An estimation method, apparatus, device, and storage medium for XL-MIMO near-field compressed channels.

Abstract

This invention discloses a method, apparatus, device, and storage medium for estimating XL-MIMO near-field compressed channels. The method includes: first, obtaining the channel vector of the near-field compressed channel to be estimated; inputting the channel vector into a constructed channel estimation model, causing the channel estimation model to convert the channel vector into a polar-domain channel vector through a built-in sparse transformation matrix; compressing the polar-domain channel vector into a signal vector through a built-in sensing matrix; adding noise to the signal vector to obtain a received signal vector; inputting the received signal vector into a built-in LAMP layer, causing the LAMP layer to iterate over the received signal vector through a built-in soft thresholding function to calculate the estimated polar-domain channel vector corresponding to the polar-domain channel vector; and converting the estimated polar-domain channel vector into an estimate of the near-field compressed channel to be estimated through a built-in sparse transformation matrix. By implementing this invention, the number of pilot signals required by the system can be reduced, thereby reducing pilot overhead.

Description

Technical Field

[0001] This invention relates to the field of channel estimation technology, and in particular to an estimation method, apparatus, device, and storage medium for XL-MIMO near-field compressed channels. Background Technology

[0002] XL-MIMO technology was developed to meet the stringent requirements of emerging services in 6G communications and is expected to significantly improve performance. Similar to 5G's massive MIMO, the accuracy of channel state information has a significant impact on the wireless communication performance of XL-MIMO technology; therefore, the accuracy of channel estimation is crucial. Existing near-field channel estimation schemes for XL-MIMO systems use polar domains to represent the sparsity of the near-field channel, obtain the transformation matrix through joint angle and range space sampling, and utilize the Orthogonal Matching Pursuit (OMP) algorithm from the CS-type methods for near-field channel estimation.

[0003] In XL-MIMO technology, the surge in the number of antennas leads to enormous overhead in traditional pilot design, which affects the accuracy of channel estimation.

[0004] Therefore, how to reduce the pilot overhead of near-field channel estimation is an urgent problem to be solved. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and storage medium for estimating XL-MIMO near-field compressed channels, which can reduce the pilot overhead of near-field channel estimation.

[0006] One embodiment of the present invention provides a method for estimating the near-field compressed channel of XL-MIMO, comprising:

[0007] Obtain the channel vector of the near-field compressed channel to be estimated;

[0008] The aforementioned channel vector is input into a channel estimation model constructed by a deep neural network, so that the channel estimation model can convert the aforementioned channel vector into a polar-domain channel vector through a built-in sparse transformation matrix.

[0009] The above polar-domain channel vector is compressed into a signal vector using the built-in sensing matrix;

[0010] Noise is added to the above signal vector to obtain the received signal vector;

[0011] The received signal vector is input into the built-in LAMP layer, so that the LAMP layer performs iterative calculations on the received signal vector using the built-in soft thresholding function to obtain the estimated polar channel vector corresponding to the above polar channel vector; wherein, the above soft thresholding function is calculated based on the built-in linear transformation parameters, nonlinear transformation parameters and linear transformation matrix;

[0012] The estimated polar channel vector is converted into an estimate of the near-field compressed channel to be estimated using a built-in sparse transformation matrix.

[0013] Furthermore, the channel estimation model is trained in two stages during the training process.

[0014] In the first stage, the initial perception matrix and the initial sparse transformation matrix are adjusted according to the first loss function until the loss function value corresponding to the first loss function converges.

[0015] In the second stage, the initial linear transformation parameters, initial nonlinear transformation parameters, and initial linear transformation matrix are adjusted according to the second loss function until the loss function value corresponding to the second loss function converges.

[0016] Furthermore, in the first stage, the initial perceptual matrix and the initial sparse transformation matrix are adjusted according to the first loss function until the loss function value corresponding to the first loss function converges, including:

[0017] Obtain several first channel vectors with first true labels, and initialize the values ​​of the perceptual matrix, sparse transformation matrix, linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix in the channel estimation model to be trained; wherein, the first true label is used to represent the true value of the estimated quantity of the first channel vector;

[0018] The first channel vector is input into the channel estimation model to be trained;

[0019] The channel estimation model to be trained transforms the first channel vector into the first polar domain channel vector using an initial sparse transformation matrix.

[0020] The first polar domain channel vector is compressed into a first signal vector using the initial sensing matrix;

[0021] Adding first noise to the first signal vector yields the first received signal vector;

[0022] The first received signal vector is input to the built-in LAMP algorithm layer to calculate the final estimated polar-domain channel vector of the LAMP algorithm layer based on the initial linear transformation parameters, the initial nonlinear transformation parameters, the initial linear transformation matrix, and the initial sensing matrix.

[0023] Based on the initial sparse transformation matrix and the final estimated polar-domain channel vector, the first estimate corresponding to the first channel vector is obtained.

[0024] Based on the first channel vector, the first estimate, and the first loss function formula described above, calculate the value of the first loss function;

[0025] For each calculated value of the first loss function, it is determined whether the first loss function has converged. If not, the values ​​of the perceptual matrix and the sparse transformation matrix are adjusted, and the channel estimation model to be trained continues to be trained. If yes, the first stage of training of the channel estimation model to be trained is completed, and the optimized perceptual matrix and the optimized sparse transformation matrix are obtained.

