WO2024088134A1

WO2024088134A1 - Channel prediction method and apparatus

Info

Publication number: WO2024088134A1
Application number: PCT/CN2023/125215
Authority: WO
Inventors: 赵喆; 李增; 赵嘉怡
Original assignee: 中兴通讯股份有限公司
Priority date: 2022-10-27
Filing date: 2023-10-18
Publication date: 2024-05-02
Also published as: CN115396055A; CN115396055B

Abstract

Disclosed in the embodiments of the present application are a channel prediction method and apparatus. The method comprises: acquiring first channel estimation values corresponding to m adjacent historical SRS periods, wherein m is a model order of a channel prediction model based on kernel recursive least squares; according to the m first channel estimation values, determining first sample data for the channel prediction model to predict a channel within the current SRS period, wherein the first sample data comprises m first channel estimation values; inputting the first sample data into the channel prediction model, updating a kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, and updating model parameters of the channel prediction model; and on the basis of the updated kernel dictionary and the updated model parameters, predicting the channel within the current SRS period by means of the channel prediction model.

Description

Channel prediction method and device

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims priority to a Chinese patent application filed with the Chinese Patent Office on October 27, 2022, with application number 202211327001.8 and invention name “Channel Prediction Method and Device”. The entire contents of the Chinese patent application are incorporated herein by reference.

Technical Field

The present invention relates to the field of communication technology, and in particular to a channel prediction method and device.

Background technique

Multi-antenna technology can make full use of spatial dimension resources and exponentially increase the transmission capacity of wireless communication systems without increasing transmission power and bandwidth. At the same time, beamforming technology is widely used because it can compensate for signal fading and distortion introduced by spatial loss and multipath effects during wireless propagation and reduce interference between users on the same channel. Under the TDD (Time Division Duplexing) standard, users send SRS (Sounding Reference Signal), and the base station obtains uplink CSI (Channel State Information) through channel estimation. Due to the channel reciprocity of the TDD system, the measured uplink channel CSI can be directly used in the downlink Beamforming design to achieve collaborative signal processing. However, when the terminal is moving at high speed (as shown in Figure 1), the Doppler shift of the point-to-point link becomes larger, the channel coherence time decreases, the channel time-variability is severe, and the measured uplink channel CSI is outdated and cannot represent the actual channel state within the 33SRS period. As a result, the downlink Beamforming designed based on the estimated CSI is mismatched with the actual channel, resulting in performance degradation.

In order to overcome the performance degradation caused by the drastic time-varying channel, in some cases, the refresh frequency of the channel state information is increased by reducing the system's SRS period. However, in high-speed scenarios, the channel coherence time is much smaller than the current application value of the SRS period, and excessive reduction of the SRS period will occupy too many time and frequency resources, thereby limiting system performance and is not feasible.

To solve the above problems, a feasible way is to predict the future channel state based on the known channel estimation value, and design a beamforming algorithm based on the predicted value of the channel to match the time-varying downlink channel. Commonly used channel prediction algorithms are: 1) Channel prediction method based on radio parameters; 2) Prediction method based on autoregressive model; 3) Channel prediction method based on neural network. Among them, the channel prediction method based on radio parameters assumes a relatively ideal channel model, and the static parameters are assumed to remain unchanged during the prediction time. However, the effective time of the static parameters is inversely proportional to the terminal movement speed, making the channel prediction based on radio parameters difficult to apply in high-speed scenarios. The prediction method based on the autoregressive model has the advantages of low complexity, but the prediction performance of channels with nonlinear time characteristics is limited. Channel prediction algorithms based on neural networks are often more complex. It can be seen that there is an urgent need to provide a channel prediction method that can be applied to scenarios where terminals move at high speed.

Summary of the invention

The purpose of the embodiments of the present application is to provide a channel prediction method and device to solve the problem that the channel cannot be accurately predicted in a terminal high-speed movement scenario.

On the one hand, an embodiment of the present application provides a channel prediction method, including: obtaining first channel estimation values corresponding to m adjacent historical SRS periods; wherein m is the model order of the channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1; based on the m first channel estimation values, determining the first sample data for predicting the channel in this SRS period by the channel prediction model; the first sample data includes m first channel estimation values; inputting the first sample data into the channel prediction model, updating the kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, and updating the model parameters of the channel prediction model; wherein the updated kernel dictionary includes the first sample data; based on the updated kernel dictionary and model parameters, and through the channel prediction model, predicting the channel in the current SRS period, obtaining a second channel estimation value.

On the other hand, an embodiment of the present application provides a channel prediction device, comprising: a first acquisition module, used to obtain a first channel estimation value corresponding to m adjacent historical SRS cycles; wherein m is based on The model order of the channel prediction model of kernel recursive least squares, and m is an integer greater than or equal to 1; a first determination module, used to determine the first sample data for predicting the channel in this SRS period by the channel prediction model based on the m first channel estimation values; the first sample data includes m first channel estimation values; a first updating module, used to input the first sample data into the channel prediction model, update the kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, and update the model parameters of the channel prediction model; wherein the updated kernel dictionary includes the first sample data; a prediction module, used to predict the channel in the current SRS period based on the updated kernel dictionary and model parameters and through the channel prediction model to obtain a second channel estimation value.

On the other hand, an embodiment of the present application provides a channel prediction device, including a processor and a memory electrically connected to the processor, the memory storing a computer program, and the processor being used to call and execute the computer program from the memory to implement the above-mentioned channel prediction method.

On the other hand, an embodiment of the present application provides a storage medium for storing a computer program, where the computer program can be executed by a processor to implement the above-mentioned channel prediction method.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in one or more embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

FIG1 is a schematic scene diagram of a channel prediction method according to an embodiment of this specification;

FIG2 is a schematic flow chart of a channel prediction method according to an embodiment of this specification;

FIG3 is a schematic flow chart of a channel prediction method according to another embodiment of the present specification;

FIG4 is a schematic effect diagram of a channel prediction method according to an embodiment of this specification;

FIG5 is a schematic effect diagram of a channel prediction method according to another embodiment of the present specification;

FIG6 is a schematic block diagram of a channel prediction device according to an embodiment of this specification;

FIG. 7 is a schematic block diagram of a channel prediction device according to an embodiment of this specification.

