantization lossy, which reduces the accuracy of the feedback channel information, and further reduces the precoding performance. If the full channel information of the channel estimation is fed back, the CSI feedback overhead is larger. Therefore, how to balance the feedback accuracy of the channel information and the CSI feedback overhead is an urgent problem to be solved.

SUMMARY OF THE INVENTION

The present application provides a channel information feedback method, a transmitting end device and a receiving end device, which can take into account the feedback accuracy of the channel information and the CSI feedback overhead.

In a first aspect, a method for channel information feedback is provided, comprising: a transmitting end device receiving a reference signal sent by a receiving end device; performing channel estimation according to the reference signal to obtain a relationship between the transmitting end device and the receiving end device. the channel information between the two; perform feature decomposition on the channel information to obtain at least one first feature vector; encode the at least one first feature vector through a neural network to obtain a target bit stream; describe the target bitstream.

In a second aspect, a method for channel information feedback is provided, comprising: a receiving end device receiving a target bit stream sent by a transmitting end device, where the target bit stream is obtained by encoding at least one first feature vector by the transmitting end device , the at least one first feature vector is obtained by performing feature decomposition on the channel estimation result by the transmitting end device; the target bit stream is decoded through a neural network to obtain at least one target feature vector.

In a third aspect, a transmitting end device is provided, which is configured to execute the method in the above-mentioned first aspect or each of its implementations.

Specifically, the sending end device includes a functional module for executing the method in the above-mentioned first aspect or each implementation manner thereof.

In a fourth aspect, a receiving end device is provided, which is configured to execute the method in the second aspect or each of its implementations.

Specifically, the receiving end device includes a functional module for executing the method in the second aspect or each implementation manner thereof.

In a fifth aspect, a transmitter device is provided, including a processor and a memory. The memory is used for storing a computer program, and the processor is used for calling and running the computer program stored in the memory to execute the method in the above-mentioned first aspect or each implementation manner thereof.

In a sixth aspect, a receiver device is provided, including a processor and a memory. The memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to execute the method in the second aspect or each of its implementations.

In a seventh aspect, a chip is provided for implementing any one of the above-mentioned first aspect to the second aspect or the method in each implementation manner thereof.

Specifically, the chip includes: a processor for invoking and running a computer program from a memory, so that a device in which the device is installed executes any one of the above-mentioned first to second aspects or each of its implementations method.

In an eighth aspect, a computer-readable storage medium is provided for storing a computer program, the computer program causing a computer to execute the method in any one of the above-mentioned first aspect to the second aspect or each of its implementations.

In a ninth aspect, a computer program product is provided, comprising computer program instructions, the computer program instructions causing a computer to execute the method in any one of the above-mentioned first to second aspects or the implementations thereof.

In a tenth aspect, there is provided a computer program which, when run on a computer, causes the computer to perform the method in any one of the above-mentioned first to second aspects or the respective implementations thereof.

Through the above technical solution, the transmitting end device obtains at least one eigenvector by decomposing the full channel information of the channel estimation, and further encodes the eigenvector by using a neural network to obtain a target bit stream, and sends the target bit stream to the receiving end, Correspondingly, the receiving end decodes the target bit stream to obtain the target feature vector. On the one hand, when channel information is sent, it only needs to send the target bit stream encoded with the feature vector obtained by decomposing the full channel information, which is beneficial to reduce CSI overhead. On the other hand, encoding the eigenvectors of the full-channel information instead of directly encoding the full-channel information, considering the correlation characteristics between the eigenvectors, is beneficial to avoid compressing too much redundant information and reduce the compression efficiency. Improve encoding performance.

Description of drawings

FIG. 1 is a schematic diagram of a communication system architecture provided by an embodiment of the present application.

FIG. 2 is a schematic interaction diagram of a method for feeding back channel information according to an embodiment of the present application.

FIG. 3 is a system architecture diagram of a method for feeding back channel information according to an embodiment of the present application.

Figure 4 is a schematic diagram of the structure of a neural network.

FIG. 5 is a schematic structural diagram of an encoder according to an embodiment of the present application.

FIG. 6 is a schematic structural diagram of a decoder according to an embodiment of the present application.

FIG. 7 is a schematic structural diagram of the common module in FIG. 6 .

FIG. 8 is a schematic structural diagram of a recurrent neural network.

FIG. 9 is a schematic structural diagram of an encoder according to another embodiment of the present application.

FIG. 10 is a schematic structural diagram of a decoder according to another embodiment of the present application.

FIG. 11 is a schematic structural diagram of an encoder according to yet another embodiment of the present application.

FIG. 12 is an exemplary structural diagram of the self-attention module in FIG. 11 .

FIG. 13 is a schematic structural diagram of a decoder according to yet another embodiment of the present application.

FIG. 14 is a schematic structural diagram of the mask block in FIG. 13 .

FIG. 15 is a schematic structural diagram of the residual block in FIG. 13 .

FIG. 16 is a schematic block diagram of a transmitting end device according to an embodiment of the present application.

FIG. 17 is a schematic block diagram of a receiving end device according to an embodiment of the present application.

FIG. 18 is a schematic block diagram of a communication device provided according to an embodiment of the present application.

FIG. 19 is a schematic block diagram of a chip provided according to an embodiment of the present application.

FIG. 20 is a schematic block diagram of a communication system provided according to an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. With regard to the embodiments in the present application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

The technical solutions of the embodiments of the present application can be applied to various communication systems, for example: a Global System of Mobile communication (GSM) system, a Code Division Multiple Access (CDMA) system, a wideband Code Division Multiple Access (CDMA) system (Wideband Code Division Multiple Access, WCDMA) system, General Packet Radio Service (General Packet Radio Service, GPRS), Long Term Evolution (Long Term Evolution, LTE) system, Advanced Long Term Evolution (Advanced long term evolution, LTE-A) system , New Radio (NR) system, evolution system of NR system, LTE (LTE-based access to unlicensed spectrum, LTE-U) system on unlicensed spectrum, NR (NR-based access to unlicensed spectrum) unlicensed spectrum, NR-U) system, Non-Terrestrial Networks (NTN) system, Universal Mobile Telecommunication System (UMTS), Wireless Local Area Networks (WLAN), Wireless Fidelity (Wireless Fidelity, WiFi), fifth-generation communication (5th-Generation, 5G) system or other communication systems, etc.

Generally speaking, traditional communication systems support a limited number of connections and are easy to implement. However, with the development of communication technology, mobile communication systems will not only support traditional communication, but also support, for example, Device to Device (Device to Device, D2D) communication, Machine to Machine (M2M) communication, Machine Type Communication (MTC), Vehicle to Vehicle (V2V) communication, or Vehicle to everything (V2X) communication, etc. , the embodiments of the present application can also be applied to these communication systems.

Optionally, the communication system in this embodiment of the present application may be applied to a carrier aggregation (Carrier Aggregation, CA) scenario, a dual connectivity (Dual Connectivity, DC) scenario, or a standalone (Standalone, SA) distribution. web scene.

Optionally, the communication system in the embodiment of the present application may be applied to an unlicensed spectrum, where the unlicensed spectrum may also be considered as a shared spectrum; or, the communication system in the embodiment of the present application may also be applied to a licensed spectrum, where, Licensed spectrum can also be considered unshared spectrum.

The embodiments of the present application describe various embodiments in conjunction with network equipment and terminal equipment, where the terminal equipment may also be referred to as user equipment (User Equipment, UE), access terminal, subscriber unit, subscriber station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent or user device, etc.