[0026] Furthermore, the LAMP algorithm layer described above contains several sub-algorithm layers;

[0027] For the first sub-algorithm layer in the LAMP algorithm layer, the first soft threshold function of the first sub-algorithm layer is calculated using the initial linear transformation parameters, the initial nonlinear transformation parameters, and the initial linear transformation matrix; based on the first soft threshold function and the first signal vector, the first estimated polar domain channel vector of the first sub-algorithm layer is calculated.

[0028] For a sub-algorithm layer that is not the first layer in the LAMP algorithm layer, the estimated polar domain channel vector of the current sub-algorithm layer is calculated based on the current perception matrix, the estimated polar domain channel vector of the previous layer, and the soft thresholding function of the previous layer.

[0029] Based on the estimated polar-domain channel vectors of all sub-algorithm layers in the LAMP algorithm layer, the final estimated polar-domain channel vector of the LAMP algorithm layer is calculated.

[0030] Furthermore, in the second stage, the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix are adjusted according to the second loss function until the loss function value corresponding to the second loss function converges, including:

[0031] Obtain several second channel vectors with second true labels; wherein the second true labels are used to represent the true values ​​of the estimated quantities of the second channel vectors.

[0032] The second channel vector is input into the channel estimation model of the second stage.

[0033] The channel estimation model to be trained described above transforms the second channel vector into a second polar domain channel vector using the optimized sparse transformation matrix.

[0034] The optimized perception matrix is ​​used to compress the second polar domain channel vector into a second signal vector.

[0035] A second noise is added to the second signal vector to obtain a second received signal vector;

[0036] The second received signal vector is input to the built-in LAMP algorithm layer to calculate the second estimated polar domain channel vector of each sub-algorithm layer of the LAMP algorithm layer based on the linear transformation parameters, nonlinear transformation parameters, linear transformation matrix and the optimized perception matrix.

[0037] Based on the optimized sparse transformation matrix and the second estimated polar domain channel vector of each sub-algorithm layer, the second estimator of each sub-algorithm layer is obtained.

[0038] Each time the second estimate of a sub-algorithm layer is calculated, the second loss function value of the current sub-algorithm layer is calculated based on the second channel vector, the second estimate, and the second loss function formula.

[0039] Determine whether the second loss function value of the current sub-algorithm layer has converged. If not, fix the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix in the previous sub-algorithm layer, adjust the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix of the current sub-algorithm layer, and continue training the channel estimation model to be trained. If yes, determine that the first stage of training of the channel estimation model to be trained is complete, and obtain the optimized linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix.

[0040] Furthermore, the second received signal vector is input to the built-in LAMP algorithm layer to calculate the estimated polar-domain channel vector for each sub-algorithm layer of the LAMP algorithm layer based on the linear transformation parameters, nonlinear transformation parameters, linear transformation matrix, and the optimized sensing matrix, including:

[0041] For the first sub-algorithm layer in the LAMP algorithm layer described above, the second estimated polar domain channel vector of the first sub-algorithm layer is calculated using the initial linear transformation parameters, the initial nonlinear transformation parameters, and the initial linear transformation matrix.

[0042] For sub-algorithm layers that are not the first level in the LAMP algorithm layer, first fix the values ​​of the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrices of all sub-algorithm layers before the current layer. Then, based on the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrices of the previous sub-algorithm layer, calculate the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrices of the current sub-algorithm layer. Finally, based on the linear transformation parameters, nonlinear transformation parameters, and linear transformation moments of the current sub-algorithm layer, calculate the estimated polar domain channel vector of the current sub-algorithm layer.

[0043] Furthermore, based on the optimized sparse transformation matrix, optimized sensing matrix, optimized linear transformation parameters, optimized nonlinear transformation parameters, optimized linear transformation matrix, and the estimator of the near-field compressed channel to be estimated, the near-field compressed channel to be estimated is reconstructed.

[0044] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments;

[0045] This invention provides an estimation device for XL-MIMO near-field compressed channels, comprising:

[0046] Channel vector acquisition module and estimation calculation module:

[0047] The aforementioned channel vector acquisition module is used to acquire the channel vector of the near-field compressed channel to be estimated;

[0048] The aforementioned estimation calculation module is used to input the aforementioned channel vector into a channel estimation model constructed by a deep neural network, so that the channel estimation model converts the aforementioned channel vector into a polar-domain channel vector through a built-in sparse transformation matrix; compresses the aforementioned polar-domain channel vector into a signal vector through a built-in sensing matrix; adds noise to the aforementioned signal vector to obtain a received signal vector; inputs the aforementioned received signal vector into a built-in LAMP layer, so that the LAMP layer iteratively calculates the received signal vector through a built-in soft thresholding function to obtain the estimated polar-domain channel vector corresponding to the aforementioned polar-domain channel vector; wherein the aforementioned soft thresholding function is calculated based on built-in linear transformation parameters, nonlinear transformation parameters, and a linear transformation matrix; and converts the aforementioned estimated polar-domain channel vector into an estimate of the aforementioned near-field compressed channel to be estimated through a built-in sparse transformation matrix.

[0049] Based on the above method embodiments, the present invention provides a corresponding terminal device embodiment;

[0050] The present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an estimation method for an XL-MIMO near-field compressed channel as described in any embodiment of the present invention.

[0051] Based on the above method embodiments, the present invention provides a corresponding storage medium embodiment;

[0052] The present invention provides a storage medium including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an estimation method for an XL-MIMO near-field compressed channel as described in any embodiment of the present invention.