Detailed ways

The embodiments of the present application provide a channel prediction method and device to solve the problem that the channel cannot be accurately predicted in a terminal high-speed movement scenario.

In order to enable those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

FIG2 is a schematic flow chart of a channel prediction method according to an embodiment of the present application. As shown in FIG2 , the method includes the following steps.

S202, obtaining first channel estimation values corresponding to m adjacent historical SRS periods; wherein m is the model order of the channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1.

The m adjacent historical SRS cycles usually select the m historical SRS cycles closest to the current cycle. The first channel estimation value corresponding to the historical SRS cycle refers to the channel estimation value corresponding to the historical SRS cycle and obtained through accurate measurement.

S204, determining first sample data for predicting the channel in this SRS cycle using a channel prediction model according to the m first channel estimation values; the first sample data includes the m first channel estimation values.

In one embodiment, m first channel estimation values are combined to obtain the first sample data of the channel prediction model. For example, when performing the i-th channel prediction, the first sample data can be expressed as u(i)=[H _i H _i+1 ...H _i+m-1 ], where H _i represents the i-th channel estimation value.

S206, inputting the first sample data into the channel prediction model, updating the core dictionary of the channel prediction model according to the first sample data and the number of elements in the core dictionary of the channel prediction model, and updating the core dictionary of the channel prediction model. The updated kernel dictionary includes the first sample data.

In one embodiment, when updating the kernel dictionary of the channel prediction model based on the first sample data and the number of elements in the kernel dictionary of the channel prediction model, it is necessary to ensure that the number of elements in the kernel dictionary does not exceed the corresponding preset element number threshold. If it exceeds, when updating the kernel dictionary, it is necessary to delete some sample data in the kernel dictionary so that the number of elements in the updated kernel dictionary does not exceed the preset element number threshold, thereby achieving the effect of controlling the amount of sample data.

S208 , predicting the channel in this SRS period based on the updated kernel dictionary and model parameters and using the channel prediction model to obtain a second channel estimation value.

In this embodiment, the model order m means that the channel estimation value in the next SRS period is related to the channel estimation values in the previous m historical SRS periods, thus forming a set of input-output pairs: Wherein, u(i)=[H _i H _i+1 ……H _i+m-1 ], H _i represents the ith channel estimation value. y(i)=H _i+m , M is the preset element number threshold of the kernel dictionary, indicating the maximum value of the sample data that can be stored in the kernel dictionary.

In this embodiment, before using the channel estimation model for channel prediction, the channel estimation model can be initialized. Specifically, first determine the model attribute information of the channel prediction model, the model attribute information includes the model order, the kernel function and the kernel dictionary; then, obtain the initial channel estimation value of the channel, and initialize the model parameters of the channel prediction model according to the initial channel estimation value, the model parameters include the intermediate matrix, the weight coefficient and the forgetting matrix. Among them, the initial channel estimation value refers to the channel estimation value obtained by measuring the channel for the first time. In this embodiment, any existing kernel function can be selected, such as the Gaussian radial basis kernel function, the linear kernel function or the polynomial kernel function, etc. The linear kernel function is suitable for scenarios where the new and old channels are linearly related, the polynomial kernel function is suitable for scenarios where the new and old channels are multi-power related, and the Gaussian kernel function is suitable for scenarios where the new and old channels are highly nonlinearly related.

Assume that the intermediate matrix Q(i), weight coefficient α(i), forgetting matrix B(i) and kernel dictionary D(i) of the channel prediction model are initialized. Specifically, let i=1, that is, the first channel estimation value (i.e., the initial channel estimation value) is used to predict the second channel estimation value. Then the intermediate matrix Q(1)=[λβ+K(u(1),u(1))] ^-1 , where λ is the regularization parameter introduced to prevent overfitting, β is the forgetting factor, and K(.) represents the kernel function. The weight coefficient α(1)=Q(1)y(1), the forgetting matrix B(1)=[1], and the kernel dictionary D(1)=[u(1)].

The technical solution of the embodiment of the present application is adopted, by obtaining the first channel estimation values corresponding to m adjacent historical SRS periods; wherein m is the model order of the channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1; according to the m first channel estimation values, determining the first sample data of the channel prediction model for predicting the channel in this SRS period; the first sample data includes m first channel estimation values; inputting the first sample data into the channel prediction model, updating the kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, and updating the model parameters of the channel prediction model; wherein the updated kernel dictionary includes the first sample data; based on the updated kernel dictionary and model parameters, and through the channel prediction model, predicting the channel in this SRS period. It can be seen that when predicting the channel estimation value in this SRS cycle based on the historical channel estimation value, the technical solution can make predictions through the channel prediction model based on kernel recursive least squares to ensure that the error between the channel estimation value and the actual channel value is minimized, thereby improving the accuracy of the channel prediction; and, by updating the dictionary of the channel prediction model based on the number of elements in the kernel dictionary, the amount of sample data in the kernel dictionary can be effectively controlled, avoiding the problem of increasing the amount of signal processing calculations that continue to arrive when the amount of sample data continues to increase, thereby greatly reducing the amount of channel prediction calculations and improving the real-time performance of channel prediction. Therefore, the technical solution can accurately predict nonlinear time-varying channels even in scenarios where the terminal is moving at high speed through efficient and real-time channel prediction.

In one embodiment, the kernel dictionary of the channel prediction model is updated according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, which may be specifically performed as the following actions.

Action A1: if the number of elements in the kernel dictionary is less than a preset element number threshold of the kernel dictionary, update the kernel dictionary according to the correlation between the first sample data and each element in the kernel dictionary.

Action A2: if the number of elements in the kernel dictionary is greater than or equal to a preset element number threshold, update the kernel dictionary according to the first sample data and the kernel function value of each element in the kernel dictionary.

Assume that the number of elements in the kernel dictionary is denoted by L, where L is an integer greater than or equal to 1.

In the above action A1, the correlation between the first sample data and each element in the kernel dictionary is first calculated to obtain L correlations. If the maximum correlation is less than or equal to the preset correlation threshold, the element corresponding to the maximum correlation in the nuclear dictionary is deleted, and the first sample data is added as a new element to the nuclear dictionary to obtain an updated nuclear dictionary; if the maximum correlation is less than or equal to the preset correlation threshold, the first sample data is added as a new element to the nuclear dictionary to obtain an updated nuclear dictionary.

