Method performed by first node in wireless communication system and related device

Abstract

The invention provides a method executed by a first node in a wireless communication system and related equipment. The method comprises the following steps: determining target position information through a first AI network; through a second AI network, channel prediction is carried out based on the channel information at multiple historical moments and the target position information; wherein the target position information is used for representing the similarity between the channels and the correlation between the time-frequency position information of the corresponding channels. According to the channel prediction method provided by the invention, by constructing the target position information, the time-frequency position information of the channels and the similarity between the channels are effectively utilized, and the prediction performance is improved. Optionally, the method executed by the first node can be executed by using an artificial intelligence model.

Description

Technical Field

[0001] This disclosure relates to the field of communication technology, and more specifically, to a method and related apparatus performed by a first node in a wireless communication system. Background Technology

[0002] Channel State Information (CSI) feedback between User Equipment (UE) and Base Station (gNB) plays a crucial role in the quality of Multi-input Multi-output (MIMO) communication systems.

[0003] However, achieving high-precision CSI with low overhead is a significant challenge due to environmental variations, Doppler and multipath effects in the channel. Summary of the Invention

[0004] The purpose of this disclosure is to at least solve one of the aforementioned technical defects. The technical solution provided by the embodiments of this disclosure is as follows:

[0005] In a first aspect, embodiments of this disclosure provide a method executed by a first node in a wireless communication system, comprising:

[0006] The target location information is determined by using the first AI network based on the time-frequency location information of the channel.

[0007] Channel prediction is performed using a second AI network based on channel information from multiple historical moments and the target location information.

[0008] The target location information is used to characterize the similarity between channels and the correlation between the time-frequency location information of the corresponding channels.

[0009] In one feasible embodiment, the channel prediction based on channel information from multiple historical moments and the target location information includes:

[0010] Channel prediction is performed based on the first channel information and the target location information to obtain a first prediction result;

[0011] Based on the first prediction result, the second channel information, and the target location information, channel prediction is performed to obtain the second prediction result;

[0012] Wherein, the first channel information includes channel information of multiple historical moments after decompression; and / or, the second channel information includes at least one of the following: channel information of multiple historical moments to be decompressed, channel information of multiple historical moments after decompression, or channel information based on the precoding index matrix of the first type codebook or the second type codebook.

[0013] In one feasible embodiment, determining the target location information includes:

[0014] Determine the relative position information between the time position and frequency position corresponding to the channel information at the historical moment;

[0015] Based on the channel distribution information of the preset scenario, a grid is divided to obtain a two-dimensional grid related to Doppler delay;

[0016] Based on the two-dimensional grid, the relative position information is mapped to obtain the target position information.

[0017] In one feasible embodiment, the process of dividing the channel distribution information based on a preset scenario into a grid to obtain a two-dimensional grid related to Doppler delay includes:

[0018] Based on multiple first change information of the channel changing with time and multiple second change information of the channel changing with frequency under a preset scenario, a two-dimensional grid is divided, and each grid corresponds to one first change information and one second change information.

[0019] Each of the first change information is related to the Doppler factor corresponding to each channel path; each of the second change information is related to the time delay factor corresponding to each channel path.

[0020] In one feasible embodiment, mapping the relative position information based on the two-dimensional grid to obtain the target position information includes:

[0021] Based on the two-dimensional grid, the relative position information corresponding to different historical moments and the channel-related information between the channels are determined;

[0022] Based on the channel-related information, the target location information is determined.

[0023] In one feasible embodiment, determining the channel-related information between the relative position information corresponding to different historical moments and the channel based on the two-dimensional grid includes:

[0024] Based on the two-dimensional grid, the similarity between the real part of the channel correlation coefficient between the relative position information at different historical moments and the channel is used to determine the first position coding information;

[0025] The determination of target location information based on the channel-related information includes:

[0026] Based on the first location encoding information, the target location information is determined.

[0027] In one feasible embodiment, determining the channel-related information between the relative position information corresponding to different historical moments and the channel based on the two-dimensional grid further includes:

[0028] The second position coding information is determined based on the similarity of the imaginary parts in the channel correlation coefficient.

[0029] The determination of target location information based on the channel-related information includes:

[0030] The target location information is determined based on the first location encoding information and the second location encoding information.

[0031] In a feasible embodiment, the channel prediction based on the first channel information and the target location information includes:

[0032] Based on the first channel information, a first feature related to time delay and a second feature related to Doppler are extracted;

[0033] Based on the target location information, the first feature, and the second feature, the positional similarity of the Doppler time delay is determined;

[0034] Channel prediction is performed based on the positional similarity of the Doppler delay.

[0035] In one feasible embodiment, the channel prediction based on the positional similarity of the Doppler delay includes:

[0036] Based on the first channel matrix corresponding to the transmit antenna of the first node, the channel similarity between channels is determined;

[0037] Channel prediction is performed based on the location similarity and the channel similarity.

[0038] In one feasible embodiment, determining the channel similarity between channels based on the first channel matrix corresponding to the transmit antenna of the first node includes:

[0039] Obtain a second channel matrix from the first channel matrix that corresponds to the channel information of a portion of the transmitting antennas;

[0040] A linear transformation is performed based on the second channel matrix to obtain the query value and the key value;

[0041] Based on the query value and the key value, the channel similarity between channels is determined.

[0042] In a second aspect, embodiments of this disclosure provide an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method provided in the first aspect and any of its embodiments.

[0043] Thirdly, embodiments of this disclosure provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect and any of its embodiments.

[0044] Fourthly, embodiments of this disclosure provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the method provided in the first aspect and any of its embodiments.

[0045] The beneficial effects of the technical solutions provided in this disclosure are:

[0046] This disclosure provides a channel prediction method. Specifically, a first AI network determines target location information, which characterizes the correlation between the similarity between channels and the corresponding time-frequency location information of the channels. Then, a second AI network performs channel prediction based on channel information from multiple historical moments and the target location information. This disclosure enables the effective utilization of the time-frequency location information and the similarity between channels in channel prediction by constructing an internal structured relationship between the time-frequency location information of the channels and the similarity between channels. Furthermore, this disclosure allows for the processing of channel information from multiple historical moments, which is beneficial for improving prediction performance. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments of this disclosure will be briefly introduced below.