The terminal device may be a station (STATION, ST) in the WLAN, and may be a cellular phone, a cordless phone, a Session Initiation Protocol (Session Initiation Protocol, SIP) phone, a Wireless Local Loop (WLL) station, a personal digital assistant (Personal Digital Assistant, PDA) devices, handheld devices with wireless communication capabilities, computing devices or other processing devices connected to wireless modems, in-vehicle devices, wearable devices, next-generation communication systems such as end devices in NR networks, or future Terminal equipment in the evolved public land mobile network (Public Land Mobile Network, PLMN) network, etc.

In this embodiment of the present application, the terminal device can be deployed on land, including indoor or outdoor, handheld, wearable, or vehicle-mounted; it can also be deployed on water (such as ships, etc.); it can also be deployed in the air (such as airplanes, balloons, and satellites) superior).

In this embodiment of the present application, the terminal device may be a mobile phone (Mobile Phone), a tablet computer (Pad), a computer with a wireless transceiver function, a virtual reality (Virtual Reality, VR) terminal device, and an augmented reality (Augmented Reality, AR) terminal Equipment, wireless terminal equipment in industrial control, wireless terminal equipment in self driving, wireless terminal equipment in remote medical, wireless terminal equipment in smart grid , wireless terminal equipment in transportation safety, wireless terminal equipment in smart city or wireless terminal equipment in smart home, etc.

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

In this embodiment of the present application, the network device may be a device for communicating with a mobile device, and the network device may be an access point (Access Point, AP) in WLAN, or a base station (Base Transceiver Station, BTS) in GSM or CDMA , it can also be a base station (NodeB, NB) in WCDMA, it can also be an evolved base station (Evolutional Node B, eNB or eNodeB) in LTE, or a relay station or access point, or in-vehicle equipment, wearable devices and NR networks The network equipment (gNB) in the PLMN network in the future evolution or the network equipment in the NTN network, etc.

As an example and not a limitation, in this embodiment of the present application, the network device may have a mobile feature, for example, the network device may be a mobile device. Optionally, the network device may be a satellite or a balloon station. For example, the satellite may be a low earth orbit (LEO) satellite, a medium earth orbit (MEO) satellite, a geostationary earth orbit (GEO) satellite, a High Elliptical Orbit (HEO) ) satellite etc. Optionally, the network device may also be a base station set in a location such as land or water.

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

Exemplarily, a communication system 100 to which this embodiment of the present application is applied is shown in FIG. 1 . The communication system 100 may include a network device 110, and the network device 110 may be a device that communicates with a terminal device 120 (or referred to as a communication terminal, a terminal). The network device 110 may provide communication coverage for a particular geographic area, and may communicate with terminal devices located within the coverage area.

FIG. 1 exemplarily shows one network device and two terminal devices. Optionally, the communication system 100 may include multiple network devices and the coverage of each network device may include other numbers of terminal devices. This application The embodiment does not limit this.

Optionally, the communication system 100 may further include other network entities such as a network controller and a mobility management entity, which are not limited in this embodiment of the present application.

It should be understood that, in the embodiments of the present application, a device having a communication function in the network/system may be referred to as a communication device. Taking the communication system 100 shown in FIG. 1 as an example, the communication device may include a network device 110 and a terminal device 120 with a communication function, and the network device 110 and the terminal device 120 may be the specific devices described above, which will not be repeated here. ; The communication device may also include other devices in the communication system 100, such as other network entities such as a network controller, a mobility management entity, etc., which are not limited in this embodiment of the present application.

It should be understood that the terms "system" and "network" are often used interchangeably herein. The term "and/or" in this article is only an association relationship to describe the associated objects, indicating that there can be three kinds of relationships, for example, A and/or B, it can mean that A exists alone, A and B exist at the same time, and A and B exist independently B these three cases. In addition, the character "/" in this document generally indicates that the related objects are an "or" relationship.

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.

In this embodiment of the present application, "predefinition" may be implemented by pre-saving corresponding codes, forms, or other means that can be used to indicate relevant information in devices (for example, including terminal devices and network devices). The implementation method is not limited. For example, predefined may refer to the definition in the protocol.

In the embodiments of the present application, the "protocol" may refer to a standard protocol in the communication field, for example, may include the LTE protocol, the NR protocol, and related protocols applied in future communication systems, which are not limited in this application.

In order to facilitate the understanding of the technical solutions of the embodiments of the present application, the technical solutions of the present application are described in detail below through specific embodiments. The following related technologies can be arbitrarily combined with the technical solutions of the embodiments of the present application as optional solutions, which all belong to the protection scope of the embodiments of the present application. The embodiments of the present application include at least part of the following contents.

In the NR system, the feedback of the channel information adopts a codebook-based feedback scheme. The feedback scheme is to select the optimal channel information eigenvalue vector from the precoding codebook according to the result of the channel estimation. Since the precoding codebook itself is limited, the channel estimation results in the precoding codebook are based on the channel estimation. The mapping process is quantized and lossy, which reduces the accuracy of the channel information fed back, which in turn reduces the performance of precoding. If the full channel information is fed back, the CSI feedback overhead will be increased. Therefore, how to take account of the channel information Feedback accuracy and CSI feedback overhead is an urgent problem to be solved.

FIG. 2 is a schematic interaction diagram of a method 200 for channel information feedback according to an embodiment of the present application. As shown in FIG. 2 , the method 200 includes at least some of the following contents:

S201, the transmitting end device receives the reference signal sent by the receiving end device;

S202, the transmitting end device performs channel estimation according to the reference signal, and obtains channel information between the transmitting end device and the receiving end device;

S203, the transmitting end device performs feature decomposition on the channel information to obtain at least one first feature vector;

S204, the sending end device encodes the at least one first feature vector through a neural network to obtain a target bit stream;

S205, the sending end device sends the target bit stream to the receiving end device.

S206, the receiving end device decodes the target bit stream to obtain at least one target feature vector.

In some embodiments, the sending end device is a terminal device, and the receiving end device is a network device.

In other embodiments, the sending end device is a network device, and the receiving end device is a terminal device.

In still other embodiments, the transmitting end device is a terminal device, and the receiving end device is another terminal device.

In still other embodiments, the sending end device is a network device, and the receiving end device is another network device.

It should be understood that the reference signal is also different according to the difference between the transmitting end device and the receiving end device. For example, if the transmitting end device is a terminal device and the receiving end device is a network device, the reference signal can be a demodulation reference signal. Signal (Demodulation Reference Signal, DMRS).

In the embodiment of the present application, an encoder is deployed in the transmitting end device, and a decoder is deployed in the receiving end device. In the embodiment of the present application, the encoder in the transmitting end device and the decoding device in the receiving end device are The device can be implemented by a neural network.

It should be noted that the embodiments of the present application do not limit the specific manner in which the transmitting end device performs feature decomposition on the channel information. As an example, the sending end device may perform singular value decomposition (Singular Value Decomposition, SVD) on the channel information to obtain the at least one first feature vector.

In some embodiments of the present application, the channel information is full channel information obtained by performing channel estimation on a reference signal, that is, unquantized channel information.

Therefore, in the embodiment of the present application, the transmitting end device obtains at least one feature vector by decomposing the full channel information obtained by channel estimation, and further encodes the feature vector by using a neural network to obtain a target bit stream, which is sent to the receiving end. the target bitstream. Correspondingly, the receiving end device decodes the target bit stream to obtain a target feature vector, and further performs precoding according to the target feature vector.