In one embodiment, the correlation can be characterized by using projection angle cosine, correlation coefficient, etc. For example, the projection angle cosine represented by the following formula (1) is used to characterize the correlation.

Among them, l(u(i)) and l(u(j)) represent the vectors composed of u(i) and u(j) respectively substituted into the kernel function K(u(i),u(n)), n≠i,j with other samples. The projection angle cosine depends on the kernel function used and the elements in the kernel dictionary. If the maximum projection angle cosine (i.e., the maximum correlation value) is greater than the preset correlation threshold τ, that is, the condition represented by the following formula (2) is satisfied, then the element with the maximum projection angle cosine is deleted from the kernel dictionary, and the first sample data u(i) is added to the kernel dictionary. The updated kernel dictionary is D(i)=[D(i-1)\u(j)u(i)], where "\" means deletion, i.e., the element u(j) is deleted. If the maximum projection angle cosine is less than or equal to the preset correlation threshold τ, that is, the condition represented by the following formula (3) is satisfied, then the first sample data u(i) is directly added to the kernel dictionary. The updated kernel dictionary is D(i)=[D(i-1)u(i)].

In the above action A2, first, according to the kernel function of the channel prediction model, the kernel function values of the first sample data and each element in the kernel dictionary are calculated to obtain L kernel function values. Secondly, if there is at least one kernel function value greater than the preset kernel function threshold value among the L kernel function values, the elements whose kernel function values are greater than the preset kernel function threshold value and meet the preset deletion condition are deleted from the kernel dictionary, and the first sample data is added as a new element to the kernel dictionary to obtain an updated kernel dictionary. If each kernel function value is less than or equal to the preset kernel function threshold value, the element corresponding to the maximum kernel function value is deleted from the kernel dictionary, and the first sample data is added as a new element to the kernel dictionary to obtain an updated kernel dictionary.

Let u(i) represent the first sample data, and u(j) represent the element in the kernel dictionary. Then the first sample data The kernel function value of each element in the kernel dictionary can be expressed as: K(u(i),u(j)). Assuming that the preset kernel function threshold is ε, if there is u(j) such that K(u(i),u(j))>ε, then the elements that satisfy the expression (i.e., the kernel function value is greater than the preset kernel function threshold) and the preset deletion condition are deleted from the kernel dictionary. If there is no u(j) such that K(u(i),u(j))>ε, then the element corresponding to the maximum kernel function value is deleted from the kernel dictionary.

Specifically, when there exists u(j) such that K(u(i),u(j))>ε, elements whose kernel function values are greater than a preset kernel function threshold and meet the preset deletion conditions are deleted from the kernel dictionary, which can be performed as the following actions B1-B3.

Action B1, for the candidate elements whose kernel function values are greater than a preset kernel function threshold, the first sample data is used as the kernel center, and a weighted average of the output values corresponding to each candidate element is calculated.

In the above-mentioned embodiment, the model order m of the channel prediction model means that the channel estimation value in the next SRS cycle is considered to be related to the channel estimation values in the previous m historical SRS cycles, thus forming a set of input-output pairs: Among them, y(i) is the output value corresponding to u(i).

In action B1, all elements whose kernel function values are greater than the preset kernel function threshold are taken as candidate elements, and all candidate elements constitute a candidate element set. Assuming that the candidate element set is S = [u(1), u(2) ... u(n)], the weighted average of the output values corresponding to each candidate element can be calculated according to the following formula (4) with the first sample data u(i) as the kernel center.

Wherein, n is the number of elements to be selected in the set of elements to be selected.

Action B2, determining the difference between the output value corresponding to each candidate element and the weighted average value, and calculating the product of the maximum difference and the minimum difference among the multiple differences.

The difference between the output value y(j) corresponding to the candidate element and the weighted average value can be expressed as: e _j =|y(j)-y|. By calculating the difference corresponding to each candidate element, the maximum difference e _max and the minimum difference e _min among the multiple differences are obtained. The product of the maximum difference e _max and the minimum difference e _min can be expressed as: e _max e _min .

Action B3, if the product of the maximum difference and the minimum difference is greater than the preset threshold, the candidate element corresponding to the maximum difference is deleted from the core dictionary; if the product of the maximum difference and the minimum difference is less than or equal to the preset threshold, the candidate element corresponding to the minimum difference is deleted from the core dictionary.

by Represents the preset threshold. If the product Then delete the candidate element corresponding to e _max from the kernel dictionary, thereby deleting the sample data that may appear with a small probability. Then delete the candidate element corresponding to e _min from the kernel dictionary, thereby deleting the sample data with the same information as the first sample data u(i). Assuming that the element deleted from the kernel dictionary is u(j), the updated kernel dictionary is: D(i) = [D(i-1)\u(j)u(i)]. "\" means deletion, that is, deleting the element u(j).

In this embodiment, based on the difference between the number of elements in the kernel dictionary and the preset element number threshold, the kernel dictionary is updated in a corresponding kernel dictionary updating manner, so that the elements in the kernel dictionary (i.e., the number of samples) never exceed the preset element number threshold. This effectively avoids the problem of increasing the amount of calculation for processing the continuously arriving signals when the amount of sample data continues to increase, greatly reduces the amount of calculation for channel prediction, and improves the real-time performance of channel prediction.

In one embodiment, after the core dictionary of the channel prediction model is updated, the model parameters of the channel prediction model are updated. The model parameters include at least one of an intermediate matrix, an initial coefficient, and a forgetting matrix. Specifically, after the core dictionary of the channel prediction model is updated, it is determined whether the number of elements in the updated core dictionary increases. If so, the intermediate matrix, the weight coefficient, and the forgetting matrix are updated; if not, the intermediate matrix and the weight coefficient are updated, while the forgetting matrix remains unchanged.

In this embodiment, if the number of elements in the updated kernel dictionary increases, the intermediate matrix, weight coefficient and forgetting matrix need to be updated. The specific updating process is as follows.

First, calculate the following formula (5).

q(i)＝[K(u(1),u(i))K(u(2),u(i))……K(u(L),u(i))] ^T (5)

Among them, u(1), u(2)...u(L) are elements in the kernel dictionary, and L is the length of the kernel dictionary, that is, the number of elements in the kernel dictionary.