[0048] Figure 1 A flowchart illustrating a method executed by a first node, as provided in this embodiment of the disclosure;

[0049] Figure 2 A schematic diagram illustrating a joint decompression prediction process between the UE and the base station, provided as an embodiment of this disclosure;

[0050] Figure 3a This demonstrates a feasible operating architecture;

[0051] Figure 3b A schematic diagram of an architecture provided for an embodiment of this disclosure;

[0052] Figure 3c A schematic diagram of a base station-side joint CSI decompression prediction model provided in an embodiment of this disclosure;

[0053] Figure 4 A schematic diagram of a time-frequency integrated location coding provided in an embodiment of this disclosure;

[0054] Figure 5 A schematic diagram of a time-frequency integrated two-dimensional relative position coding provided in an embodiment of this disclosure;

[0055] Figure 6 A schematic diagram illustrating channel similarity calculation based on antenna polarization characteristics, provided for an embodiment of this disclosure;

[0056] Figure 7 A schematic diagram illustrating a low-complexity channel dot product similarity calculation provided in this embodiment of the disclosure;

[0057] Figure 8 A schematic diagram of a network architecture provided in an embodiment of this disclosure;

[0058] Figure 9 This is a schematic diagram of a channel distribution provided in an embodiment of the present disclosure;

[0059] Figure 10 A schematic structural diagram of an electronic device provided in an embodiment of the present disclosure. Detailed Implementation

[0060] The following description, with reference to the accompanying drawings, is provided to aid in a thorough understanding of the various embodiments of this disclosure as defined by the claims and their equivalents. This description includes various specific details to aid understanding but should be considered exemplary only. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the various embodiments described herein without departing from the scope and spirit of this disclosure. Furthermore, for clarity and brevity, descriptions of well-known functions and structures may be omitted.

[0061] The terms and wording used in the following description and claims are not limited to their dictionary meanings, but are merely used by the inventors to enable a clear and consistent understanding of this disclosure. Therefore, it will be apparent to those skilled in the art that the following description of various embodiments of this disclosure is for illustrative purposes only and not for limiting the purpose of this disclosure as defined in the appended claims and their equivalents.

[0062] It should be understood that the singular forms of “a,” “an,” and “the” can also include plural references unless the context clearly indicates otherwise. Thus, for example, the reference to “component surface” includes referring to one or more such surfaces. When we say that an element is “connected” or “coupled” to another element, the element can be directly connected or coupled to the other element, or it can mean that the element and the other element are connected through an intermediate element. Furthermore, the use of “connected” or “coupled” herein can include wireless connections or wireless couplings.

[0063] The terms “comprising” or “may include” refer to the presence of a corresponding disclosed function, operation, or component that may be used in the various embodiments of this disclosure, rather than limiting the presence of one or more additional functions, operations, or features. Furthermore, the terms “comprising” or “having” may be interpreted as indicating certain characteristics, numbers, steps, operations, constituent elements, components, or combinations thereof, but should not be construed as excluding the possibility of the presence of one or more other characteristics, numbers, steps, operations, constituent elements, components, or combinations thereof.

[0064] The term "or" as used in the various embodiments of this disclosure includes any of the listed terms and all combinations thereof. For example, "A or B" may include A, may include B, or may include both A and B. When describing multiple (two or more) items, if the relationship between the multiple items is not explicitly defined, the multiple items may refer to one, more, or all of the multiple items. For example, the description "parameter A includes A1, A2, A3" can be implemented as parameter A includes A1 or A2 or A3, or it can be implemented as parameter A includes at least two of the three items A1, A2, and A3.

[0065] Unless otherwise defined, all terms used in this disclosure (including technical or scientific terms) have the same meaning as understood by one of those skilled in the art as described herein. Common terms as defined in dictionaries are to be interpreted as having a meaning consistent with the context in the relevant technical field and should not be interpreted ideally or overly formally unless expressly defined in this disclosure.

[0066] At least some of the functions of the device or electronic device provided in this disclosure embodiment can be implemented by an AI model, such as implementing at least one module of a plurality of modules of the device or electronic device by an AI model. AI-related functions can be executed by non-volatile memory, volatile memory, and a processor.

[0067] The processor may include one or more processors. In this case, the one or more processors may be general-purpose processors, such as central processing unit (CPU), application processor (AP), etc., or pure graphics processing unit, such as graphics processing unit (GPU), vision processing unit (VPU), and / or AI-specific processors, such as neural processing unit (NPU).

[0068] The one or more processors control the processing of input data based on predefined operating rules or artificial intelligence (AI) models stored in non-volatile and volatile memory. These predefined operating rules or AI models are provided through training or learning.

[0069] Here, "providing through learning" refers to obtaining predefined operating rules or an AI model with desired characteristics by applying a learning algorithm to multiple learning datasets. This learning can be performed within the device or electronic device itself, in which the AI ​​is executed according to the embodiment, and / or can be implemented via a separate server / system.

[0070] AI models can contain multiple neural network layers. Each layer has multiple weight values, and each layer performs neural network computations by calculating the input data of that layer (such as the computation results of the previous layer and / or the input data of the AI ​​model) and the multiple weight values ​​of the current layer. Examples of neural networks include, but are not limited to, convolutional neural networks (CNNs), deep neural networks (DNNs), recurrent neural networks (RNNs), restricted Boltzmann machines (RBMs), deep belief networks (DBNs), bidirectional recurrent deep neural networks (BRDNNs), generative adversarial networks (GANs), and deep Q-networks.

[0071] A learning algorithm is a method of training a predetermined target device (e.g., a robot) using multiple learning data sets to enable, allow, or control the target device to make determinations or predictions. Examples of such learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0072] According to this disclosure, at least one step in a channel prediction method performed in an electronic device, such as determining target location information, can be implemented using an artificial intelligence model. The processor of the electronic device can perform preprocessing operations on the data to transform it into a form suitable for use as input to the artificial intelligence model. The artificial intelligence model can be obtained through training. Here, "obtained through training" means obtaining a predefined operating rule or artificial intelligence model configured to perform desired features (or objectives) by training a basic artificial intelligence model with multiple training data using a training algorithm.