[0053] The embodiments of the present invention have the following beneficial effects:

[0054] This invention provides a method, apparatus, terminal device, and storage medium for XL-MIMO near-field compressed channel estimation based on staged dual optimization. The method first obtains the channel vector of the near-field compressed channel to be estimated; then, the channel vector is input into a channel estimation model constructed by a deep neural network, so that the channel estimation model converts the channel vector into a polar-domain channel vector through a built-in sparse transformation matrix; then, the polar-domain channel vector is compressed into a signal vector through a built-in sensing matrix; then, noise is added to the signal vector to obtain a received signal vector; subsequently, the received signal vector is input into a built-in LAMP layer, so that the LAMP layer iteratively calculates the received signal vector through a built-in soft thresholding function to obtain the estimated polar-domain channel vector corresponding to the polar-domain channel vector; wherein, the soft thresholding function is calculated based on built-in linear transformation parameters, nonlinear transformation parameters, and a linear transformation matrix; finally, the estimated polar-domain channel vector is converted into an estimate of the near-field compressed channel to be estimated through a built-in sparse transformation matrix. Therefore, this invention, by combining deep neural networks and the LAMP algorithm, can obtain more accurate channel estimation, thereby reducing the number of pilot signals required by the system and effectively reducing pilot overhead. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating an estimation method for an XL-MIMO near-field compressed channel provided in one embodiment of the invention.

[0056] Figure 2 This is an internal structure diagram of a channel estimation model provided in an embodiment of the present invention.

[0057] Figure 3 This is a schematic diagram of the NMSE performance of AMP at different stages according to an embodiment of the present invention.

[0058] Figure 4 This is a schematic diagram illustrating the NMSE performance of the present invention, the OMP algorithm, the AMP algorithm, and the LAMP network under different SNRs, provided by an embodiment of the present invention.

[0059] Figure 5 This is a schematic diagram of the NMSE performance related to pilot overhead provided in an embodiment of the present invention.

[0060] Figure 6 This is a schematic diagram of a device structure for estimating an XL-MIMO near-field compression channel according to an embodiment of the present invention. Detailed Implementation

[0061] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0062] like Figure 1 As shown, an embodiment of the present invention provides a method for estimating an XL-MIMO near-field compressed channel, comprising:

[0063] Step S101: Obtain the channel vector of the near-field compressed channel to be estimated;

[0064] Specifically, a near-field compressed channel is used to represent the channel vector between the transmit and receive antennas of a uniform linear array of N antennas configured in a base station of a downlink XL-MIMO communication system, with equal spacing between the antennas and each antenna being half the carrier wavelength.

[0065] Schematic, the received signal described above can be represented by the following formula:

[0066]

[0067] h = [h0, h1, ..., h N-1 ] T

[0068] In the formula, y represents the received signal, P represents the pilot signal of the user's transmitting antenna, and n represents Gaussian noise, which follows the expression CN(0,σ). 2 I M The distribution is given by h, where h represents the channel vector and M represents the number of users.

[0069] In this embodiment, the channel vector to be estimated is obtained by receiving signals and pilot signals from the user's transmitting antenna.

[0070] In a preferred embodiment, when the distance between the base station and the scatterer is within the Rayleigh distance, the channel vector under the spherical wave assumption can be expressed by the following formula:

[0071]

[0072]

[0073] In the formula, L represents the number of scatterer components, and α l Let θ represent the gain of the l-th scatterer. l r represents the angle of the l-th scatterer. l b(θ) represents the distance from the l-th scatterer to the center of the uniform linear array. l,r l ) represents the near-field array steering vector. Let d represent the distance from the l-th scatterer to the n-th base station antenna, and let δ represent the antenna spacing. n =(2n-N+1) / 2 is a temporary variable.

[0074] In this preferred embodiment, the channel vector to be estimated when the distance between the base station and the scatterer is within the Rayleigh distance is calculated by using the number of scatterer components, the gain and angle of each scatterer, the distance from each scatterer to the center of the uniform linear array, the near-field array steering vector, the distance from each scatterer to each base station antenna, and the antenna spacing.

[0075] Step S102: Input the aforementioned channel vector into a channel estimation model constructed by a deep neural network, so that the channel estimation model converts the aforementioned channel vector into a polar-domain channel vector through a built-in sparse transformation matrix; compresses the aforementioned polar-domain channel vector into a signal vector through a built-in sensing matrix; adds noise to the aforementioned signal vector to obtain a received signal vector; input the aforementioned received signal vector into a built-in LAMP layer, so that the LAMP layer iteratively calculates the received signal vector through a built-in soft thresholding function to obtain the estimated polar-domain channel vector corresponding to the aforementioned polar-domain channel vector; wherein, the aforementioned soft thresholding function is calculated based on built-in linear transformation parameters, nonlinear transformation parameters, and a linear transformation matrix; convert the aforementioned estimated polar-domain channel vector into an estimate of the aforementioned near-field compressed channel to be estimated through a built-in sparse transformation matrix.

[0076] The internal structure of the above channel estimation model is illustrated as follows: Figure 2 As shown, the channel estimation model described above has a first layer, a second layer, a third layer, and so on up to the Tth layer. The parameters of the first layer and the Tth layer are sparse transformation matrices, the parameters of the second layer are perception matrices, and the fourth layer to the (T-1)th layer are collectively referred to as the LAMP algorithm layers, and the fourth layer to the (T-1)th layer are the various sub-algorithm layers of the LAMP algorithm layer.