Secondly, update the forgetting matrix B(i) according to the following formula (6).

Update the intermediate matrix Q(i) according to the following formula (7).

In which, z ₁ (i) = Q (i-1) B (i-1) q (i); r (i) = λ β + K (u (i), u (i)) - β ^-1 z ₁ (i) ^T q (i); z (i) = Q (i-1) q (i).

Update the weight coefficient α(i) according to the following formula (8).

Where, e(i) = y(i) - q(i-1) ^T α(i-1).

If the number of elements in the updated kernel dictionary has not increased, the intermediate matrix and weight coefficients need to be updated. Let T(i) represent the kernel matrix, then the intermediate matrix Q(i-1) is the inverse matrix of the previous kernel matrix T(i-1). Since the number of elements in the kernel dictionary has not changed, but the elements in the kernel dictionary have been updated, the kernel function information of the deleted element u(j) needs to be deleted as well. Specifically, the kernel function information related to u(j) in T(i-1) is located in the jth row and jth column. Through row transformation and column transformation, the jth row and jth column can be moved to the first row and first column, and we get but The inverse matrix of It can be obtained by doing corresponding column transformation and row transformation.

By deleting The first row and first column of Then use the block matrix inversion Specifically, it can be calculated according to the following formula group (9).

According to the above formula group (9), the updated intermediate matrix Q(i) can be calculated as follows.

in, g＝(1+λ-b ^T -A ^-1 b) ^-1 .

The updated weight coefficient α(i) is: α(i) = Q(i)Y(i), where Y(i)＝[y(1), y(2)……y(i)] ^T .

After updating the model parameters of the channel prediction model, based on the updated kernel dictionary and model parameters, the channel in this SRS period is predicted through the channel prediction model to obtain a second channel estimation value. The second channel estimation value can be expressed as:

In one embodiment, based on the updated kernel dictionary and model parameters, and after predicting the channel in this SRS period through the channel prediction model and obtaining the second channel estimation value, steps S210-S216 as shown in FIG. 3 may also be performed.

S210 , determining a first historical prediction error corresponding to a most recent historical SRS period according to a first channel estimation value corresponding to a most recent historical SRS period.

S212, updating the first historical prediction error group corresponding to the channel prediction model according to the first historical prediction error to obtain a second historical prediction error group; the second historical prediction error group includes the first historical prediction error.

S214: Determine the current prediction error of the channel prediction model according to the second historical prediction error group and by using an error estimation model based on Gaussian process regression.

S216: Perform error compensation on the second channel estimation value according to the current prediction error to obtain a channel correction value corresponding to the second channel estimation value.

Specifically, it is assumed that the model order of the error estimation model is n, and the window size of the error samples in the error sample group is _Me . According to the first channel estimation value corresponding to the most recent historical SRS period, the first historical prediction error corresponding to the most recent historical SRS period is determined to be expressed as e _i , and a new error sample E(i)=[e _i-n+1 … e _i-1 e _i ]. If the space size of the first historical prediction error group F(i-1) is already _Me , the new error sample E(i) is added to the first historical prediction error group F(i-1), and the error sample with the longest retention time in the historical prediction error group F(i-1) is deleted, and the updated second historical prediction error group is obtained: F(i)=[E(iM _e +1) … E(i-1)E(i)]. If the space size of the first historical prediction error group F(i-1) is smaller than _Me , the new error sample E(i) is directly added to the first historical prediction error group F(i-1) to obtain the updated second historical prediction error group: F(i)=[E(i-1)E(i)].

Then, the current prediction error of the channel prediction model is determined according to the second historical prediction error group. Specifically, the kernel function K(.) is selected, and the covariance matrix P of the error vector is calculated. The element P _i,j in the i-th row and j-th column of the matrix P is ＝K(E(i),E(j)). The vector q _i is calculated as [K(E(i),E(iM _e +1))K(E(i),E(iM _e ))……K(E(i),E(i))]. Then, using the Bayesian posterior probability, the prediction error can be calculated as:

Then, according to the current prediction error, the second channel estimation value is error compensated in the ratio 0<θ<1, and the channel correction value corresponding to the second channel estimation value can be expressed as: θ is the error correction factor.

In one embodiment, after performing error compensation on the second channel estimation value according to the current prediction error to obtain a channel correction value corresponding to the second channel estimation value, steps S218-S220 as shown in FIG. 3 may also be performed.

S218, performing Wiener filtering according to the first channel estimation value and the channel correction value corresponding to the most recent historical SRS period to obtain a filtering result.

The most recent historical SRS cycle refers to the historical SRS cycle closest to the current cycle among the m adjacent historical SRS cycles. The Wiener filter can use a second-order Wiener filter or a higher-order Wiener filter. The difference is that the performance of using a higher-order Wiener filter is relatively higher, but the amount of calculation and resource consumption are higher.

S220, determining a target channel estimation value in each time slot in the current SRS cycle according to the filtering result, wherein the target channel estimation value includes channel estimation values of different time slots in each time slot in the current SRS cycle.

Specifically, S220 can be executed as the following actions C1-C3.

Action C1, determining the autocorrelation information of the channel in the time dimension according to the large-scale information of the channel.

Assume that based on the large-scale information of the channel, the autocorrelation estimate of the channel in the time dimension is determined as follows: Formula (10).

Wherein, l represents the number of averages, T is the length of an SRS cycle, R is the time correlation function of the channel, and the above autocorrelation estimate is modeled as the following zero-order Bessel function.
R(t)＝J ₀ (2πf _dmax t)

Wherein, f _dmax is the maximum Doppler frequency shift, which can be estimated using the zero-crossing point of the zero-order Bessel function.

Action C2: determining the interpolation weight corresponding to each time slot in this SRS cycle according to the channel autocorrelation information in the time dimension and a preset Wiener filter function.

Action C3, determining a channel estimation value for each time slot in this SRS period according to the second channel estimation value, the channel correction value and the interpolation weight corresponding to each time slot.