Based on the channel information feedback scheme of the embodiments of the present application, on the one hand, when transmitting channel information, the transmitting end device only needs to transmit the target bit stream obtained by encoding the eigenvector obtained by eigendecomposition of the full channel information, which is beneficial to reduce CSI overhead On the other hand, encoding the feature vector of the full channel information instead of directly compressing and encoding the full channel information can take into account the correlation characteristics between the channel information, which is beneficial to avoid compressing too much redundant information and improve the compression efficiency. , thereby improving the encoding performance.

In some embodiments of the present application, the receiving end device may perform precoding according to at least one target feature vector obtained by decoding, for example, may perform beamforming according to the at least one target feature vector.

In some scenarios, the scheduling bandwidth of the transmitting end device may include multiple subcarrier groups, respectively corresponding to multiple subbands. In some embodiments of the application, the transmitting end device may feed back the channel information on the multiple subbands of the transmitting end device respectively. .

In some embodiments, the transmitting end device may perform channel estimation according to the reference signal to obtain channel information on multiple subbands, and further perform feature decomposition on the channel information on the multiple subbands to obtain eigenvectors corresponding to the multiple subbands respectively.

That is to say, the full channel information of the channel estimation includes channel information corresponding to each subcarrier group in the multiple subcarrier groups of the transmitting end device. In other words, the full channel information may include channel information corresponding to each of the multiple subbands of the transmitting end device.

Correspondingly, the at least one first eigenvector includes an eigenvector corresponding to each subcarrier group in the multiple subcarrier groups respectively. The at least one first feature vector includes a feature vector corresponding to each of the plurality of subbands, respectively.

Since the eigenvectors of the channel information on multiple subbands are correlated, for example, the channel information on adjacent subbands has more correlation information, by jointly compressing and encoding the eigenvectors of the multiple subbands, compared to considering This correlation information is beneficial to improve compression efficiency and enhance feedback recovery performance.

FIG. 3 is a schematic system architecture diagram of a method for channel feedback according to an embodiment of the present application.

The transmitting end device can perform feature decomposition on the channel information of the n sub-bands to obtain the feature vectors corresponding to the n sub-bands, denoted as W_1, W_2, ..., W_n, and further input the feature vectors corresponding to the n sub-bands to the encoding In the encoder, the encoder uses the neural network to jointly encode the feature vectors of the n sub-bands to obtain the target bit stream, denoted as B. Further, the transmitting end device sends this target bit stream B to the receiving end device, and the decoder of the receiving end device decodes and recovers this target bit stream B, obtains n target feature vectors, denoted as W`_1, W`_2 , …, W`_n.

Therefore, in this embodiment of the present application, the transmitting end device performs joint compression and feedback on the eigenvectors of the channel information of multiple subbands on the scheduling bandwidth, and correspondingly, the receiving end device can decompress and recycle the eigenvectors of the multiple subbands. Compared with the use of the correlation information between feature vectors, the structure is beneficial to improve the compression efficiency, thereby enhancing the performance of compression, feedback and decompression of the entire system.

Optionally, in some embodiments, the scheduling bandwidth may be at least one BWP, or at least one carrier, or at least one frequency band, etc., which is not limited in this application.

It should be understood that the embodiments of the present application do not specifically limit the specific implementation manners of the encoder in the transmitting end device and the decoder in the receiving end device. Hereinafter, with reference to specific embodiments, implementations of the encoder in the transmitting end device and the decoder in the receiving end device will be described.

In some embodiments of the present application, the encoder in the sending end and the decoder in the receiving end may be implemented using the neural network structure in FIG. 4 . Wherein, the neural network includes an input layer, a hidden layer and an output layer, the input layer is used for receiving data, the hidden layer is used for processing the received data, and the processing result is generated in the output layer. In this neural network, each node represents a processing module, which can be considered to simulate a neuron, and multiple neurons form a layer of neural network, and multiple layers of information transmission and processing construct an overall neural network.

In some embodiments of the present application, the encoder of the transmitting end device may use computer vision technology to process the input at least one first feature vector, for example, use a neural network to process the at least one first feature vector as an image to be compressed Compression coding is performed to obtain the target bit stream.

Correspondingly, the decoder of the receiving end device can decode and restore the target bitstream by using the target bitstream as the information obtained by compressing the image to obtain the target image, wherein the target image includes at least one target feature vector.

In some embodiments, the encoder of the sender device may use a neural network for image processing to compress and encode the at least one first feature vector, for example, the neural network for image processing may be a convolutional neural network , or other neural networks with better image processing performance, which are not limited in this application.

In some embodiments, the encoder of the receiving end device may use a neural network for image processing to decompress the target bit stream, for example, the neural network for image processing may be a convolutional neural network, or other The neural network with better image processing performance is not limited in this application.

Optionally, the convolutional neural network includes an input layer, at least one convolutional layer, at least one pooling layer, a fully connected layer and an output layer. By introducing a convolutional layer and a pooling layer, relative to the neural network architecture in Figure 4 The rapid increase of network parameters is effectively controlled, the number of parameters is limited and the characteristics of local structures are excavated, and the robustness of the algorithm is improved.

In some embodiments, the encoding of the at least one first feature vector through a neural network to obtain a target bit stream includes:

The at least one first eigenvector is spliced into a matrix of eigenvectors and input to the neural network;

The neural network encodes the feature vector matrix as an image to be compressed to obtain the target bit stream.

For example, the feature vectors W_1, ^W_2 , . The neural network is input, and the feature vector matrix W is used as the image to be compressed to be compressed and encoded through the neural network, and the target bit stream B is obtained.

Correspondingly, at the decoding end, the target bit stream is decoded by the neural network to obtain at least one target feature vector, including:

inputting the target bit stream to the neural network;

Decoding the target bitstream by using the neural network as information obtained by encoding an image to obtain a target image, where the target image includes the at least one target feature vector.

That is, the decoder at the receiving end takes the received target bitstream B as information obtained by compressing the image, and decompresses the target bitstream B to obtain a target image including at least one target feature vector.

In some embodiments, the encoder in the transmitting end device and the decoder in the receiving end device may be implemented using the network structures shown in FIG. 5 and FIG. 6 , respectively.

In some embodiments, the encoder in the transmitting end device may include: a feature extraction module, configured to receive an input eigenvector matrix W, perform feature extraction on the eigenvector matrix W, and obtain the corresponding eigenvector matrix W feature map.

In some embodiments, the feature extraction module performs feature extraction on the feature vector matrix W through convolution kernels of different sizes to obtain feature maps of different fields of view of the feature vector matrix W, which increases the nonlinearity in the convolution process, Improves the expressiveness of convolutional neural networks.

As an example, as shown in Figure 5, the feature extraction module may include 3×3 convolutional layers, 5×5 convolutional layers and 7×7 convolutional layers. Among them, the 3x3 convolutional layer uses a 3x3 convolution kernel, the 5x5 convolutional layer uses a 5x5 convolution kernel, and the 7x7 convolutional layer uses a 7x7 convolution kernel. In practical applications, convolution kernels of other sizes can also be replaced, which is not limited in this application.

Further, the plurality of feature maps output by the feature extraction module are input to the splicing module to merge the feature maps, for example, the splicing module realizes the splicing of the channel dimensions of the plurality of feature maps.

As an example, as shown in Figure 5, the splicing module can be implemented by a 1×1 convolutional layer, where the 1×1 convolutional layer uses a 1×1 convolution kernel, and in the 1×1 convolutional layer, the convolutional convolution can be controlled by The number of kernels controls the number of channels. The convolution process using a 1×1 convolution kernel is larger than the calculation process of the fully connected layer, and a nonlinear activation function is added, which is beneficial to increase the other nonlinearity of the neural network and make the neural network express the characteristics. More complex.