[0073] In the field of communication technology, CSI (Channel State Information) can be used to evaluate and describe the characteristics of communication channels, helping transmitters and receivers make adjustments and decisions during communication. By acquiring CSI, communication systems can achieve highly reliable and high-speed communication in multi-antenna systems.

[0074] Feedback from the Customer Service Interface (CSI) between the UE and the base station plays a crucial role in the quality of MIMO communication systems. Existing CSI technologies employ Transformer-based AI (Artificial Intelligence) models for compression on the user side and decompression on the base station side. This treats the channel, including angle and latency, as an image, utilizing the Transformer's encoder and decoder for compression and decompression respectively. However, due to complex channel variations, existing processing methods suffer from low efficiency and prediction accuracy, resulting in poor generalization performance.

[0075] To address at least one of the aforementioned technical problems, this disclosure proposes a Channel Attention Mechanism based Unified CSI feedback and prediction (CAM-UniCSI) architecture. Under this architecture, the solution of this disclosure achieves unified CSI feedback and prediction based on the decoder and channel characteristics of the attention mechanism.

[0076] The following description of several optional embodiments illustrates the technical solutions of this disclosure and the technical effects produced by these solutions. It should be noted that the following embodiments can be referenced, learned from, or combined with each other. Identical terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0077] Figure 1 This is a flowchart illustrating a channel prediction method provided in an embodiment of the present disclosure. The method provided in this disclosure will be specifically described below, taking the first node as a base station as an example and the base station as the execution subject.

[0078] Among them, such as Figure 2 As shown in the embodiments of this disclosure, a scheme for joint channel decompression and prediction on the base station side is provided. The base station sends a reference signal to the UE, the UE receives the reference signal and estimates the channel, and then the UE uses a DNN (Deep Neural Network) to extract and compress channel features, and sends the compressed channel information to the base station. The base station can decompress and predict the received channel information to be decompressed.

[0079] Specifically, such as Figure 1 As shown, the method may include S101-S102:

[0080] S101. Determine the target location information through the first AI network.

[0081] The target location information is used to characterize the similarity between channels and the correlation between the time-frequency location information of the corresponding channels.

[0082] Optionally, for the time-frequency location information of the channel (which may include time location and frequency location), this embodiment of the disclosure takes into account the coupling effect of time and frequency changes in the channel response and the channel changes with time and frequency, and constructs a correlation relationship between the time location, frequency location (time-frequency location information) and the channel similarity, so as to effectively utilize the target location information in subsequent CSI prediction and improve the prediction performance.

[0083] S102. Channel prediction is performed using the second AI network based on channel information from multiple historical moments and the target location information.

[0084] Optionally, when channel information compression is performed on the UE side (which may be a second node communicating with the first node), the reported channel information is compressed one time at a time. On the base station side, channel information from multiple historical times can be accumulated and combined with target location information for joint decompression and prediction. When the target location information is effectively utilized, accurate CSI is provided, thus solving the channel aging problem.

[0085] In this embodiment of the disclosure, the first AI network and the second AI network can be independent AI networks, such as... Figure 3c As shown, the first AI network performs time-frequency integrated location coding based on the input location information 1 or location information 2 to obtain the target location information. The second AI network can perform joint decompression prediction based on the input channel information and target location information, and output the channel prediction result. Optionally, the first AI network and the second AI network can also be regarded as the same AI network deployed on the base station side, and the output of the first AI network can be used as the input of the second AI network.

[0086] In one feasible embodiment, the second AI network includes a first AI module and a second AI module. For example... Figure 3c As shown, the second AI network may include an N-layer progressively connected decoder based on an attention mechanism (Transformer), wherein the first AI module may be a channel self-attention layer and the second AI module may be a channel cross-attention layer.

[0087] In one feasible embodiment, such as Figure 3a As shown, an AI-based CSI decoder can first reconstruct a single CSI from the compressed CSI (CCSI) fed back by the UE, and then use another independent AI-based CSI predictor to predict the accumulated channels at multiple time points to obtain the predicted CSI.

[0088] In another feasible embodiment, for the proposed unified CSI architecture, such as Figure 3bAs shown, a self-attention layer (masked multi-head channel attention layer) can be used to extract the correlation between historical CSI and future CSI. Then, by introducing more CSI, a cross-attention layer (multi-head cross-channel attention) is used to predict future CSI. Figure 3b In the architecture shown, the domain transformation and feature extraction capabilities of the transformer can be utilized to directly predict future CSI from the accumulated CCSI, relative to... Figure 3a The architecture shown can reduce processing complexity and latency, and improve prediction efficiency.

[0089] Optionally, in S102, channel prediction is performed based on channel information from multiple historical moments and the target location information, including steps A1-A2:

[0090] Step A1: Perform channel prediction based on the first channel information and the target location information to obtain the first prediction result.

[0091] The first channel information includes channel information from multiple historical moments after decompression.

[0092] Step A2: Based on the first prediction result, the second channel information, and the target location information, perform channel prediction to obtain the second prediction result.

[0093] The second channel information includes at least one of the following: channel information of multiple historical moments to be decompressed, channel information of multiple historical moments after decompression, or channel information based on the precoding index matrix of the first type codebook or the second type codebook.

[0094] For example, in the joint decompression prediction at the base station side, depending on the input information, it can include at least three of the following cases:

[0095] The first type: such as Figure 5 As shown, the channel information 1 input from the channel attention layer may include a small amount of decompressed channel information, and the channel information 2 input from the channel cross-attention layer may include the channel information of the historical time to be decompressed reported by the UE. Finally, the decompressed predicted channel is output, realizing the integrated joint decompression prediction of the compressed channel. In this example, the channel decompression step can be omitted, and the compressed channel can be used directly for channel prediction, which is beneficial to improving the prediction efficiency.