Assuming that the autocorrelation information of the channel in the time dimension is the time correlation function represented by the above formula (10), and the preset Wiener filter function is a second-order Wiener filter function, the interpolation weight of the pth time slot in this SRS cycle is expressed as:

in, N is the total number of time slots in one SRS cycle, and Δt is the time length of one time slot.

Based on the above analysis, it can be determined that the channel estimation value of the p-th time slot in this SRS cycle is.

Taking the urban road in the scene shown in Figure 1 as an example, assuming that the moving speed is 20km/h. In this scene, the channel prediction method provided by the present application is performed as follows.

First, the model order m of the channel prediction model is determined to be 4, the preset element number threshold M of the kernel dictionary is 4, the kernel function is selected as the Gaussian radial basis kernel function, the model order n of the error estimation model is determined to be 4, and the window size _Me of the error sample is set to 4.

Then, the model parameters of the initial channel prediction model include the following actions: setting the preset correlation threshold τ to 0.8, the preset kernel function threshold ε to 0.5, and the preset threshold corresponding to the difference product = 0.5, the error correction factor θ is 0.8, the regularization parameter λ is 0.8, and the forgetting factor β is 0.9. Let i = 1, input the first set of channel estimation values u(1) as samples into the channel prediction model, and initialize the intermediate matrix Q(1) = [λβ+K(u(1),u(1))] ^-1 , initialize weight coefficient α(1)=Q(1)y(1), forgetting matrix B(1)=[1], kernel dictionary D(1)=[u(1)].

Then, the above S204 is executed as follows.

When i=3, the new channel estimation value is used to form a new sample u(3) (i.e., the first sample data) and input into the channel prediction model, and the channel prediction model updates the kernel dictionary. The number of kernel dictionaries is less than the preset element number threshold 4 of the kernel dictionary, and the above action A1 is performed to calculate the kernel vector projection angle cosine of u(3) and each element in the kernel dictionary to obtain 0.23, which is less than or equal to the preset correlation threshold 0.8, so the new sample u(3) is added to the kernel dictionary.

Then the model parameters of the channel prediction model are updated, and the next channel estimation value is predicted. As the kernel dictionary size increases, the intermediate matrix, weight coefficients and forgetting matrix need to be updated, and then the channel estimation value of the next SRS cycle (i.e., this SRS cycle) is calculated.

Then, based on the historical prediction error, the current prediction error is estimated and compensated using Gaussian process regression. As in the above steps S210-S216, the previous historical prediction error is obtained based on the new channel estimation value (i.e., the first channel estimation value corresponding to the most recent historical SRS cycle) to form a new error sample. Assuming that the space size of the original error sample is smaller than the window size 4, the new error sample is added to the error sample space. The covariance matrix and kernel vector of the error vector are calculated based on the selected kernel function, and the current prediction error is obtained in combination with the new error sample, and then the current prediction error is multiplied by the error correction factor 0.8 to correct the channel estimation value.

Then, a second-order Wiener filter is performed based on a channel correction value obtained by correcting the first channel estimation value corresponding to the most recent historical SRS period to obtain channel predictions for different time slots within the SRS period.

When i=7, the new channel estimation value is used to form a new sample u(7) (i.e., the first sample data) and input into the channel prediction model, and the channel prediction model updates the kernel dictionary. The number of kernel dictionaries is greater than or equal to the preset element number threshold 4 of the kernel dictionary, and the above action A2 is performed to calculate the kernel function of u(7) and each element in the kernel dictionary, where the maximum kernel function value is 0.47. Since there is no kernel function value greater than the preset kernel function threshold 0.5, the dictionary element corresponding to the maximum kernel function value can be directly deleted, and the new sample u(7) is added to the kernel dictionary.

Then the model parameters of the channel prediction model are updated, and the next channel estimation value is predicted. Since the kernel dictionary size remains unchanged, it is necessary to update the intermediate matrix and weight coefficients while keeping the forgetting matrix inconvenient, and then calculate the channel estimation value of the next SRS cycle.

Then, based on the historical prediction error, the current prediction error is estimated and compensated using Gaussian process regression. As in the above steps S210-S216, the previous historical prediction error is obtained based on the new channel estimation value (i.e., the first channel estimation value corresponding to the most recent historical SRS cycle) to form a new error sample. Assuming that the space size of the original error sample is equal to the window size 4, the new error sample is added to the error sample space, and the error sample with the longest retention time in the error sample space is removed. The covariance matrix and kernel vector of the error vector are calculated based on the selected kernel function, and the current prediction error is obtained in combination with the new error sample, and then the current prediction error is multiplied by the error correction factor 0.8 to correct the channel estimation value.

Attached Figures 4 and 5 show the implementation effects of the channel prediction method provided by the present application when the terminal moving speed is 60km/h and 120km/h, respectively. In Attached Figures 4 and 5, the solid line indicates the channel value measured most recently, that is, the channel measurement value within the most recent SRS cycle; the dotted line indicates the channel estimation value predicted by the channel prediction method provided by the present application. It can be seen from the figure that compared with the channel measurement value within the most recent SRS cycle, the channel prediction method provided by the present application can improve the correlation with the real channel.

In summary, specific embodiments of the present subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recorded in the claims can be performed in a different order and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing can be advantageous.

The above is a channel prediction method provided in an embodiment of the present application. Based on the same idea, an embodiment of the present application also provides a channel prediction device.

FIG6 is a schematic block diagram of a channel prediction device according to an embodiment of the present application. As shown in FIG6 , the device includes the following modules.

The first acquisition module 61 is used to obtain first channel estimation values corresponding to m adjacent historical SRS periods; wherein m is the model order of the channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1.

The first determination module 62 is used to determine the first sample data for predicting the channel in this SRS period by the channel prediction model according to the m first channel estimation values; the first sample data includes the m first channel estimation values.

The first updating module 63 is used to input the first sample data into the channel prediction model, update the kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, and update the model parameters of the channel prediction model; wherein the updated kernel dictionary includes the first sample data.

The prediction module 64 is used to predict the channel in the current SRS period based on the updated kernel dictionary and model parameters and through the channel prediction model to obtain a second channel estimation value.

In one embodiment, the first update module 63 includes: a first update unit, which is used to update the kernel dictionary according to the correlation between the first sample data and each element in the kernel dictionary if the number of elements is less than a preset element number threshold of the kernel dictionary; and a second update unit, which is used to update the kernel dictionary according to the kernel function value of the first sample data and each element in the kernel dictionary if the number of elements is greater than or equal to the preset element number threshold.