[0077] In a preferred embodiment, the channel estimation model is trained in two stages.

[0078] In the first stage, the initial perception matrix and the initial sparse transformation matrix are adjusted according to the first loss function until the loss function value corresponding to the first loss function converges.

[0079] In another preferred embodiment, the adjustment of the initial perceptual matrix and the initial sparse transformation matrix according to the first loss function in the first stage, until the loss function value corresponding to the first loss function converges, includes:

[0080] Obtain several first channel vectors with first true labels, and initialize the values ​​of the perceptual matrix, sparse transformation matrix, linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix in the channel estimation model to be trained; wherein, the first true label is used to represent the true value of the estimated quantity of the first channel vector;

[0081] The first channel vector is input into the channel estimation model to be trained;

[0082] Specifically, the first channel vector mentioned above is represented by the following formula:

[0083] h1 = [h0, h1, ..., h N-1 ] T

[0084] The channel estimation model to be trained transforms the first channel vector into the first polar domain channel vector using an initial sparse transformation matrix.

[0085] Specifically, the above sparse transformation matrix can be represented by the following formula:

[0086]

[0087] In the formula, S n Indicates at angle θ n The number of sampling distances.

[0088] The channel vector is transformed into a polar-domain channel vector using the following formula:

[0089]

[0090] In the formula, H1 represents the first polar domain channel vector, W1 represents the initial sparse transformation matrix, and S represents the total number of sampling grids.

[0091] The first polar domain channel vector is compressed into a first signal vector using the initial sensing matrix;

[0092] Specifically, the aforementioned perception matrix can be represented by the following formula:

[0093]

[0094] In the formula, A1 represents the initial perception matrix.

[0095] Specifically, the polar-domain channel vector is compressed into a signal vector using the following formula;

[0096] r1=PW1H1=A1H1

[0097] In the formula, r1 represents the signal vector.

[0098] Adding first noise to the first signal vector yields the first received signal vector;

[0099] Specifically, the received signal vector is obtained using the following formula:

[0100] y1=A1H1+n1

[0101] In the formula, y1 represents the first received signal vector, and n1 represents the first noise.

[0102] The first received signal vector is input to the built-in LAMP algorithm layer to calculate the final estimated polar-domain channel vector of the LAMP algorithm layer based on the initial linear transformation parameters, the initial nonlinear transformation parameters, the initial linear transformation matrix, and the initial sensing matrix.

[0103] In another preferred embodiment, the LAMP algorithm layer described above includes several sub-algorithm layers;

[0104] For the first sub-algorithm layer in the LAMP algorithm layer, the first soft threshold function of the first sub-algorithm layer is calculated using the initial linear transformation parameters, the initial nonlinear transformation parameters, and the initial linear transformation matrix; based on the first soft threshold function and the first signal vector, the first estimated polar domain channel vector of the first sub-algorithm layer is calculated.

[0105] Specifically, the initial linear transformation parameter is set to 1, and the initial nonlinear transformation parameter is set to 1.1402. The first estimated polar-domain channel vector of the first sub-algorithm layer is calculated using the following formula:

[0106] v′1=y1

[0107]

[0108] c′1=B′1v′1

[0109]

[0110] In the formula, v′1 represents the received signal vector of the first sub-algorithm layer of the first stage LAMP algorithm layer, and (σ′1) 2 Let B'1 represent the noise variance of the first sub-algorithm layer of the first-stage LAMP algorithm layer, and let B'1 represent the initial linear transformation matrix of the first sub-algorithm layer of the first-stage LAMP algorithm layer. Let θ′ represent the first estimated polar-domain channel vector in the first stage. 11 θ′ represents the initial linear transformation parameters of the first sub-algorithm layer of the LAMP algorithm. 12 η represents the initial nonlinear transformation parameters of the first sub-algorithm layer of the LAMP algorithm. sst(c′1;θ′1;(σ′1) 2 ) represents the first soft threshold function of the first stage, and c1′ represents the intermediate variable of the first sub-algorithm layer of the LAMP algorithm layer in the first stage.

[0111] For a sub-algorithm layer that is not the first layer in the LAMP algorithm layer, the estimated polar domain channel vector of the current sub-algorithm layer is calculated based on the current perception matrix, the estimated polar domain channel vector of the previous layer, and the soft thresholding function of the previous layer.

[0112] Specifically, the linear parameters, nonlinear parameters, and linear transformation matrix in the soft thresholding function are fixed and all have the initial values ​​mentioned above. The estimated polar-domain channel vectors of the sub-algorithm layers (excluding the first layer) in the LAMP algorithm are calculated sequentially using the following formula:

[0113]

[0114]

[0115] In the formula, b′ k v′ represents the offset of the k-th sub-algorithm layer in the first stage of the LAMP algorithm. k This represents the updated received signal vector of the k-th sub-algorithm layer in the first stage of the LAMP algorithm, (σ′). k-1 ) 2 Let σ' represent the noise variance of the (k-1)th sub-algorithm layer in the first stage. k ) 2 Let c′ represent the noise variance of the k-th sub-algorithm layer in the first stage. k This represents the intermediate variable of the k-th sub-algorithm layer in the first stage of the LAMP algorithm. This represents the estimated polar-domain channel vector of the (k-1)th sub-algorithm layer in the first stage of the LAMP algorithm. η represents the estimated polar-domain channel vector of the k-th sub-algorithm layer in the first stage of the LAMP algorithm. sst (c′ k ;θ′1,(σ′ k ) 2 ) represents the soft threshold function of the k-th sub-algorithm layer of the first-stage LAMP algorithm layer.