Further, as shown in FIG. 5 , the feature map output from the splicing module is processed by the fully connected layer and the quantization layer and converted into the target bitstream B.

It should be understood that the convolutional neural network structure in Figure 5 is only an example. In practical applications, it can be flexibly designed according to the number of subbands, encoding and decoding performance requirements, etc., for example, adding an activation layer between the convolutional layers, normalizing The application is not limited to network layers such as the ization layer.

In some embodiments, as shown in FIG. 6 , the decoder may include: a fully connected layer, a dimension adjustment module and a residual module, wherein the target bitstream B received from the sender device is first input to the fully connected layer and the dimension The adjustment module is converted into the dimension of the feature vector matrix W, which is further input to the residual module, and outputs the feature vector matrix W` composed of the at least one target feature vector.

In some embodiments, as shown in FIG. 6 , the residual module may include a convolutional layer, an L-order common module, a convolutional layer and a summation module, where L is a positive integer. First, the output of the dimension adjustment module is sampled and copied, one input is input to the summation module, and the other is input to the convolutional layer, the convolution kernel of the convolutional layer is used to enlarge the number of channels, and the feature information is further extracted through L times of public modules. . The output of the L public modules is further reduced in number of channels through the convolutional layer, and the output of the convolutional layer is input to the summation module and the output of the dimension adjustment module for sampling and replication is summed to obtain the feature vector matrix W'.

It should be understood that the embodiments of the present application do not specifically limit the number and structural composition of the common modules. As an example, the common modules may adopt the structure shown in FIG. 7 , but the present application is not limited thereto.

It should be understood that the convolutional neural network structure in FIG. 6 is only an example. In practical applications, it can be flexibly configured according to information such as the number of subbands, encoding and decoding performance requirements, and the present application is not limited thereto.

In some embodiments of the present application, the model parameters of the neural network of the encoder and the decoder are obtained by joint training. For example, firstly, the model parameters of the neural network of the encoder and the decoder are initialized, and multiple sets of feature vector matrices are input to the neural network of the encoder. The samples are encoded to obtain multiple target bitstreams, and the multiple target bitstreams are further input to the decoding end for decoding, and the model parameters of the encoder and decoder are adjusted according to the decoding results until the feature vector output by the decoder and the input to the encoding. The eigenvectors of the neural network of the generator satisfy the convergence conditions.

Therefore, in the embodiment of the present application, the transmitting end device uses the convolutional neural network to compress and encode the feature vectors corresponding to the channel information of the multiple subbands as images to obtain the target bit stream. Correspondingly, the receiving end device decompresses and restores the target bit stream to the target image through the convolutional neural network, thereby obtaining at least one target feature vector. On the one hand, compared with directly encoding the full channel information of the channel estimation, encoding the feature vector of the full channel information is beneficial to avoid compressing excessive redundant information and reduce the CSI feedback overhead. On the other hand, joint compression feedback is performed according to the cross-correlation information between the feature vectors of multiple subbands in the frequency domain, which is beneficial to improve the compression feedback performance.

In other embodiments of the present application, the encoder of the sender device may use a cyclic neural network (RNN) to process the input at least one first feature vector, for example, use the neural network to use the at least one first feature vector as a The elements of the sequence are compressed to obtain an encoding result, that is, the target bit stream.

Correspondingly, the encoder of the receiving end device can use the cyclic neural network to decompress and recover the input target bit stream as the information obtained by compressing and encoding the sequence to obtain a target sequence consisting of at least one target bit vector. .

As shown above, the neural network includes an input layer, a hidden layer, and an output layer. The output is controlled by an activation function, and the layers are connected by weights. The activation function is predetermined, and what the neural network model learns through training is contained in the weights. The basic neural network only establishes weight connections between layers. The biggest difference between RNN and basic neural networks is that weight connections are also established between neurons between layers.

Figure 8 is a typical RNN structure diagram. Each arrow in Figure 8 represents a transformation, that is to say, the arrow connections have weights. The left picture in Figure 8 is the RNN structure folded diagram, the right side is the RNN structure expansion diagram, the arrow next to h in the left picture represents the "cycle" in this structure is reflected in the hidden layer.

It can be seen from the expanded RRN structure diagram that the neurons in the hidden layer also have weights. That is, as the sequence progresses, earlier hidden layers will affect later hidden layers. In Figure 8, x represents the input, h represents the hidden unit, O represents the output, y represents the label of the training set, L represents the loss function, K, V and U represent the weights, t represents time t, and t-1 represents time t-1. , t+1 represents time t+1, from which it can be seen that the "loss" also accumulates with the recommendation of the sequence. Based on the above structure, RNN can have good performance in processing sequence data, that is, RNN is a recursive neural network that performs recursion in the evolution direction of the sequence and all nodes (recurrent units) are connected in a chain.

Long Short-Term Memory (Long Shor Term Memory, LSTM) is an evolved RNN. Different from the typical RNN structure, LSTM introduces the concept of cell state. Unlike RNN, which only considers the most recent state, the cell state of LSTM will determine Which states should be left behind and which states should be forgotten solves the shortcomings of traditional RNNs in long-term memory.

In some embodiments of the present application, the at least one first feature vector may be processed through a basic RNN to obtain the target bit stream B, or the at least one first feature vector may also be processed through an LSTM to obtain the target bit stream B stream B.

inputting each feature vector in the at least one first feature vector into the recurrent neural network in turn;

The target bit stream is obtained by encoding each feature vector as an element of a sequence through the recurrent neural network.

For example, each feature vector in the at least one first feature vector is sequentially input to the LSTM, and the LSTM encodes each feature vector as an element of a sequence to obtain the target bitstream.

As an example, the feature vectors W_1, W_2, .

inputting the target bit stream to the recurrent neural network;

The target bit stream is decoded by using the cyclic neural network as information obtained by encoding the sequence to obtain a target sequence, where the target sequence includes the at least one target feature vector.

That is, the decoder at the receiving end takes the received target bit stream B as information obtained by compressing the sequence, and decompresses the target bit stream B to obtain a target sequence including at least one target feature vector.

Hereinafter, the specific encoding and decoding processes will be described by taking the encoder in the transmitting end device and the decoder in the receiving end device adopting the network structures in FIG. 9 and FIG. 10 respectively as an example. It should be understood that the neural network structures illustrated in FIG. 9 and FIG. 10 are only examples, and in practical applications, they can be flexibly configured according to information such as the number of subbands, encoding and decoding performance requirements, and the present application is not limited thereto.

In some embodiments, as shown in FIG. 9 , the encoder may include: an LSTM module for sequentially receiving each feature vector in the at least one first feature vector, and using each feature vector as an element of the sequence to be processed.

Further, the encoder may include a fully connected layer and a quantization layer for converting the result processed by the LSTM module to obtain the target bitstream B.

In some embodiments, as shown in FIG. 10 , the decoder may include a fully connected layer, a plurality of LSTM modules, and a fully connected layer connected to each LSTM module of the plurality of LSTM modules, wherein the first A fully connected layer is used to perform dimension transformation on the target bitstream B. The output of the first fully connected layer is used as the input of the first LSTM module, and the output of each LSTM module is used as the input of the next LSTM module. After the outputs of the multiple LSTM modules pass through the corresponding fully connected layers, the corresponding feature vector matrix W`={W`_1, W`_2, . . . , W`_n} is output.