In one embodiment, the first updating unit is used to: calculate the correlation between the first sample data and each element in the core dictionary to obtain L correlations; wherein L is the number of elements and L is an integer greater than or equal to 1; if the maximum correlation among the L correlations is greater than a preset correlation threshold, delete the element in the core dictionary corresponding to the maximum correlation, and add the first sample data as a new element to the core dictionary to obtain the updated core dictionary; if the maximum correlation is less than or equal to the preset correlation threshold, add the first sample data as a new element to the core dictionary to obtain the updated core dictionary.

In one embodiment, the second updating unit is used to: calculate the kernel function values of the first sample data and each element in the kernel dictionary according to the kernel function of the channel prediction model to obtain L kernel function values; wherein L is the number of elements and L is an integer greater than or equal to 1; if there is at least one kernel function value greater than a preset kernel function threshold among the L kernel function values, then delete the elements whose kernel function values are greater than the preset kernel function threshold and meet the preset deletion condition from the kernel dictionary, and add the first sample data as a new element to the kernel dictionary to obtain the updated kernel dictionary; if each of the kernel function values is less than or equal to the preset kernel function threshold, then delete the element corresponding to the maximum kernel function value from the kernel dictionary, and add the first sample data as a new element to the kernel dictionary to obtain the updated kernel dictionary.

In one embodiment, the second updating unit is used to: for the candidate elements whose kernel function values are greater than the preset kernel function threshold, use the first sample data as the kernel center, and calculate the weighted average of the output values corresponding to each of the candidate elements; determine the difference between the output value corresponding to each of the candidate elements and the weighted average, and calculate the product of the maximum difference and the minimum difference among multiple differences; if the product is greater than the preset threshold, delete the candidate element corresponding to the maximum difference from the kernel dictionary; if the product is less than or equal to the preset threshold, delete the candidate element corresponding to the minimum difference from the kernel dictionary.

In one embodiment, the model parameters include at least one of an intermediate matrix, a weight coefficient, and a forgetting matrix.

The first updating module 63 includes: a judging unit, used to judge whether the number of elements in the updated core dictionary increases; a third updating unit, used to update the intermediate matrix, the weight coefficient and the forgetting matrix if yes; if no, update the intermediate matrix and the weight coefficient.

In one embodiment, the device further includes: a second determination module, which is used to predict the channel in the current SRS period based on the updated kernel dictionary and model parameters and through the channel prediction model, and after obtaining the second channel estimation value, determine the first historical prediction error corresponding to the most recent historical SRS period according to the first channel estimation value corresponding to the most recent historical SRS period; a second updating module, which is used to update the first historical prediction error according to the first historical prediction error. A first historical prediction error group corresponding to the channel prediction model is updated to obtain a second historical prediction error group; the second historical prediction error group includes the first historical prediction error; a third determination module is used to determine the current prediction error of the channel prediction model according to the second historical prediction error group and through an error estimation model based on Gaussian process regression; a compensation module is used to perform error compensation on the second channel estimation value according to the current prediction error to obtain a channel correction value corresponding to the second channel estimation value.

In one embodiment, the device also includes: a filtering module, which is used to compensate the second channel estimation value according to the current prediction error, and after obtaining the channel correction value corresponding to the second channel estimation value, perform Wiener filtering according to the first channel estimation value and the channel correction value corresponding to the most recent historical SRS cycle to obtain a filtering result; a fourth determination module, which is used to determine the target channel estimation value in each time slot of the SRS cycle in the current SRS cycle according to the filtering result; the target channel estimation value includes the channel estimation values of different time slots in each time slot of the SRS cycle in the current SRS cycle.

In one embodiment, the fourth determination module includes: a first determination unit, used to determine the autocorrelation information of the channel in the time dimension based on the large-scale information of the channel; a second determination unit, used to determine the interpolation weight corresponding to each time slot in the current SRS cycle based on the autocorrelation information and a preset Wiener filter function; and a third determination unit, used to determine the channel estimation value of each time slot in the current SRS cycle based on the second channel estimation value, the channel correction value and the interpolation weight corresponding to each time slot.

In one embodiment, the device also includes: a fifth determination module, used to determine the model attribute information of the channel prediction model before obtaining the first channel estimation value corresponding to m adjacent historical SRS periods, the model attribute information including the model order, the kernel function and the kernel dictionary; a second acquisition module, used to obtain the initial channel estimation value of the channel; an initialization module, used to initialize the model parameters of the channel prediction model according to the initial channel estimation value.

The device of the embodiment of the present application is used to obtain first channel estimation values corresponding to m adjacent historical SRS periods; wherein m is the model order of the channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1; according to the m first channel estimation values, determine the first sample data for the channel prediction model to predict the channel in this SRS cycle; the first sample data includes m first channel estimation values; input the first sample data into the channel prediction model, update the kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, and update the model parameters of the channel prediction model; wherein the updated kernel dictionary includes the first sample data; based on the updated kernel dictionary and the model parameters, and through the channel prediction model, predict the channel in this SRS cycle. It can be seen that when the device predicts the channel estimation value in this SRS cycle based on the historical channel estimation value, it can predict through the channel prediction model based on the kernel recursive least squares to ensure that the error between the channel estimation value and the true channel value is minimized, thereby improving the accuracy of the channel prediction; and by updating the dictionary of the channel prediction model based on the number of elements in the kernel dictionary, the amount of sample data in the kernel dictionary can be effectively controlled, avoiding the problem of increasing the amount of sample data causing the amount of calculation for the continuously arriving signal processing, thereby greatly reducing the amount of calculation for the channel prediction and improving the real-time performance of the channel prediction. Therefore, the device can accurately predict the nonlinear time-varying channel even in the scenario of high-speed terminal movement through efficient and real-time channel prediction.

Those skilled in the art should understand that the channel prediction device in Figure 6 can be used to implement the channel prediction method described above, and the detailed description should be similar to the description of the method part above. To avoid redundancy, it will not be repeated here.