[0116] Based on the estimated polar-domain channel vectors of all sub-algorithm layers in the LAMP algorithm layer, the final estimated polar-domain channel vector of the LAMP algorithm layer is calculated.

[0117] Specifically, the estimated polar-domain channel vector obtained from the last sub-algorithm layer of the LAMP algorithm is used as the estimated polar-domain channel vector corresponding to the aforementioned polar-domain channel vector.

[0118] Based on the initial sparse transformation matrix and the final estimated polar-domain channel vector, the first estimate corresponding to the first channel vector is obtained.

[0119] Specifically, the estimated polar-domain channel vector is transformed into a first estimator of the near-field compressed channel to be estimated using the following formula:

[0120]

[0121] In the formula, This represents the first estimate.

[0122] Based on the first channel vector, the first estimate, and the first loss function formula described above, calculate the value of the first loss function;

[0123] Specifically, the value of the first loss function is calculated using the following formula:

[0124]

[0125] In the formula, This represents the value of the first loss function mentioned above.

[0126] For each calculated value of the first loss function, it is determined whether the first loss function has converged. If not, the values ​​of the perceptual matrix and the sparse transformation matrix are adjusted, and the channel estimation model to be trained continues to be trained. If yes, the first stage of training of the channel estimation model to be trained is completed, and the optimized perceptual matrix and the optimized sparse transformation matrix are obtained.

[0127] Preferably, through the first stage of training, the optimized sparse transformation matrix and the optimized sensing matrix are better matched to the complex characteristics of the near-field channel. At the same time, the approximation error of the polar domain transformation matrix on the sensing matrix is ​​reduced, the transformation error between different channel domains is suppressed, and the tolerance to the sparsity of the polar domain channel is improved.

[0128] In the second stage, the initial linear transformation parameters, initial nonlinear transformation parameters, and initial linear transformation matrix are adjusted according to the second loss function until the loss function value corresponding to the second loss function converges.

[0129] In a preferred embodiment, in the second stage, adjusting the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix according to the second loss function until the loss function value corresponding to the second loss function converges includes:

[0130] Obtain several second channel vectors with second true labels; wherein the second true labels are used to represent the true values ​​of the estimates of the second channel vectors.

[0131] Specifically, the second channel vector mentioned above is represented by the following formula:

[0132] h2 = [h0, h1, ..., h N-1 ] T

[0133] In the formula, h2 represents the second channel vector mentioned above.

[0134] The second channel vector is input into the channel estimation model of the second stage.

[0135] The channel estimation model to be trained described above transforms the second channel vector into a second polar domain channel vector using the optimized sparse transformation matrix.

[0136] Specifically, the second channel vector is transformed into the second polar domain channel vector using the following formula:

[0137] h2=WH2

[0138] In the formula, W represents the optimized sparse transformation matrix, and H2 represents the second polar domain channel vector.

[0139] The optimized perception matrix is ​​used to compress the second polar domain channel vector into a second signal vector.

[0140] Specifically, the second polar domain channel vector is compressed into a second signal vector using the following formula:

[0141]

[0142] r2=PWH2=AH2

[0143] In the formula, A represents the optimized perception matrix and r2 represents the first signal vector.

[0144] A second noise is added to the second signal vector to obtain a second received signal vector;

[0145] Specifically, the second received signal vector is obtained through the following formula:

[0146] y2=AH2+n2

[0147] In the formula, y2 represents the received signal vector, and n2 represents the second noise.

[0148] The second received signal vector is input to the built-in LAMP algorithm layer to calculate the second estimated polar domain channel vector of each sub-algorithm layer of the LAMP algorithm layer based on the linear transformation parameters, nonlinear transformation parameters, linear transformation matrix and the optimized perception matrix.

[0149] Preferably, the second received signal vector is input to the built-in LAMP algorithm layer to calculate the estimated polar-domain channel vector of each sub-algorithm layer of the LAMP algorithm layer based on the linear transformation parameters, nonlinear transformation parameters, linear transformation matrix, and the optimized sensing matrix, including:

[0150] For the first sub-algorithm layer in the LAMP algorithm layer described above, the second estimated polar domain channel vector of the first sub-algorithm layer is calculated using the initial linear transformation parameters, the initial nonlinear transformation parameters, and the initial linear transformation matrix.

[0151] Specifically, the second estimated polar-domain channel vector of the first-layer sub-algorithm layer is calculated using the following formula:

[0152] v″1=y2

[0153]

[0154] c″1=B″1v′1

[0155]

[0156] In the formula, y2 represents the second received signal, v″1 represents the received signal vector of the first sub-algorithm layer of the second-stage LAMP algorithm layer, and (σ″1) 2 Let c''1 represent the noise variance of the first sub-algorithm layer of the second-stage LAMP algorithm layer, c''1 represent the intermediate variables of the first sub-algorithm layer of the second-stage LAMP algorithm layer, and B''1 represent the linear transformation matrix of the first sub-algorithm layer of the second-stage LAMP algorithm layer. θ″1 represents the second estimated polar-domain channel vector in the second stage, and θ″1 represents the linear and nonlinear transformation parameters of the first sub-algorithm layer of the LAMP algorithm layer in the second stage. 11 θ″ represents the linear transformation parameter of the first sub-algorithm layer in the second-stage LAMP algorithm layer. 12 This represents the nonlinear transformation parameters of the first sub-algorithm layer in the second-stage LAMP algorithm layer.