[0096] The second type: such as Figure 5 As shown, the channel information of historical moments reported by the UE can be decompressed first, and then the accumulated decompressed historical moment channels can be used as channel information 2 as input, and the channels of multiple decompressed historical moments can be used as channel information 1 as input. This can enable the prediction of the decompressed channels and output the predicted channels.

[0097] The third type: such as Figure 5 As shown, the input to channel information 1 can include the channel information of the decompressed historical time, and the input to channel information 2 can include the channel information based on the precoding matrix indicator (PMI) of the first type I codebook or the second type II codebook. This can achieve technical compatibility with the Type I codebook used for CSI feedback in the 3GPP Rel.15 protocol and the Type II codebook in 3GPP Rel.16, and realize the prediction enhancement of PMI.

[0098] In the embodiments of this disclosure, the AI ​​model for channel compression on the UE side can be determined with reference to relevant technologies, and this disclosure does not limit it. The base station side can employ an improved attention-based decoder (Transformer decoder model) to perform joint decompression and prediction inference using accumulated channel information from multiple historical moments. During the model training phase, the UE side can perform compression and decompression training with reference to relevant technologies. The intermediate data and some real channel information generated during the training and testing of the UE-side AI model are used for training the base station-side model.

[0099] Before detailing the steps involved in constructing target location information and channel prediction, we will explain some function definitions used in the embodiments of this disclosure.

[0100] First, the attention mechanism described by kernel functions.

[0101] In this embodiment of the disclosure, the output of the attention layer in the Transformer can be defined such that, for a specific query q, the attention score between it and a certain key k is:

[0102]

[0103] The SoftMax function is defined as follows:

[0104]

[0105] If we describe the transform domain q and k using the input x = h + p and the corresponding transform matrix W, then we have:

[0106]

[0107] In the formula, Let x be the set of all values ​​v and keys k used to compute the weighted superposition.

[0108] In classical machine learning, a kernel function is a way to characterize similarity. The kernel function k(x,x′) is a mapping of the similarity between binary inputs x and x′ to their corresponding function values ​​y = f(x) and y′ = f(x′). From the perspective of kernel functions, the natural exponential kernel function can be defined as:

[0109]

[0110] The above formula can then be written as

[0111]

[0112] Based on the aforementioned calculation process, the attention mechanism in Transformer can be rewritten as an exponential kernel function as follows:

[0113]

[0114] In the formula v(x) k )=x k W k As can be seen, the output of the attention layer For all possible values ​​v(x) k The weighted summation of v(x) and v(x) k The corresponding weight is the normalized q = x. q W q and k = x k W k Similarity, i.e., attention score In the attention mechanism from the perspective of the kernel function described above, similarity is determined by the kernel function k(x). q ,x k To depict.

[0115] Alternatively, considering the definition of x, x = h + p, x can be written in the form of channel information h and corresponding position code p:

[0116] k(x q ,x k )=k exp (h q +p q ,h k +p k )

[0117] In the kernel function above, the result of superimposing the channel response h of the input antenna domain and the corresponding position coding vector p is used as the exponential kernel function k. exp (h q +p q ,h k +p k The input is ) and the similarity result k(x) is obtained. q,x k In this context, the similarity of the channel response h and the similarity calculated from the location information are inseparable, and the two similarities cannot be calculated independently. Therefore, a channel synthesis similarity kernel function that allows for independent calculation of channel similarity and location similarity can be considered, as follows:

[0118] k(x q ,x k )=k exp (h q ,h k )·k(s q ,s k )

[0119] In the formula s q ,s k They represent h respectively q ,h k The corresponding positional state information shows the kernel function k(x) q ,x k It consists of the product of two similarities: a) the state similarity kernel part k(s) related to the location and state information. q ,s k b) The channel similarity kernel function part k, which is independent of location information. exp (h q ,h k ).

[0120] Secondly, the channel complex correlation coefficient

[0121] In this embodiment of the disclosure, a channel response vector in the antenna domain can be defined. The function expression for the channel variation with time and subcarrier frequency is:

[0122]

[0123] In the formula, function g is a mapping from a binary variable to a channel response vector. In the formula, f is... c The center frequency of the propagating signal, c is the speed of light, and P is the number of multipath paths in the channel ρ. p v is the power corresponding to the p-th path. p The Doppler velocity corresponding to the p-th path, τ p The time delay corresponding to the p-th path, a p For N tx A 1×1 dimensional array response vector.

[0124] As can be seen from the above equation, the bivariate function of channel variation with time and frequency can be viewed as the result of the superposition of multiple component functions with different rates of change. However, the relationship between a single channel and its position is insufficient to characterize the attention information between two channels. Therefore, the correlation between the similarity between two channels and their corresponding target position information, i.e., state information, can be derived. This involves two time-frequency states (t... q ,f q ), (t k ,f k The correlation between the channel response vectors is the dot product of the two vectors, i.e., the complex-valued correlation coefficient.

[0125] α=h(t q ,f q )h(t k ,f k ) H

[0126] definition Its physical meaning is the p-th component of the rate of change of the channel in the time domain (the first change information of the channel over time). Different values ​​of p mean that the channel has different time-varying rate components, stemming from the different Doppler factors of each path; definition The p-th component represents the rate of change of the channel with respect to the carrier frequency (the second change information of the channel with respect to the carrier frequency). Different values ​​of p mean that the channel has different rate of change components that vary with the carrier frequency, which comes from the different multipath delays of each path.

[0127] Based on the above definition, the complex correlation coefficient α of the channel can be further expanded as follows:

[0128]

[0129] The correlation coefficient h(t) between the two channel response vectors in the above formula q ,f q )h(t k ,f k ) H For the exponentially scalar value of e, according to Euler's formula:

[0130] e iφ =cosφ+isinφ

[0131] Where cosφ is the real part and sinφ is the imaginary part, the correlation coefficient scalar can be written as a combination of the real and imaginary parts:

[0132]

[0133] in Let represent the real part of the multiple correlation coefficient. This represents the imaginary part of the multiple correlation coefficient.