Based on the same idea, an embodiment of the present application also provides a channel prediction device, as shown in FIG7 . The channel prediction device may have relatively large differences due to different configurations or performances, and may include one or more processors 701 and a memory 702, and the memory 702 may store one or more application programs or data. Among them, the memory 702 may be a short-term storage or a permanent storage. The application stored in the memory 702 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the channel prediction device. Furthermore, the processor 701 may be configured to communicate with the memory 702 to execute a series of computer executable instructions in the memory 702 on the channel prediction device. The channel prediction device may also include one or more power supplies 703, one or more wired or wireless network interfaces 704, and one or more input and output Interface 705 , one or more keyboards 706 .

Specifically in this embodiment, the channel prediction device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions for the channel prediction device, and is configured to be executed by one or more processors. The one or more programs include the following computer executable instructions: obtaining first channel estimation values corresponding to m adjacent historical SRS periods; wherein m is the model order of the channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1; according to the m obtained The first channel estimation value is used to determine the first sample data used by the channel prediction model to predict the channel in this SRS period; the first sample data includes m first channel estimation values; the first sample data is input into the channel prediction model, and according to the first sample data and the number of elements in the core dictionary of the channel prediction model, the core dictionary of the channel prediction model is updated, and the model parameters of the channel prediction model are updated; wherein the updated core dictionary includes the first sample data; based on the updated core dictionary and model parameters, the channel in this SRS period is predicted through the channel prediction model to obtain a second channel estimation value.

The technical solution of the embodiment of the present application is adopted, by obtaining the first channel estimation value corresponding to m adjacent historical SRS periods; wherein m is the model order of the channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1; according to the m first channel estimation values, the first sample data for the channel prediction model to predict the channel in this SRS period is determined; the first sample data includes m first channel estimation values; the first sample data is input into the channel prediction model, and according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, the kernel dictionary of the channel prediction model is updated, and the model parameters of the channel prediction model are updated; wherein the updated kernel dictionary includes the first sample data; based on the updated kernel dictionary and model parameters, the channel in this SRS period is predicted through the channel prediction model. It can be seen that when the technical solution predicts the channel estimation value in this SRS period based on the historical channel estimation value, it can make a prediction through the channel prediction model based on the kernel recursive least squares, so as to ensure that the error between the channel estimation value and the true channel value is minimized, thereby improving the accuracy of channel prediction. degree; and, by updating the dictionary of the channel prediction model based on the number of elements in the kernel dictionary, the amount of sample data in the kernel dictionary can be effectively controlled, avoiding the problem of increasing the amount of calculation for processing the continuously arriving signals when the amount of sample data continues to increase, thereby greatly reducing the amount of calculation for channel prediction and improving the real-time performance of channel prediction. Therefore, this technical solution can accurately predict nonlinear time-varying channels even in high-speed terminal movement scenarios through efficient and real-time channel prediction.

The embodiment of the present application also proposes a storage medium, which stores one or more computer programs, and the one or more computer programs include instructions. When the instructions are executed by an electronic device including multiple application programs, the electronic device can execute each process of the above-mentioned channel prediction method embodiment, and are specifically used to execute: obtaining first channel estimation values corresponding to m adjacent historical SRS periods; wherein m is the model order of the channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1; according to the m first channel estimation values, determining the first sample data for the channel prediction model to predict the channel in this SRS period; the first sample data includes m first channel estimation values; inputting the first sample data into the channel prediction model, updating the kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, and updating the model parameters of the channel prediction model; wherein the updated kernel dictionary includes the first sample data; based on the updated kernel dictionary and model parameters, and through the channel prediction model, predicting the channel in the current SRS period, to obtain a second channel estimation value.

The technical solution of the embodiment of the present application is adopted, by obtaining first channel estimation values corresponding to m adjacent historical SRS periods; wherein m is the model order of the channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1; according to the m first channel estimation values, determining the first sample data for predicting the channel in this SRS period by the channel prediction model; the first sample data includes m first channel estimation values; inputting the first sample data into the channel prediction model, updating the kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, and updating the model parameters of the channel prediction model; wherein the updated kernel dictionary includes the first sample data; based on the updated kernel dictionary and the model parameters, and through the channel prediction model for predicting the channel in this SRS period It can be seen that when predicting the channel estimation value in this SRS period based on the historical channel estimation value, the technical solution can make predictions through the channel prediction model based on kernel recursive least squares to ensure that the error between the channel estimation value and the true channel value is minimized, thereby improving the accuracy of channel prediction; and, by updating the dictionary of the channel prediction model based on the number of elements in the kernel dictionary, the amount of sample data in the kernel dictionary can be effectively controlled to avoid the problem of increasing the amount of signal processing calculations for the continuously arriving signals when the amount of sample data continues to increase, thereby greatly reducing the amount of channel prediction calculations and improving the real-time performance of channel prediction. Therefore, the technical solution can accurately predict nonlinear time-varying channels even in scenarios where the terminal is moving at high speed through efficient and real-time channel prediction.

The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

For the convenience of description, the above device is described in terms of functions and is divided into various units and described separately. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and/or hardware.

Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

The present application is described with reference to the flowcharts and/or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and/or block in the flowchart and/or block diagram, as well as the combination of the processes and/or blocks in the flowchart and/or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flow chart. A flowchart may include one process or multiple processes and/or a block diagram may include one block or multiple blocks that specify functions.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

In a typical configuration, a computing device includes one or more processors (CPU), input/output interfaces, network interfaces, and memory.

Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to include The term "includes" or "includes" a non-exclusive inclusion, so that a process, method, commodity, or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity, or device. In the absence of more restrictions, an element defined by the phrase "includes a ..." does not exclude the existence of other identical elements in the process, method, commodity, or device including the element.