[0157] For sub-algorithm layers that are not the first level in the LAMP algorithm layer, first fix the values ​​of the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrices of all sub-algorithm layers before the current layer. Then, based on the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrices of the previous sub-algorithm layer, calculate the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrices of the current sub-algorithm layer. Finally, based on the linear transformation parameters, nonlinear transformation parameters, and linear transformation moments of the current sub-algorithm layer, calculate the estimated polar domain channel vector of the current sub-algorithm layer.

[0158] Specifically, the estimated polar-domain channel vector of the current sub-algorithm layer is calculated using the following formula:

[0159] θ′ k ′=θ′ k ′ -1

[0160]

[0161] B′ k ′=B′ k ′ -1

[0162]

[0163] In the formula, b″ k θ″ represents the offset of the k-th sub-algorithm layer in the second-stage LAMP algorithm layer. k c″ represents the linear and nonlinear transformation parameters of the k-th sub-algorithm layer in the second-stage LAMP algorithm layer. k This represents the intermediate variable of the k-th sub-algorithm layer in the second stage of the LAMP algorithm. Let θ″ represent the estimated polar-domain channel vector of the k-th sub-algorithm layer in the second-stage LAMP algorithm layer. k1 θ″ represents the linear transformation parameter of the k-th sub-algorithm layer in the second-stage LAMP algorithm layer. k2 This represents the nonlinear transformation parameter of the k-th sub-algorithm layer in the second-stage LAMP algorithm layer.

[0164] Based on the optimized sparse transformation matrix and the second estimated polar domain channel vector of each sub-algorithm layer, the second estimator of each sub-algorithm layer is obtained.

[0165] Specifically, the second estimator for each sub-algorithm layer is calculated using the following formula:

[0166]

[0167] In the formula, This indicates the second estimate.

[0168] Each time the second estimate of a sub-algorithm layer is calculated, the second loss function value of the current sub-algorithm layer is calculated based on the second channel vector, the second estimate, and the second loss function formula.

[0169] Specifically, the second loss function value of the current sub-algorithm layer is calculated using the following formula:

[0170]

[0171] In the formula, This represents the value of the second loss function mentioned above.

[0172] Determine whether the second loss function value of the current sub-algorithm layer has converged. If not, fix the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix in the previous sub-algorithm layer, adjust the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix of the current sub-algorithm layer, and continue training the channel estimation model to be trained. If yes, determine that the first stage of training of the channel estimation model to be trained is complete, and obtain the optimized linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix.

[0173] Preferably, by using the optimized perception matrix and sparse transformation matrix, the linear and nonlinear parameters in the LAMP network are optimized, which improves the system's detail and superiority. Such a two-stage training model can meet the strict RIP requirements of AMP.

[0174] Specifically, the sparse transformation matrix mentioned above is the optimized sparse transformation matrix.

[0175] In a preferred embodiment, the channel vector is transformed into a polar-domain channel vector using the following formula:

[0176]

[0177] In the formula, H represents the polar domain channel vector, and S represents the total number of sampling grids.

[0178] In this embodiment, the channel vector is transformed into a polar-domain channel vector using the parameters of the first layer in the channel estimation model, namely the sparse transformation matrix.

[0179] Specifically, the aforementioned built-in perception matrix is ​​the optimized perception matrix, and the perception matrix consists of the parameters of the second layer of the aforementioned channel estimation model.

[0180] Specifically, the aforementioned perception matrix can be represented by the following formula:

[0181]

[0182] In a preferred embodiment, the polar-domain channel vector is compressed into a signal vector using the following formula;

[0183] r = PWH = AH

[0184] In the formula, r represents the signal vector.

[0185] In this preferred embodiment, the channel vector is transformed into a polar-domain channel vector using the parameters of the second layer in the channel estimation model, namely the sparse transformation matrix.

[0186] In a preferred embodiment, the received signal vector is obtained using the following formula:

[0187] y = AH + n

[0188] In the formula, y represents the received signal vector.

[0189] In this preferred embodiment, the received signal vector is obtained by adding noise to the third layer of the channel estimation model.

[0190] Specifically, the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix mentioned above are optimized versions of the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix.

[0191] In a preferred embodiment, the estimated polar-domain channel vector of the first sub-algorithm layer of the LAMP algorithm layer is calculated using the following formula:

[0192] v1 = y

[0193]

[0194] c1 = Bv1

[0195]

[0196] In the formula, v1 represents the received signal vector of the first sub-algorithm layer of the LAMP algorithm layer described above. Let c represent the noise variance of the first layer, c1 represent the intermediate variables of the first layer, and B represent the optimized linear transformation matrix. θ represents the estimated polar-domain channel vector corresponding to the aforementioned polar-domain channel vector of the first sub-algorithm layer. 11 Let θ represent the optimized linear transformation parameters. 12 η represents the optimized nonlinear transformation parameters. sst This represents the soft threshold function.