[0134]

[0135] Considering the multipath sparsity of the channel, the different ω components are approximately orthogonal, meaning the cross term is close to 0 and can be regarded as a bias term. Using b1, we can obtain:

[0136]

[0137] The imaginary part of the channel correlation coefficient can be obtained in the same way:

[0138]

[0139] Third, channel location similarity

[0140] In the application scenario of this disclosure embodiment, the mapping from two different time-frequency states to the correlation between channels characterizes the similarity between them. Considering that the correlation coefficient of two channels is a complex number, and the similarity needs to be defined in the real number space, the complex dot product can be decomposed into the sum of the dot products of the real and imaginary parts. Therefore, for channels h(t) under two different states (time and frequency positions), p ,f p ) and h(t k ,f k ) ， The correlation between the two channels, i.e., the channels are:

[0141]

[0142] in, For two different states of the channel, t q ,f q Let q be the time and frequency position, and t be the position of frequency. k ,f k Let k be the time and frequency position.

[0143] First, consider the correlation result of the dot product of real vectors, and define... Its physical meaning is the p-th component of the rate of change of the channel in the time domain. Different values ​​of p mean that the channel has different time-varying rate components, stemming from the different Doppler factors of each path; definition The p-th component represents the rate of change of the channel with respect to the carrier frequency. Different values ​​of p in the equation imply that the channel has different rate-of-change components that vary with the carrier frequency, stemming from the different multipath delays of each path. The similarity of the real parts can be expressed as:

[0144]

[0145] Considering the multipath sparsity of the channel, the different ω components are approximately orthogonal, meaning the cross term is close to 0 and can be regarded as a bias term. Using b1, we can obtain:

[0146]

[0147] Similarly, the similarity between the imaginary parts of the channels can be derived as follows:

[0148]

[0149] The result of merging the real and imaginary parts is:

[0150]

[0151] Let t q -t k =Δt,f q -f k =Δf, the above equation can be further simplified to

[0152]

[0153] As can be seen from the above equation, the similarity between the antenna domain channel responses under any two different time-frequency states can be written in the form of a structured kernel function. The binary independent variables of this function are the time difference and frequency difference between the two states, and the value of the function represents the channel correlation level corresponding to the difference between the two states. From the function structure, the kernel function characterizing the channel similarity between two state positions is formed by the superposition of L multiplicative quantities, and the product factors of the p-th component are respectively... and Multipath energy The magnitude is related to the distribution characteristics of the channel itself; different user channels correspond to different... therefore The value of needs to be related to the corresponding channel.

[0154] The following is a detailed description of a channel time-frequency integrated location coding method provided in the embodiments of this disclosure.

[0155] Optionally, such as Figure 4As shown, the base station can obtain channel information and its corresponding two-dimensional time-domain and frequency-domain integrated location coding information (time-frequency integrated location coding information) through decompression, prediction, and SRS (Sounding Reference Signal), and input the two types of information into the self-attention part of the decoding layer of the Transformer.

[0156] In a feasible embodiment, the determination of the target location information in S101 includes steps B1-B3:

[0157] Step B1: Determine the relative position information between the time position and frequency position corresponding to the channel information at the historical moment.

[0158] Optionally, such as Figure 5 As shown, for N slot When processing the channel at time N, we can define N corresponding to each time point. c One carrier, all N c N slot 2D position state information corresponding to each input element The possible values ​​are:

[0159]

[0160] For a specific query q and a certain key k, their corresponding 2D absolute positions are:

[0161]

[0162] The relative positions are:

[0163]

[0164] Based on the above formula, the correlation mapping can be performed by combining the relative position information {Δt, Δf}.

[0165] Step B2: Based on the channel distribution information of the preset scenario, perform grid division to obtain a two-dimensional grid related to Doppler delay.

[0166] In this embodiment, considering that the parameters in the model need to adapt to all possible UE channels, information corresponding to all possible UE channels in a given application scenario can be considered as the model parameters. Furthermore, it is also considered that signals may traverse multiple channel paths during transmission, and multipath channels are often time-varying channels whose characteristics change over time, such as signal frequency changes caused by the relative motion between the UE and the base station. Based on this, the two-dimensional location information of the channel distribution information in a given scenario can be meshed to obtain a two-dimensional mesh related to Doppler delay.

[0167] Optionally, in step B2, the channel distribution information of the preset scenario is used to divide the grid to obtain a two-dimensional grid related to Doppler delay. This includes step B22: based on multiple first change information of the channel changing with time and multiple second change information of the channel changing with frequency under the preset scenario, a two-dimensional grid is divided, and each grid obtained corresponds to one first change information and one second change information.

[0168] Each of the first change information is related to the Doppler factor corresponding to each channel path; each of the second change information is related to the time delay factor corresponding to each channel path.

[0169] In this embodiment of the disclosure, according to the channel multipath model, the complex correlation coefficient α between the channel response vectors of two antenna domains can be rewritten as a vector dot product based on its multinomial weighted superposition form:

[0170]

[0171]

[0172] in

[0173]

[0174]

[0175] In this embodiment of the disclosure, the purpose of position encoding is to obtain all possible β. p and γ p Since the parameters in the model need to be adapted to all possible UE channels, it is not simply a matter of taking P parameters corresponding to a single UE channel. and Instead of using the value as a parameter of the model, it considers all possible UE channels corresponding to a given application scenario. and The range of values ​​for ω, and the possible values ​​of ω t and ω f The value space is effectively divided into grids, into d... k Each subdivided grid corresponds to one and a In order to cover the channel parameters of all UEs, that is, to satisfy i∈[1,d k ]and j∈[1,d k ],in Its physical meaning can be determined through the channel model in 3GPP 38.901.

[0176] In this embodiment of the disclosure, to make the model applicable to multi-UE channel data that may occur in different scenarios, the channel distribution can be obtained by estimating the combination of channel Doppler and delay under different scenarios. For example... Figure 9 As shown, the two-dimensional space can be partitioned based on the estimated channel distribution (distribution range) and empirical probability distribution, resulting in better generalization performance for the AI ​​model. Multipath factor {ω} t ,ω f The two-dimensional space of} can be divided as shown by the following formula:

[0177]

[0178]

[0179] In the above formula, the maximum and minimum values ​​of the multipath factor can be determined by the UE's speed range, LOS mode, channel scenario, etc.