Therefore, in the embodiment of the present application, the transmitting end device uses the cyclic neural network to compress the feature vectors of multiple subbands as elements of the sequence to obtain the target bit stream. The receiving end device decompresses and restores the target bit stream to elements of the sequence through the cyclic neural network, thereby obtaining at least one target feature vector. On the one hand, compared with directly encoding the full channel information of the channel estimation, encoding the feature vector of the full channel information is beneficial to avoid compressing excessive redundant information and reduce the CSI feedback overhead. On the other hand, joint compression feedback is performed according to the cross-correlation information between the feature vectors of multiple subbands in the frequency domain, which is beneficial to improve the compression feedback performance.

In still other embodiments of the present application, the encoding of the at least one first feature vector through a neural network to obtain a target bit stream includes:

The at least one first feature vector is encoded based on an attention mechanism to obtain the target bitstream.

Optionally, in this embodiment of the present application, the attention mechanism may be a self-attention mechanism, or may also be another attention mechanism, which is not limited in this application.

The self-attention mechanism adopts a "query-key-value" model.

The calculation of attention is mainly divided into three steps:

Step 1: Calculate the similarity between the query and each key to obtain the weight. Commonly used similarity functions are dot product, splicing, perceptron, etc.;

Step 2: Use a softmax function to normalize these weights;

Finally, the weight and the corresponding key value are weighted and summed to obtain the final attention.

In some embodiments of the present application, the transmitting end device extracts correlation features between feature vectors of multiple subbands and correlation features between elements of feature vectors based on an attention mechanism, to further improve encoding performance and improve decompression performance. For example, the sender device extracts the correlation feature of the elements in the at least one first feature vector based on the attention mechanism, obtains at least one second feature vector, and further compresses the at least one second feature vector to obtain the Describe the target bitstream B.

Correspondingly, at the decoding end, the device at the receiving end may decode the target bitstream based on the attention mechanism to obtain at least one target feature vector.

In some embodiments of the present application, the decoder first performs feature extraction on the target bitstream to obtain a first feature map of the target bitstream; further determines, based on an attention mechanism, the features in the first feature map of the target bitstream. The weight of the element; the first feature map of the target bitstream and the weights of the elements in the first feature map of the target bitstream are dot-multiplied to obtain the second feature map of the target bitstream; The second feature map of the target bitstream is decompressed to obtain the at least one target feature vector.

Hereinafter, the specific encoding and decoding processes will be described by taking the network structures shown in FIG. 11 and FIG. 13 as an example for the encoder in the transmitting end device and the decoder in the receiving end device respectively. It should be understood that the neural network structures illustrated in FIG. 11 and FIG. 13 are only examples, and in practical applications, they can be flexibly configured according to information such as the number of subbands, codec performance requirements, and the present application is not limited thereto.

In some embodiments, as shown in FIG. 11 , the encoder may include: s-time self-attention modules, concatenation modules, fully connected layers and quantization layers, where s is a positive integer, and each self-attention module is used for The self-attention mechanism is used to extract the correlation features of the elements in the input multiple first feature vectors, and after the s cascaded sub-attention modules, multiple second feature vectors are output. Further, the plurality of second feature vectors are input to the splicing module, spliced into one feature vector, and the target bit stream B is obtained after processing by the fully connected layer and the quantization layer.

In some embodiments, the attention module can be implemented using the structure shown in FIG. 12 , but the present application is not limited thereto.

As an example, as shown in Figure 12, the attention module includes: n fully-connected layers, an attention layer and n groups of two-layer fully-connected layers, wherein the n fully-connected layers are respectively used to receive a feature vector, the characteristic The vector can be the first feature vector in the preceding paragraph, or it can be the feature vector output by the previous attention module; the attention layer is used to receive multiple feature vectors output from n fully connected layers, and extract the feature vectors of the multiple feature vectors. Correlation characteristics between elements. Each output of the self-attention layer is first copied and sampled, and each output is summed through a corresponding set of two-layer fully connected layers and the sampled outputs as the output of the self-attention module.

In some embodiments, as shown in FIG. 13 , the decoder may include: a fully connected layer, a dimension adjustment module, a feature extraction module and a self-attention module, and the target bitstream B is first input to the fully connected layer and the dimension adjustment module, It is converted into the dimension of the feature vector matrix W, and further input to the feature extraction module for feature extraction to obtain a first feature map, and the output of the dimension adjustment module is sampled and copied.

Optionally, as shown in FIG. 13 , the feature extraction module may include a convolution layer, an activation function, a q-order residual block and a t-order residual block, where q and t are positive integers.

The convolutional layer is used to adjust the number of channels of the output of the module. The output of the convolutional layer is input to the q-th residual block through the activation function, and the output of the activation function is sampled and copied.

After the output of the activation function passes through the q residual blocks, it is divided into two paths, one passes through the t residual blocks to obtain the second feature map, and the other is input to the self-attention module to extract the attention of the elements in the feature vector matrix. Weights.

In some embodiments, the self-attention module can be implemented by a mask block, or can also be implemented by the attention module in FIG. 11 , which is not limited in this application.

For example, the output of the q-th residual block can be input to the mask block, which extracts attention masks for elements in the feature map.

Further, the output of the mask block and the output of the t-th residual block are subjected to dot multiplication processing, and the dot multiplication result is added to the sampled copy result of the output of the activation function.

Then, the addition result is input to the q-th residual block for result correction, and then the channel dimension is reduced through the convolution layer, and the result of the sampling copy of the second eigenvector matrix is added to output the restored eigenvector matrix W `=[W`_1, W`_2, ..., W`_n].

FIG. 14 is an exemplary structural diagram of the mask block, but the present application is not limited thereto.

As shown in Fig. 14, the mask block can include m-order residual block, down-sampling module, 2m-order residual block, up-sampling module, m-order residual block, convolution layer and sigmoid activation function, namely the mask The block can be realized by the 4mth residual block, and the upsampling and downsampling processing are performed after the mth residual block and the 3mth residual block, respectively. Feature extraction, and finally match the channel dimension through the convolution layer, and then map the output of the convolution layer between 0 and 1 through the sigmoid activation function to obtain the attention weight.

It should be understood that the residual block in this embodiment of the present application may be implemented by using the structure shown in FIG. 15 , or may also be implemented by using other equivalent structures, which is not limited in this application.

As an example, as shown in Figure 15, the residual block may include: a convolution layer, a normalization layer, an activation function, a convolution layer and a summation module, wherein the input of the residual block is sampled and copied all the way to the calculation The sum module, the other way through the convolution layer, the normalization layer, the activation function and the output of the convolution layer is sent to the summation module, and the input of the residual block and the output of the convolutional layer are summed through the summation module, as The output of the residual block. Designing a suitable residual block in the neural network is beneficial to solve the gradient problem in the neural network.

Therefore, in the embodiment of the present application, the transmitting end device uses the attention mechanism to compress the feature vectors of multiple subbands as elements of the sequence to obtain the target bit stream. Correspondingly, the receiving end device uses the attention mechanism to decompress and restore the target bit stream, thereby obtaining at least one target feature vector. On the one hand, compared with directly encoding the full channel information of the channel estimation, encoding the feature vector of the full channel information is beneficial to avoid compressing excessive redundant information and reduce the CSI feedback overhead. On the other hand, the attention mechanism is used to jointly compress the cross-correlation information between the elements of the eigenvectors of multiple subbands, taking into account the correlation features between the eigenvectors of multiple subbands, as well as the correlation between the elements of the eigenvectors. The correlation feature is beneficial to improve the compression feedback performance.

The method embodiments of the present application are described in detail above with reference to FIGS. 2 to 15 , and the device embodiments of the present application are described in detail below with reference to FIGS. 16 to 20 . It should be understood that the device embodiments and the method embodiments correspond to each other, and are similar to For the description, refer to the method embodiment.