[0197] The estimated polar-domain channel vectors of the sub-algorithm layers (excluding the first layer) in the LAMP algorithm are calculated sequentially using the following formula, and the estimated polar-domain channel vector obtained from the last sub-algorithm layer in the LAMP algorithm is taken as the estimated polar-domain channel vector corresponding to the above-mentioned polar-domain channel vector:

[0198]

[0199] In this preferred embodiment, the estimated polar-domain channel vector corresponding to the polar-domain channel vector is obtained by iterative calculation using a built-in soft threshold function.

[0200] Specifically, the sparse transformation matrix mentioned above is the optimized sparse transformation matrix.

[0201] In a preferred embodiment, the estimated polar-domain channel vector is transformed into an estimate of the near-field compressed channel to be estimated using the following formula:

[0202]

[0203] In the formula, It represents an estimate.

[0204] In this preferred embodiment, the estimated polar channel vector is transformed into an estimate of the near-field compressed channel to be estimated using an optimized sparse transformation matrix.

[0205] In another preferred embodiment, the near-field compressed channel to be estimated is reconstructed based on the optimized sparse transformation matrix, the optimized sensing matrix, the optimized linear transformation parameters, the optimized nonlinear transformation parameters, the optimized linear transformation matrix, and the estimator of the near-field compressed channel to be estimated.

[0206] Preferably, the reconstructed near-field compressed channel can be used for subsequent signal detection and related data monitoring.

[0207] In this preferred embodiment, the near-field compressed channel to be estimated is reconstructed using the optimized sparse transformation matrix, the optimized sensing matrix, the optimized linear transformation parameters, the optimized nonlinear transformation parameters, the optimized linear transformation matrix, and the estimator of the near-field compressed channel to be estimated.

[0208] Indicative, such as Figure 3 As shown, through the optimization process in the first and second stages, the error caused by the increase of signal-to-noise ratio in the original signal is greatly reduced. This design not only improves performance but also prevents undesirable local optima.

[0209] Indicative, such as Figure 4 As shown, the proposed scheme exhibits lower error compared to the other three existing schemes across all considered signal-to-noise ratio regions. Furthermore, by optimizing the sensing matrix and sparse transformation matrix in the first stage, the adverse effects of sparse channel approximation are mitigated while effectively capturing near-field channel characteristics, ensuring higher channel estimation accuracy.

[0210] Indicative, such as Figure 5 As shown, with SNR = 8dB, the pilot overhead increases from 96 to 256, corresponding to a compression ratio increase from 0.375 to 1. In Figure 5Within the specified range, the optimized scheme proposed in this invention achieves the same channel estimation accuracy as other schemes while reducing pilot overhead. In particular, the LAMP scheme requires approximately 224 pilot lengths of overhead to achieve -7dB NMSE, while the scheme proposed in this invention only requires approximately 128.

[0211] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0212] like Figure 6 As shown, an embodiment of the present invention provides an apparatus for XL-MIMO near-field compressed channel estimation based on staged dual optimization, comprising: a channel vector acquisition module and an estimation calculation module.

[0213] The aforementioned channel vector acquisition module is used to acquire the channel vector of the near-field compressed channel to be estimated;

[0214] The aforementioned estimation calculation module is used to input the aforementioned channel vector into a channel estimation model constructed by a deep neural network, so that the channel estimation model converts the aforementioned channel vector into a polar-domain channel vector through a built-in sparse transformation matrix; compresses the aforementioned polar-domain channel vector into a signal vector through a built-in sensing matrix; adds noise to the aforementioned signal vector to obtain a received signal vector; inputs the aforementioned received signal vector into a built-in LAMP layer, so that the LAMP layer iteratively calculates the received signal vector through a built-in soft thresholding function to obtain the estimated polar-domain channel vector corresponding to the aforementioned polar-domain channel vector; wherein the aforementioned soft thresholding function is calculated based on built-in linear transformation parameters, nonlinear transformation parameters, and a linear transformation matrix; and converts the aforementioned estimated polar-domain channel vector into an estimate of the aforementioned near-field compressed channel to be estimated through a built-in sparse transformation matrix.

[0215] It should be noted that the device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without creative effort. The above schematic diagrams are merely examples of a personalized feedback device for patient outpatient information and do not constitute a limitation on the personalized feedback device for patient outpatient information. It may include more or fewer components than illustrated, or combine certain components, or use different components.

[0216] Based on the above method embodiments, the present invention provides corresponding terminal device embodiments.

[0217] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an estimation method for an XL-MIMO near-field compressed channel as described in any embodiment of the present invention.

[0218] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the device.

[0219] The aforementioned terminal devices can be computing devices such as desktop computers, laptops, handheld computers, and cloud servers. These devices may include, but are not limited to, processors and memory.

[0220] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the device, connecting various parts of the device via various interfaces and lines.

[0221] The aforementioned memory can be used to store the aforementioned computer programs and / or modules. The aforementioned processor implements various functions of the aforementioned device by running or executing the computer programs and / or modules stored in the aforementioned memory, and by calling data stored in the memory. The aforementioned memory may mainly include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application program required for a function, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0222] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments.

[0223] Another embodiment of the present invention provides a storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the storage medium is located to execute an estimation method for an XL-MIMO near-field compression channel as described in any embodiment of the present invention.

[0224] In this embodiment, the storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0225] Compared with the prior art, by implementing the above embodiments of the present invention, the pilot overhead of near-field channel estimation can be reduced.