[0180] Step B3: Based on the two-dimensional grid, map the relative position information to obtain the target position information.

[0181] Optionally, the mapping process can be a correlation from time-frequency state to inter-channel correlation. This can be achieved by constructing a correlation between relative position information and corresponding channel similarity, i.e., a correlation between state information, based on a two-dimensional grid, to obtain the target position information. Optionally, the mapping can be a non-linear, high-dimensional mapping.

[0182] Optionally, in step B3, the relative position information is mapped based on the two-dimensional grid to obtain the target position information, including steps B31-B32:

[0183] Step B31: Based on the two-dimensional grid, determine the relative position information and channel-related information between the channels corresponding to different historical moments.

[0184] Optionally, step B31 involves determining the channel-related information between the relative position information corresponding to different historical moments and the channel based on the two-dimensional grid, including step B311:

[0185] Based on the two-dimensional grid, the similarity of the real part of the channel correlation coefficient between the relative position information at different historical moments and the channel is used to determine the first position coding information.

[0186] Among them, the result based on the real part of the channel correlation coefficient Ignoring the bias b1 and comparing it with the one-dimensional positional encoding, it can be seen that the part constituting the 2D positional encoding is... The corresponding encoding process is as follows:

[0187]

[0188] in

[0189] Among them, for location encoding with only one dimension of location information, N c Location encoding vector of location information The encoding result p at position i i for:

[0190]

[0191] Among them, it can be determined that d = d model .

[0192] Optionally, step B31 may also include step B312: determining the second position coding information based on the similarity of the imaginary part in the channel correlation coefficient.

[0193] Based on the similarity between the real parts of the channel Ignoring the bias b1 and comparing it with the 1D position code, it can be seen that the part constituting the 2D position code is The corresponding encoding process is as follows:

[0194]

[0195] in

[0196] Step B32: Determine the target location information based on the channel-related information.

[0197] In the above embodiment, the encoding result is the result of separating the real part and the imaginary part of the correlation coefficient. Considering that the calculation of real number similarity is supported when based on the Transformer decoder structure, it is necessary to merge the encoding of the real part and the imaginary part.

[0198] Optionally, step B32, based on the channel-related information, determines the target location information, including steps B321 and / or B322:

[0199] Step B321: Determine the target location information based on the first location encoding information.

[0200] Converting the encoded real part into a vector, we get:

[0201]

[0202] Step B322: Determine the target location information based on the first location encoding information and the second location encoding information.

[0203] Optionally, when processing the result of superimposing the real and imaginary parts, i.e., for a certain time-frequency two-dimensional state... And another time-frequency two-dimensional state The final encoding result of the relative positions of the two. For: summing the encoding matrices of the real and imaginary parts and stretching them into a vector, i.e.

[0204]

[0205] Based on the above embodiments, and considering the possibility of complex attention mechanisms, the encoding results that simultaneously retain both the real and imaginary parts can also be preserved. and

[0206]

[0207]

[0208] Encoding results that preserve both real and imaginary parts can be compatible with subsequent attention mechanisms for complex numbers.

[0209] In this embodiment of the disclosure, a comprehensive coding result is obtained based on the channel multipath propagation model and relative position information, taking into account the coupling effect of time and frequency changes in the channel response.

[0210] The following describes a method for calculating positional similarity considering Doppler time delay, as proposed in the embodiments of this disclosure.

[0211] Optionally, time delay features and Doppler features can be extracted using channel information from multiple time points, and the channel time-frequency position similarity of Doppler and time delay features can be obtained based on the result of time-frequency integrated position coding (target position information).

[0212] In a feasible embodiment, step A1 involves channel prediction based on the first channel information and the target location information, including steps A11-A13:

[0213] Step A11: Based on the first channel information, extract the first feature related to time delay and the second feature related to Doppler.

[0214] In this embodiment of the disclosure, the input channel sequence information (first channel information) H = [h1, h2, ..., h L ], L=N c N slot and the corresponding time-frequency two-dimensional state sequence information S=[s1,s2,…,s L First, channel features in the Doppler-delay domain need to be extracted using channel information corresponding to all carriers at all times, resulting in feature matrix C:

[0215]

[0216] W1 and W2 are learnable feature extraction parameters.

[0217] Step A12: Based on the target location information, the first feature, and the second feature, determine the positional similarity of the Doppler time delay.

[0218] Optionally, based on the result of two-dimensional relative position encoding in the time-frequency domain (target position information) Calculate the similarity of the state kernel function related to location state information: Convert the feature matrix in the Doppler-time-delay domain into a vector, and take the dot product with the relative position encoding vector in the time-frequency domain to obtain the location similarity score related to the location information.

[0219]

[0220] Considering the possibility of complex attention mechanisms, such as the position encoding part obtaining separate results for the real and imaginary parts, we will obtain two results here: the real part and the imaginary part of the position similarity score.

[0221]

[0222]

[0223] Optionally, the real and imaginary parts of the positional similarity score are computed separately to support a complex attention mechanism.

[0224] Step A13: Perform channel prediction based on the positional similarity of the Doppler delay.

[0225] Among them, such as Figure 3c As shown, after obtaining the location similarity results based on Doppler delay features, subsequent channel-integrated attention scoring can be performed.

[0226] In this embodiment of the disclosure, the multipath propagation physical model based on MIMO channels can extract Doppler and time delay features based on all channel information, and use the Doppler and time delay information to calculate the location similarity. This can accurately characterize the similarity between channels that are far apart based on the Doppler and time delay information of the channels, thereby improving prediction performance.

[0227] In the embodiments disclosed herein, such as Figure 8As shown, a Channel Attention Mechanism (CAM) is adopted. To better characterize the attention information between channels, MIMO channel characteristics are incorporated into channel position similarity and channel dot product similarity. In this process, the self-attention layer can be used to obtain the channel attention score in the following formula (1), which is a combination of channel position similarity and channel dot product similarity. The similarity between the two can be calculated by the following formula (1):

[0228]

[0229] Here, q and k come from the single-polarization channel response.