FIG. 16 shows a schematic block diagram of a transmitting end device 400 according to an embodiment of the present application. As shown in FIG. 16, the sender device 400 includes:

The communication module 410 is used for receiving the reference signal sent by the receiving end device;

The processing module 420 is configured to perform channel estimation according to the reference signal, obtain channel information between the transmitting end device and the receiving end device, and perform feature decomposition on the channel information to obtain at least one first eigenvector ;

an encoding module 430, configured to encode the at least one first feature vector through a neural network to obtain a target bit stream;

The communication module 410 is further configured to: send the target bit stream to the receiving end device.

In some embodiments of the present application, the channel information includes channel information corresponding to each of the multiple subcarrier groups of the transmitting end device, and the at least one first feature vector includes the multiple subcarrier groups Eigenvectors corresponding to each subcarrier group in .

In some embodiments of the present application, the encoding module 430 is specifically configured to:

The target bit stream is obtained by encoding the feature vector matrix as an image to be compressed through the neural network.

In some embodiments of the present application, the neural network is a convolutional neural network.

In some embodiments of the present application, the neural network is a recurrent neural network, and the encoding module 430 is further configured to:

In some embodiments of the present application, the recurrent neural network includes a long short-term memory LSTM neural network.

In some embodiments of the present application, the encoding module 430 is further configured to:

In some embodiments of the present application, the encoding module 430 further includes:

an attention module, configured to extract the correlation feature between elements in the at least one first feature vector based on the attention mechanism, to obtain at least one second feature vector;

A feature vector compression module, configured to perform feature compression on the at least one second feature vector to obtain the target bit stream.

In some embodiments of the present application, the transmitting end device is a terminal device, and the receiving end device is a network device.

Optionally, in some embodiments, the above-mentioned communication module may be a communication interface or a transceiver, or an input/output interface of a communication chip or a system-on-chip. The above-mentioned processing module and encoding module may be one or more processors.

It should be understood that the transmitting end device 400 according to the embodiment of the present application may correspond to the transmitting end device in the method embodiment of the present application, and the above-mentioned and other operations and/or functions of each unit in the transmitting end device 400 are for the purpose of realizing FIG. 2 . The corresponding flow of the transmitting end device in the method 200 shown in FIG. 15 will not be repeated here for brevity.

FIG. 17 is a schematic block diagram of a receiving end device according to an embodiment of the present application. The receiving end device 500 of FIG. 17 includes:

The communication module 510 is configured to receive a target bit stream sent by the sending end device, where the target bit stream is obtained by encoding at least one first feature vector by the sending end device, and the at least one first feature vector is the It is obtained by the terminal equipment performing eigendecomposition on the channel estimation result;

The decoding module 520 is configured to decode the target bit stream through a neural network to obtain at least one target feature vector.

In some embodiments of the present application, the decoding module 520 is configured to:

inputting the target bit stream to the neural network;

In some embodiments of the present application, the neural network is a recurrent neural network, and the decoding module 520 is further configured to:

inputting the target bit stream to the recurrent neural network;

In some embodiments of the present application, the decoding module 520 is further configured to:

The target bitstream is decoded based on the attention mechanism to obtain at least one target feature vector.

In some embodiments of the present application, the encoding module 520 includes:

a feature extraction module, configured to perform feature extraction on the target bitstream to obtain a first feature map;

an attention module, configured to determine the weight of the elements in the first feature map based on the attention mechanism;

a dot product module, configured to perform dot product on the first feature map and the weights of the elements in the first feature map to obtain a second feature map;

A decompression module, configured to decompress the second feature map to obtain the at least one target feature vector.

In some embodiments of the present application, the receiving end device 500 further includes:

A processing module, configured to perform precoding according to the at least one target feature vector.

It should be understood that the receiving end device 500 according to the embodiment of the present application may correspond to the receiving end device in the method embodiment of the present application, and the above-mentioned and other operations and/or functions of each unit in the receiving end device 500 are for the purpose of realizing FIG. 2 . The corresponding flow of the receiving end device in the method 200 shown in FIG. 15 is not repeated here for brevity.

In summary, the transmitting end device obtains at least one feature vector by decomposing the full channel information obtained by channel estimation, and further encodes the feature vector by using a neural network to obtain a target bit stream, and sends the target bit stream to the receiving end. Correspondingly, the receiving end device decodes the target bit stream to obtain a target feature vector, and further performs precoding according to the target feature vector. On the one hand, when transmitting the channel information, the transmitting end device only needs to send the target bit stream obtained by encoding the eigenvector obtained by eigendecomposition of the full channel information, which is beneficial to reduce the CSI overhead. Encoding the vector instead of directly compressing and encoding the full channel information can take into account the correlation characteristics between the channel information, which is beneficial to avoid compressing too much redundant information, improve the compression efficiency, and then improve the encoding performance.

FIG. 18 is a schematic structural diagram of a communication device 600 provided by an embodiment of the present application. The communication device 600 shown in FIG. 18 includes a processor 610, and the processor 610 can call and run a computer program from a memory, so as to implement the method in the embodiment of the present application.

Optionally, as shown in FIG. 18 , the communication device 600 may further include a memory 620 . The processor 610 may call and run a computer program from the memory 620 to implement the methods in the embodiments of the present application.

The memory 620 may be a separate device independent of the processor 610 , or may be integrated in the processor 610 .

Optionally, as shown in FIG. 6 , the communication device 600 may further include a transceiver 630, and the processor 610 may control the transceiver 630 to communicate with other devices, specifically, may send information or data to other devices, or receive other devices Information or data sent by a device.

Among them, the transceiver 630 may include a transmitter and a receiver. The transceiver 630 may further include antennas, and the number of the antennas may be one or more.

Optionally, the communication device 600 may specifically be the sending end device of the embodiment of the present application, and the communication device 600 may implement the corresponding processes implemented by the sending end device in each method of the embodiment of the present application. Repeat.

In some embodiments, the transceiver 630 in the communication device 600 may be used to implement the related operations of the communication module 410 in the transmitting end device 400 shown in FIG. 16 , which will not be repeated here for brevity.

In some embodiments, the processor 610 in the communication device 600 may be used to implement the related operations of the processing module 420 and the encoding module 430 in the transmitting end device 400 shown in FIG. 16 , which are not repeated here for brevity.

Optionally, the communication device 600 may specifically be the receiving end device of the embodiment of the present application, and the communication device 600 may implement the corresponding processes implemented by the receiving end device in each method of the embodiment of the present application. Repeat.

In some embodiments, the transceiver 630 in the communication device 600 may be used to implement the related operations of the communication module 510 in the receiving end device 500 shown in FIG. 17 , which will not be repeated here for brevity.

In some embodiments, the processor 610 in the communication device 600 may be used to implement the related operations of the decoding module 520 and the processing module in the receiver device 500 shown in FIG. 17 , which are not repeated here for brevity.

FIG. 19 is a schematic structural diagram of a chip according to an embodiment of the present application. The chip 700 shown in FIG. 19 includes a processor 710, and the processor 710 can call and run a computer program from a memory, so as to implement the method in the embodiment of the present application.

Optionally, as shown in FIG. 19 , the chip 700 may further include a memory 720 . The processor 710 may call and run a computer program from the memory 720 to implement the methods in the embodiments of the present application.

The memory 720 may be a separate device independent of the processor 710 , or may be integrated in the processor 710 .

Optionally, the chip 700 may further include an input interface 730 . The processor 710 may control the input interface 730 to communicate with other devices or chips, and specifically, may acquire information or data sent by other devices or chips.