The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

A channel prediction method, comprising:

Obtaining first channel estimation values corresponding to m adjacent historical SRS periods; wherein m is the model order of the channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1;

Determine, according to the m first channel estimation values, first sample data for predicting the channel in this SRS cycle by the channel prediction model; the first sample data includes the m first channel estimation values;

inputting the first sample data into the channel prediction model, updating the kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, and updating the model parameters of the channel prediction model; wherein the updated kernel dictionary includes the first sample data;

Based on the updated kernel dictionary and model parameters, the channel in the current SRS period is predicted by the channel prediction model to obtain a second channel estimation value.
The method according to claim 1, wherein the updating the kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model comprises:

If the number of elements is less than a preset element number threshold of the kernel dictionary, updating the kernel dictionary according to the correlation between the first sample data and each element in the kernel dictionary;

If the number of elements is greater than or equal to the preset element number threshold, the kernel dictionary is updated according to the first sample data and the kernel function value of each element in the kernel dictionary.
The method according to claim 2, wherein updating the kernel dictionary according to the correlation between the first sample data and each element in the kernel dictionary comprises:

Calculating the correlation between the first sample data and each element in the kernel dictionary to obtain L correlations, where L is the number of elements and L is an integer greater than or equal to 1;

If the maximum correlation among the L correlations is greater than a preset correlation threshold, the element corresponding to the maximum correlation in the core dictionary is deleted, and the first sample data is added as a new element. Add to the kernel dictionary to obtain the updated kernel dictionary;

If the maximum correlation is less than or equal to the preset correlation threshold, the first sample data is added as a new element to the kernel dictionary to obtain the updated kernel dictionary.
The method according to claim 2, wherein updating the kernel dictionary according to the first sample data and the kernel function value of each element in the kernel dictionary comprises:

According to the kernel function of the channel prediction model, calculating the kernel function value of each element in the kernel dictionary and the first sample data to obtain L kernel function values; wherein L is the number of elements and L is an integer greater than or equal to 1;

If at least one kernel function value among the L kernel function values is greater than a preset kernel function threshold, then deleting the elements whose kernel function values are greater than the preset kernel function threshold and satisfy a preset deletion condition from the kernel dictionary, and adding the first sample data as a new element to the kernel dictionary to obtain the updated kernel dictionary;

If each of the kernel function values is less than or equal to the preset kernel function threshold, the element corresponding to the maximum kernel function value is deleted from the kernel dictionary, and the first sample data is added as a new element to the kernel dictionary to obtain the updated kernel dictionary.
The method according to claim 4, wherein the deleting from the kernel dictionary the elements whose kernel function values are greater than the preset kernel function threshold and satisfy the preset deletion condition comprises:

For the candidate elements whose kernel function values are greater than the preset kernel function threshold, taking the first sample data as the kernel center, calculating the weighted average of the output values corresponding to each of the candidate elements;

Determine the difference between the output value corresponding to each of the to-be-selected elements and the weighted average value, and calculate the product of the maximum difference and the minimum difference among the multiple differences;

If the product is greater than a preset threshold, deleting the candidate element corresponding to the maximum difference from the kernel dictionary;

If the product is less than or equal to the preset threshold, the candidate element corresponding to the minimum difference is deleted from the kernel dictionary.
The method according to claim 1, wherein the model parameters include intermediate matrices, weights Heavy coefficient, at least one item in the forgetting matrix;

The updating of the model parameters of the channel prediction model comprises:

Determining whether the number of elements in the updated kernel dictionary increases;

If yes, then update the intermediate matrix, the weight coefficient and the forgetting matrix;

If not, update the intermediate matrix and the weight coefficients.
The method according to claim 1, wherein, after predicting the channel in the current SRS period based on the updated kernel dictionary and model parameters and using the channel prediction model to obtain the second channel estimation value, the method further comprises:

Determining a first historical prediction error corresponding to the most recent historical SRS period according to the first channel estimation value corresponding to the most recent historical SRS period;

According to the first historical prediction error, updating the first historical prediction error group corresponding to the channel prediction model to obtain a second historical prediction error group; the second historical prediction error group includes the first historical prediction error;

Determining the current prediction error of the channel prediction model according to the second historical prediction error group and by using an error estimation model based on Gaussian process regression;

According to the current prediction error, error compensation is performed on the second channel estimation value to obtain a channel correction value corresponding to the second channel estimation value.
The method according to claim 7, wherein after compensating the second channel estimation value according to the current prediction error to obtain a channel correction value corresponding to the second channel estimation value, the method further comprises:

Performing Wiener filtering according to the first channel estimation value and the channel correction value corresponding to the most recent historical SRS period to obtain a filtering result;

According to the filtering result, a target channel estimation value in each time slot of the current SRS cycle is determined; the target channel estimation value includes channel estimation values of different time slots in each time slot of the current SRS cycle.
The method according to claim 8, wherein the filtering result is used to determine the The target channel estimation value in this SRS cycle is described, including:

Determining the autocorrelation information of the channel in the time dimension according to the large-scale information of the channel;

Determine, according to the autocorrelated information and a preset Wiener filter function, an interpolation weight corresponding to each time slot in the current SRS cycle;

The channel estimation value of each time slot in the current SRS cycle is determined according to the second channel estimation value, the channel correction value and the interpolation weight corresponding to each time slot.
The method according to claim 6, wherein before obtaining the first channel estimation values corresponding to m adjacent historical SRS periods, the method further comprises:

Determining model attribute information of the channel prediction model, the model attribute information including the model order, the kernel function and the kernel dictionary;

Obtaining an initial channel estimation value of the channel;

The model parameters of the channel prediction model are initialized according to the initial channel estimation value.
A channel prediction device, comprising:

A first acquisition module, used to acquire first channel estimation values corresponding to m adjacent historical SRS periods; wherein m is a model order of a channel prediction model based on kernel recursive least squares, and m is an integer greater than or equal to 1;

A first determination module is used to determine first sample data for predicting the channel in this SRS period by the channel prediction model according to the m first channel estimation values; the first sample data includes the m first channel estimation values;

a first updating module, configured to input the first sample data into the channel prediction model, update the kernel dictionary of the channel prediction model according to the first sample data and the number of elements in the kernel dictionary of the channel prediction model, and update the model parameters of the channel prediction model; wherein the updated kernel dictionary includes the first sample data;

A prediction module is used to predict the channel in the current SRS period based on the updated kernel dictionary and model parameters and through the channel prediction model to obtain a second channel estimation value.
A channel prediction device comprises a processor and a memory electrically connected to the processor. The memory stores a computer program, and the processor is used to call and execute the computer program from the memory to implement the channel estimation method according to any one of claims 1 to 10.
A storage medium, wherein the storage medium is used to store a computer program, wherein the computer program can be executed by a processor to implement the channel estimation method according to any one of claims 1 to 10.