[0230] For channel dot product similarity, the features of the dual-polarized antenna can be incorporated into the dot product of q and k to simplify the calculation. For channel position similarity, the kernel function of the channel position can be derived from the MIMO multipath channel (MPC) model, and the channel position similarity can be expressed as the following formula (2):

[0231]

[0232] Based on the kernel function, the multipath characteristics of the channel can be incorporated into the attention mechanism.

[0233] Optionally, antenna polarization characteristics are also taken into consideration in the embodiments of this disclosure. Therefore, when performing channel integration attention scoring, channel similarity can be further combined for calculation.

[0234] Optionally, the channel prediction based on the positional similarity of the Doppler delay in step A13 includes steps A131-A132:

[0235] Step A131: Determine the channel similarity between channels based on the first channel matrix corresponding to the transmit antenna of the first node.

[0236] The first node can be a transmitter, such as a base station. Based on its corresponding transmitting antenna, a first channel matrix (channel response matrix in the antenna domain) H can be constructed, with dimension N. c ×N tx N c It is the number of carriers, N tx It refers to the number of base station transmitting antennas.

[0237] Optionally, after linear transformation, matrix H yields query and key values. These query and key values ​​can then be used to calculate channel similarity independent of location information, such as... Figure 6 The calculation method is based on all antennas.

[0238] Optionally, step A131, based on the first channel matrix corresponding to the transmit antenna of the first node, determines the channel similarity between channels, including steps C1-C2:

[0239] Step C1: Obtain the second channel matrix corresponding to the channel information of a portion of the transmitting antennas from the first channel matrix.

[0240] Step C2: Perform a linear transformation based on the second channel matrix to obtain the query value and the key value.

[0241] Step C3: Determine the channel similarity between channels based on the query value and the key value.

[0242] In the AI ​​model provided in this disclosure, the mapping from state information to channel state similarity can be realized by constructing the intrinsic relationship between relative position information and channel response similarity through the state kernel function. The characterization of channel similarity no longer depends solely on the dot product attention between channel responses. Therefore, only a portion of the channel information can be used for dot product calculation without affecting the final attention score, which can also effectively reduce the amount of computation and computational complexity.

[0243] like Figure 6 and Figure 7 As shown, channel calculations can be performed using only a portion of the antennas, i.e., using only N. c ×αN tx The calculation of H, α∈(0,1], reduces computational complexity. One method for selecting a subset of antennas is to start by selecting αN from the first antenna. tx Root antenna, spacing αN tx Root antenna, randomly select αN tx The selected channel matrix, including the root antenna, is denoted as H′.

[0244] H′ is transformed by the linear matrix W k W q W v Get query key Sum The calculation is as follows:

[0245] Query: K = H′W k Key: Q = H′W q Value: V = H′W v

[0246] Obtain low-complexity channel dot product results and calculate channel similarity k independent of location information. exp (h q ,h k )

[0247] <q,k> = <h′W′q ,h′W′ k >

[0248]

[0249] Step A132: Perform channel prediction based on the location similarity and the channel similarity.

[0250] In the integrated channel attention mechanism, the integrated channel similarity can be obtained by merging the position similarity result and the channel similarity, i.e., by multiplying the two.

[0251] The comprehensive channel similarity is normalized to obtain an attention score. Similar to traditional attention mechanisms, this score will be used as a weight for each value v(x). k )=x k W k The hidden layer is weighted and stacked, and then the result of the hidden layer is output. The result of the hidden layer will be input to the next attention mechanism layer or the next decoding layer.

[0252] In this embodiment of the disclosure, complex attention mechanisms can be used for processing.

[0253] In this embodiment of the disclosure, channel location similarity information is introduced into the attention scoring scheme, which reduces the dependence on channel dot product similarity in the attention scoring calculation. Therefore, only channels on some antennas can be used to calculate the channel dot product, which can effectively reduce computational complexity and improve prediction performance.

[0254] In one feasible embodiment, such as Figure 3c and Figure 8 As shown, after merging the time-frequency position similarity (position similarity) and the channel dot product similarity (channel similarity), the channel comprehensive attention score of the self-attention layer can be obtained. After performing operations such as SoftMax, Add and normalization (Norm), the hidden state can be output and input into the subsequent cross-attention layer.

[0255] In the channel cross-attention layer processing, the base station can input the compressed channel reported by the UE and the encoding result of the two-dimensional time and frequency domain position corresponding to the compressed channel (refer to the encoding process of the target position information shown in the above embodiment). Optionally, as Figure 3cAs shown, the input to the channel cross-attention layer can be formed by combining the channel information 2 reported by the UE and the corresponding location information 2 with the hidden layer information output from the channel self-attention layer. The information reported by the UE is used to generate K and V in the attention mechanism, and the hidden information output from the self-attention layer is used to generate Q. The information from the three transform domains is processed by the channel cross-attention layer, and the output is new hidden information that is then input to the next Transformer decoder layer.

[0256] Optionally, the operation flow of the channel cross-attention layer is similar to that of the channel self-attention layer, both consisting of steps such as calculating the channel comprehensive attention score, SoftMax, and Add&Norm. The difference lies in that the input of the channel self-attention layer is channel information 1 and location information 1, while the input of the channel cross-attention layer is channel information 2, location information 2, and the output of the previous channel self-attention layer.

[0257] The information output from the channel cross-attention layer is used as the input to the channel self-attention layer in the next Transformer decoder layer. This process is repeated until N Transformer decoder layers have been passed, at which point the predicted channel result is finally output.

[0258] Optionally, the input to the channel cross-attention layer can also be the channel information decompressed by the base station.

[0259] In the solutions provided in this disclosure, on the one hand, the provided CAM-UniCSI adopts a single-level architecture, directly predicting future CSI based on Cumulative Compressed CSI (CCSI), thereby reducing model complexity and processing latency. On the other hand, by integrating complex time-frequency spatial variation functions into the calculation of attention scores, the provided CAM can better capture the correlation between channel and time-frequency, thus achieving more accurate channel prediction. Furthermore, by estimating the Doppler and delay distribution ranges of all possible communication scenarios, the proposed AI model can adapt to different scenarios, possessing good generative capabilities and improving the model's generalization performance. The CAM-UniCSI provided in this disclosure can solve the practical problems of AI-based CSI feedback and prediction, realizing the combination of the underlying mechanism of the AI ​​model and communication characteristics, improving CSI feedback and prediction performance, and enhancing the deployment possibility in large-scale MIMO systems. Specifically, by using single-level processing and integrating channel characteristics and communication scenario information into the attention mechanism and parameter adjustment of the AI ​​model, a solution with improved throughput, low complexity, high performance, and strong generalization is provided, making the solution easy to use commercially.