Optionally, the chip 700 may further include an output interface 740 . The processor 710 can control the output interface 740 to communicate with other devices or chips, and specifically, can output information or data to other devices or chips.

Optionally, the chip can be applied to the transmitting end device in the embodiment of the present application, and the chip can implement the corresponding processes implemented by the transmitting end device in each method of the embodiment of the present application, which is not repeated here for brevity.

In some embodiments, the input interface 730 and the output interface 740 in the chip 700 may be used to implement related operations of the communication module 410 in the transmitting end device 400 shown in FIG. 16 , which are not repeated here for brevity.

In some embodiments, the processor 710 in the chip 700 may be used to implement the related operations of the processing module 420 and the encoding module 430 in the transmitting end device 400 shown in FIG. 16 , which are not repeated here for brevity.

Optionally, the chip can be applied to the receiving end device in the embodiment of the present application, and the chip can implement the corresponding processes implemented by the receiving end device in each method of the embodiment of the present application, which is not repeated here for brevity.

In some embodiments, the input interface 730 and the output interface 740 in the chip 700 may be used to implement the related operations of the communication module 510 in the receiving end device 500 shown in FIG. 17 , which are not repeated here for brevity.

In some embodiments, the processor 710 in the chip 700 may be used to implement the related operations of the decoding module 520 and the processing module in the receiving end device 500 shown in FIG. 17 , which are not repeated here for brevity.

It should be understood that the chip mentioned in the embodiments of the present application may also be referred to as a system-on-chip, a system-on-chip, a system-on-chip, or a system-on-a-chip, or the like.

FIG. 20 is a schematic block diagram of a communication system 900 provided by an embodiment of the present application. As shown in FIG. 20 , the communication system 900 includes a transmitter device 910 and a receiver device 920 .

Wherein, the transmitting end device 910 can be used to realize the corresponding functions realized by the transmitting end device in the above method, and the receiving end device 920 can be used to realize the corresponding functions realized by the receiving end device in the above method. For brevity, in This will not be repeated here.

It should be understood that the processor in this embodiment of the present application may be an integrated circuit chip, which has a signal processing capability. In the implementation process, each step of the above method embodiments may be completed by a hardware integrated logic circuit in a processor or an instruction in the form of software. The above-mentioned processor can be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), an off-the-shelf programmable gate array (Field Programmable Gate Array, FPGA) or other available Programming logic devices, discrete gate or transistor logic devices, discrete hardware components. The methods, steps, and logic block diagrams disclosed in the embodiments of this application can be implemented or executed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in conjunction with the embodiments of the present application may be directly embodied as executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other storage media mature in the art. The storage medium is located in the memory, and the processor reads the information in the memory, and completes the steps of the above method in combination with its hardware.

It can be understood that the memory in this embodiment of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. Wherein, the non-volatile memory may be a read-only memory (Read-Only Memory, ROM), a programmable read-only memory (Programmable ROM, PROM), an erasable programmable read-only memory (Erasable PROM, EPROM), an electrically programmable read-only memory (Erasable PROM, EPROM). Erase programmable read-only memory (Electrically EPROM, EEPROM) or flash memory. Volatile memory may be Random Access Memory (RAM), which acts as an external cache. By way of illustration and not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (Double Data Rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (Enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (Synchlink DRAM, SLDRAM) ) and direct memory bus random access memory (Direct Rambus RAM, DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but not be limited to, these and any other suitable types of memory.

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

Embodiments of the present application further provide a computer-readable storage medium for storing a computer program.

Optionally, the computer-readable storage medium can be applied to the sending end device in the embodiments of the present application, and the computer program enables the computer to execute the corresponding processes implemented by the sending end device in each method of the embodiments of the present application. For brevity, It is not repeated here.

Optionally, the computer-readable storage medium can be applied to the receiving-end device in the embodiments of the present application, and the computer program enables the computer to execute the corresponding processes implemented by the receiving-end device in each method of the embodiments of the present application. For brevity, It is not repeated here.

Embodiments of the present application also provide a computer program product, including computer program instructions.

Optionally, the computer program product can be applied to the sending end device in the embodiments of the present application, and the computer program instructions cause the computer to execute the corresponding processes implemented by the sending end device in each method of the embodiments of the present application. This will not be repeated here.

Optionally, the computer program product can be applied to the receiving end device in the embodiments of the present application, and the computer program instructions cause the computer to execute the corresponding processes implemented by the receiving end device in the various methods of the embodiments of the present application. This will not be repeated here.

The embodiments of the present application also provide a computer program.

Optionally, the computer program can be applied to the sending end device in the embodiments of the present application, and when the computer program is run on the computer, the computer is made to execute the corresponding processes implemented by the sending end device in each method of the embodiments of the present application, For brevity, details are not repeated here.

Optionally, the computer program can be applied to the receiving-end device in the embodiments of the present application, and when the computer program is run on the computer, the computer is made to execute the corresponding processes implemented by the receiving-end device in the various methods of the embodiments of the present application, For brevity, details are not repeated here.

Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each particular application, but such implementations should not be considered beyond the scope of this application.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described systems, devices and units may refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

In the several embodiments provided in this application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or Can be integrated into another system, or some features can be ignored, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing module, or each unit may exist physically alone, or two or more units may be integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as independent products, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other media that can store program codes .

The above are only specific embodiments of the present application, but the protection scope of the present application is not limited to this. should be covered within the scope of protection of this application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

A method for channel information feedback, comprising:

The transmitting end device receives the reference signal sent by the receiving end device;

Perform channel estimation according to the reference signal to obtain channel information between the transmitting end device and the receiving end device;

Perform feature decomposition on the channel information to obtain at least one first feature vector;

The at least one first feature vector is encoded by a neural network to obtain a target bit stream;

Send the target bitstream to the receiver device.
The method according to claim 1, wherein the channel information includes channel information corresponding to each subcarrier group in the multiple subcarrier groups of the transmitting end device, and the at least one first eigenvector includes the Eigenvectors corresponding to each subcarrier group in the multiple subcarrier groups respectively.
The method according to claim 1 or 2, wherein the encoding the at least one first feature vector through a neural network to obtain a target bit stream comprises:

The at least one first eigenvector is spliced into a matrix of eigenvectors and input to the neural network;

The target bit stream is obtained by encoding the feature vector matrix as an image to be compressed through the neural network.
The method according to claim 3, wherein the neural network is a convolutional neural network.
The method according to claim 1 or 2, wherein the neural network is a recurrent neural network, and the at least one first feature vector is encoded by the neural network to obtain a target bit stream, comprising:

inputting each feature vector in the at least one first feature vector into the recurrent neural network in sequence;

The target bit stream is obtained by encoding each feature vector as an element of a sequence through the recurrent neural network.
The method of claim 5, wherein the recurrent neural network comprises a long short-term memory LSTM neural network.
The method according to any one of claims 1-6, wherein the encoding the at least one first feature vector through a neural network to obtain a target bit stream comprises:

The at least one first feature vector is encoded based on an attention mechanism to obtain the target bitstream.
The method according to claim 7, wherein the encoding of the at least one first feature vector based on an attention mechanism to obtain the target bitstream comprises:

Extracting correlation features between elements in the at least one first feature vector based on the attention mechanism, to obtain at least one second feature vector;

Feature compression is performed on the at least one second feature vector to obtain the target bit stream.
The method according to any one of claims 1-8, wherein the transmitting end device is a terminal device, and the receiving end device is a network device.
A method for channel information feedback, comprising:

The receiving end device receives the target bit stream sent by the sending end device, the target bit stream is obtained by the sending end device by encoding at least one first feature vector, and the at least one first feature vector is the pair of the sending end device. The result of channel estimation is obtained by eigendecomposition;

The target bit stream is decoded through a neural network to obtain at least one target feature vector.
The method according to claim 10, wherein the channel information includes channel information corresponding to each subcarrier group in the multiple subcarrier groups of the transmitting end device, and the at least one first eigenvector includes all the subcarrier groups. Eigenvectors corresponding to each subcarrier group in the multiple subcarrier groups respectively.
The method according to claim 10 or 11, wherein the decoding the target bit stream through a neural network to obtain at least one target feature vector, comprising:

inputting the target bit stream to the neural network;

Decoding the target bitstream by using the neural network as information obtained by encoding an image to obtain a target image, where the target image includes the at least one target feature vector.
The method according to claim 12, wherein the neural network is a convolutional neural network.
The method according to claim 10 or 11, wherein the neural network is a recurrent neural network, and the target bit stream is decoded by the neural network to obtain at least one target feature vector, comprising:

inputting the target bit stream to the recurrent neural network;

The target bit stream is decoded by using the cyclic neural network as information obtained by encoding the sequence to obtain a target sequence, where the target sequence includes the at least one target feature vector.
The method of claim 14, wherein the recurrent neural network comprises a long short term memory LSTM neural network.
The method according to claim 10 or 11, wherein the decoding the target bit stream through a neural network to obtain at least one target feature vector, comprising:

The target bitstream is decoded based on the attention mechanism to obtain at least one target feature vector.
The method according to claim 16, wherein the decoding of the target bitstream based on the attention mechanism to obtain at least one target feature vector, comprising:

Perform feature extraction on the target bit stream to obtain a first feature map;

Determine the weight of the element in the first feature map based on the attention mechanism;

Dot multiplication of the weights of the elements in the first feature map and the first feature map to obtain a second feature map;

Decompress the second feature map to obtain the at least one target feature vector.
The method according to any one of claims 10-17, wherein the method further comprises:

The receiving end device performs precoding according to the at least one target feature vector.
A sending end device, characterized in that it includes:

The communication module is used to receive the reference signal sent by the receiver device;

a processing module, configured to perform channel estimation according to the reference signal, obtain channel information between the transmitting end device and the receiving end device, and perform feature decomposition on the channel information to obtain at least one first feature vector;

an encoding module, configured to encode the at least one first feature vector through a neural network to obtain a target bit stream;

The communication module is further configured to: send the target bit stream to the receiving end device.
The transmitting end device according to claim 19, wherein the channel information comprises channel information corresponding to each subcarrier group in the plurality of subcarrier groups of the transmitting end device, the at least one first eigenvector The eigenvectors corresponding to each subcarrier group in the multiple subcarrier groups are included.
The transmitting end device according to claim 19 or 20, wherein the encoding module is specifically used for:

The at least one first eigenvector is spliced into a matrix of eigenvectors and input to the neural network;

The target bit stream is obtained by encoding the feature vector matrix as an image to be compressed through the neural network.
The sender device according to claim 21, wherein the neural network is a convolutional neural network.
The sender device according to claim 19 or 21, wherein the neural network is a recurrent neural network, and the encoding module is further used for:

inputting each feature vector in the at least one first feature vector into the recurrent neural network in turn;

The target bit stream is obtained by encoding each feature vector as an element of a sequence through the recurrent neural network.
The sender device according to claim 23, wherein the recurrent neural network comprises a long short-term memory (LSTM) neural network.
The transmitting end device according to any one of claims 19-24, wherein the encoding module is further configured to:

The at least one first feature vector is encoded based on an attention mechanism to obtain the target bitstream.
The transmitting end device according to claim 25, wherein the encoding module further comprises:

an attention module, configured to extract the correlation feature between elements in the at least one first feature vector based on the attention mechanism, to obtain at least one second feature vector;

A feature vector compression module, configured to perform feature compression on the at least one second feature vector to obtain the target bit stream.
The transmitting end device according to any one of claims 19-26, wherein the transmitting end device is a terminal device, and the receiving end device is a network device.
A receiving end device, comprising:

A communication module, configured to receive a target bit stream sent by the sending end device, where the target bit stream is obtained by encoding at least one first feature vector by the sending end device, and the at least one first feature vector is the sending end device Obtained by the device performing eigendecomposition on the channel estimation result;

The decoding module is used for decoding the target bit stream through the neural network to obtain at least one target feature vector.
The receiving end device according to claim 28, wherein the channel information comprises channel information corresponding to each subcarrier group in the plurality of subcarrier groups of the transmitting end device, the at least one first eigenvector The eigenvectors corresponding to each subcarrier group in the multiple subcarrier groups are included.
The receiver device according to claim 28 or 29, wherein the decoding module is used for:

inputting the target bit stream to the neural network;

Decoding the target bitstream by using the neural network as information obtained by encoding an image to obtain a target image, where the target image includes the at least one target feature vector.
The receiver device according to claim 30, wherein the neural network is a convolutional neural network.
receiver equipment according to claim 28 or 29, is characterized in that, described neural network is cyclic neural network, and described decoding module is also used for:

inputting the target bit stream to the recurrent neural network;

The target bit stream is decoded by using the cyclic neural network as information obtained by encoding the sequence to obtain a target sequence, where the target sequence includes the at least one target feature vector.
The receiver device according to claim 32, wherein the recurrent neural network comprises a long short-term memory (LSTM) neural network.
The receiver device according to any one of claims 28-33, wherein the decoding module is further configured to:

The target bitstream is decoded based on the attention mechanism to obtain at least one target feature vector.
The receiver device according to claim 34, wherein the encoding module comprises:

a feature extraction module, configured to perform feature extraction on the target bitstream to obtain a first feature map;

an attention module, configured to determine the weight of the elements in the first feature map based on the attention mechanism;

a dot product module, configured to perform dot product on the first feature map and the weights of the elements in the first feature map to obtain a second feature map;

A decompression module, configured to decompress the second feature map to obtain the at least one target feature vector.
The receiving end device according to any one of claims 28-35, wherein the receiving end device further comprises:

A processing module, configured to perform precoding according to the at least one target feature vector.
A sending end device is characterized in that, comprising: a processor and a memory, the memory is used for storing a computer program, the processor is used for calling and running the computer program stored in the memory, and executes the program as claimed in claims 1 to 9 The method of any one.
A chip, characterized by comprising: a processor for calling and running a computer program from a memory, so that a device installed with the chip executes the method according to any one of claims 1 to 9.
A computer-readable storage medium, characterized by being used for storing a computer program, the computer program causing a computer to execute the method according to any one of claims 1 to 9.
A computer program product comprising computer program instructions, the computer program instructions causing a computer to perform the method of any one of claims 1 to 9.
A computer program, characterized in that the computer program causes a computer to perform the method according to any one of claims 1 to 9.
A network device, characterized in that it comprises: a processor and a memory, the memory is used to store a computer program, the processor is used to call and run the computer program stored in the memory, and execute any one of claims 10 to 18. one of the methods described.
A chip, characterized by comprising: a processor for calling and running a computer program from a memory, so that a device installed with the chip executes the method according to any one of claims 10 to 18 .
A computer-readable storage medium, characterized by being used for storing a computer program, the computer program causing a computer to execute the method according to any one of claims 10 to 18.
A computer program product comprising computer program instructions, the computer program instructions causing a computer to perform the method of any one of claims 10 to 18.
A computer program, characterized in that the computer program causes a computer to perform the method according to any one of claims 10 to 18.