[0260] This disclosure also provides an electronic device including a processor, and optionally, a transceiver and / or memory coupled to the processor, the processor being configured to perform the steps of the method provided in any optional embodiment of this disclosure.

[0261] Figure 10 The diagram shows a structural schematic of an electronic device to which an embodiment of the present invention applies, such as... Figure 10 As shown, Figure 10 The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of this disclosure. Optionally, the electronic device may be at least a first node and a second node.

[0262] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with this disclosure. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0263] Bus 4002 may include a pathway for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 10 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0264] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing computer programs and capable of being read by a computer, without limitation herein.

[0265] The memory 4003 is used to store computer programs that execute embodiments of the present disclosure, and is controlled by the processor 4001 to execute them. The processor 4001 is used to execute the computer programs stored in the memory 4003 to implement the steps shown in the foregoing method embodiments.

[0266] This disclosure provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0267] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0268] The terms “first,” “second,” “third,” “fourth,” “1,” “2,” etc. (if present) in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in a sequence other than that shown in the figures or text.

[0269] It should be understood that although arrows indicate various operation steps in the flowcharts of the embodiments of this disclosure, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of the embodiments of this disclosure, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured as required, and the embodiments of this disclosure do not limit this.

[0270] The above text and accompanying drawings are provided as examples only to help the reader understand this disclosure. They are not intended and should not be construed as limiting the scope of this disclosure in any way. Although certain embodiments and examples have been provided, it will be apparent to those skilled in the art, based on the content disclosed herein, that changes can be made to the illustrated embodiments and examples, and other similar implementations based on the technical concept of this disclosure can be adopted without departing from the scope of this disclosure, and these modifications and modifications are also within the protection scope of the embodiments of this disclosure.

Claims

1. A method executed by a first node in a wireless communication system, characterized in that, include: The target location information is determined by using the first artificial intelligence (AI) network based on the time-frequency location information of the channel. Channel prediction is performed using a second AI network based on channel information from multiple historical moments and the target location information. The target location information is used to characterize the similarity between channels and the correlation between the time-frequency location information of the corresponding channels.

2. The method according to claim 1, characterized in that, The channel prediction based on channel information from multiple historical moments and the target location information includes: Channel prediction is performed based on the first channel information and the target location information to obtain a first prediction result; Based on the first prediction result, the second channel information, and the target location information, channel prediction is performed to obtain the second prediction result; Wherein, the first channel information includes channel information of multiple historical moments after decompression; and / or, the second channel information includes at least one of the following: channel information of multiple historical moments to be decompressed, channel information of multiple historical moments after decompression, or channel information based on the precoding index matrix of the first type codebook or the second type codebook.

3. The method according to claim 1, characterized in that, The determination of the target location information includes: Determine the relative position information between the time position and frequency position corresponding to the channel information at the historical moment; Based on the channel distribution information of the preset scenario, a grid is divided to obtain a two-dimensional grid related to Doppler delay; Based on the two-dimensional grid, the relative position information is mapped to obtain the target position information.

4. The method according to claim 3, characterized in that, The channel distribution information based on the preset scenario is used to perform grid division to obtain a two-dimensional grid related to Doppler delay, including: Based on multiple first change information of the channel changing with time and multiple second change information of the channel changing with frequency under a preset scenario, a two-dimensional grid is divided, and each grid corresponds to one first change information and one second change information. Each of the first change information is related to the Doppler factor corresponding to each channel path; each of the second change information is related to the time delay factor corresponding to each channel path.

5. The method according to claim 3, characterized in that, The process of mapping the relative position information based on the two-dimensional grid to obtain the target position information includes: Based on the two-dimensional grid, the relative position information corresponding to different historical moments and the channel-related information between the channels are determined; Based on the channel-related information, the target location information is determined.

6. The method according to claim 5, characterized in that, The determination of channel-related information between relative position information and the channel at different historical moments based on the two-dimensional grid includes: Based on the two-dimensional grid, the similarity between the real part of the channel correlation coefficient between the relative position information at different historical moments and the channel is used to determine the first position coding information; The determination of target location information based on the channel-related information includes: Based on the first location encoding information, the target location information is determined.

7. The method according to claim 6, characterized in that, The step of determining the relative position information and channel-related information between the channel and the channel corresponding to different historical moments based on the two-dimensional grid also includes: The second position coding information is determined based on the similarity of the imaginary parts in the channel correlation coefficient. The determination of target location information based on the channel-related information includes: The target location information is determined based on the first location encoding information and the second location encoding information.

8. The method according to any one of claims 2-7, characterized in that, The channel prediction based on the first channel information and the target location information includes: Based on the first channel information, a first feature related to time delay and a second feature related to Doppler are extracted; Based on the target location information, the first feature, and the second feature, the positional similarity of the Doppler time delay is determined; Channel prediction is performed based on the positional similarity of the Doppler delay.

9. The method according to claim 8, characterized in that, The channel prediction based on the positional similarity of the Doppler delay includes: Based on the first channel matrix corresponding to the transmit antenna of the first node, the channel similarity between channels is determined; Channel prediction is performed based on the location similarity and the channel similarity.

10. The method according to claim 9, characterized in that, Determining the channel similarity between channels based on the first channel matrix corresponding to the transmitting antenna of the first node includes: Obtain a second channel matrix from the first channel matrix that corresponds to the channel information of a portion of the transmitting antennas; A linear transformation is performed based on the second channel matrix to obtain the query value and the key value; Based on the query value and the key value, the channel similarity between channels is determined.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 10.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 10.

13. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 10.

Images