[0226] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for estimating the near-field compressed channel of XL-MIMO, characterized in that, include: Obtain the channel vector of the near-field compressed channel to be estimated; The channel vector is input to a channel estimation model constructed by a deep neural network, so that the channel estimation model transforms the channel vector into a polar-domain channel vector through a built-in sparse transformation matrix; compresses the polar-domain channel vector into a signal vector through a built-in sensing matrix; noise is added to the signal vector to obtain a received signal vector; the received signal vector is input to a built-in LAMP layer, so that the LAMP layer iteratively calculates the received signal vector through a built-in soft thresholding function to obtain the estimated polar-domain channel vector corresponding to the polar-domain channel vector; wherein, the soft thresholding function is calculated based on built-in linear transformation parameters, nonlinear transformation parameters, and a linear transformation matrix; the estimated polar-domain channel vector is transformed into an estimate of the near-field compressed channel to be estimated through a built-in sparse transformation matrix; Specifically, the channel estimation model is trained in two stages during training. In the first stage, several first channel vectors with first true labels are obtained, and the values ​​of the perceptual matrix, sparse transformation matrix, linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix in the channel estimation model to be trained are initialized; wherein, the first true label is used to represent the true value of the estimate of the first channel vector; The first channel vector is input to the channel estimation model to be trained; the channel estimation model to be trained transforms the first channel vector into a first polar domain channel vector through an initial sparse transformation matrix; the first polar domain channel vector is compressed into a first signal vector through an initial sensing matrix; first noise is added to the first signal vector to obtain a first received signal vector; the first received signal vector is input to the built-in LAMP algorithm layer to calculate the final estimated polar domain channel vector of the LAMP algorithm layer based on the initial linear transformation parameters, initial nonlinear transformation parameters, initial linear transformation matrix, and initial sensing matrix; Based on the initial sparse transformation matrix and the final estimated polar-domain channel vector, a first estimator corresponding to the first channel vector is obtained; based on the first channel vector, the first estimator, and the first loss function formula, the value of the first loss function is calculated; after each calculation of the first loss function value, it is determined whether the first loss function has converged; if not, the values ​​of the perceptual matrix and the sparse transformation matrix are adjusted, and the channel estimation model to be trained continues to be trained; if yes, it is determined that the first stage of training of the channel estimation model to be trained is completed, and the optimized perceptual matrix and the optimized sparse transformation matrix are obtained. In the second stage, the initial linear transformation parameters, initial nonlinear transformation parameters, and initial linear transformation matrix are adjusted according to the second loss function until the loss function value corresponding to the second loss function converges.

2. The estimation method for an XL-MIMO near-field compressed channel according to claim 1, characterized in that, The LAMP algorithm layer contains several sub-algorithm layers; For the first sub-algorithm layer in the LAMP algorithm layer, the first soft threshold function of the first sub-algorithm layer is calculated using the initial linear transformation parameters, the initial nonlinear transformation parameters, and the initial linear transformation matrix; and the first estimated polar domain channel vector of the first sub-algorithm layer is calculated based on the first soft threshold function and the first signal vector. For a sub-algorithm layer that is not the first layer in the LAMP algorithm layer, the estimated polar domain channel vector of the current sub-algorithm layer is calculated based on the current perception matrix, the estimated polar domain channel vector of the previous layer, and the soft thresholding function of the previous layer. The final estimated polar-domain channel vector of the LAMP algorithm layer is calculated based on the estimated polar-domain channel vectors of all sub-algorithm layers in the LAMP algorithm layer.

3. The estimation method for an XL-MIMO near-field compressed channel according to claim 2, characterized in that, In the second stage, the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix are adjusted according to the second loss function until the loss function value corresponding to the second loss function converges, including: Obtain several second channel vectors with second true labels; wherein the second true labels are used to represent the true values ​​of the estimates of the second channel vectors; The second channel vector is input into the channel estimation model of the second stage; The channel estimation model to be trained transforms the second channel vector into a second polar domain channel vector using an optimized sparse transformation matrix. The second polar domain channel vector is compressed into a second signal vector using the optimized perception matrix; A second noise is added to the second signal vector to obtain a second received signal vector; The second received signal vector is input to the built-in LAMP algorithm layer to calculate the second estimated polar domain channel vector of each sub-algorithm layer of the LAMP algorithm layer based on the linear transformation parameters, nonlinear transformation parameters, linear transformation matrix and the optimized sensing matrix. Based on the optimized sparse transformation matrix and the second estimated polar-domain channel vector of each sub-algorithm layer, the second estimate of each sub-algorithm layer is obtained; Each time the second estimate of a sub-algorithm layer is calculated, the second loss function value of the current sub-algorithm layer is calculated based on the second channel vector, the second estimate, and the second loss function formula. Determine whether the second loss function value of the current sub-algorithm layer has converged. If not, fix the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix in the previous sub-algorithm layer, adjust the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix of the current sub-algorithm layer, and continue training the channel estimation model to be trained. If yes, determine that the first stage of training of the channel estimation model to be trained is complete, and obtain the optimized linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix.

4. The estimation method for an XL-MIMO near-field compressed channel according to claim 3, characterized in that, The step of inputting the second received signal vector to the built-in LAMP algorithm layer to calculate the estimated polar-domain channel vector of each sub-algorithm layer of the LAMP algorithm layer based on the linear transformation parameters, nonlinear transformation parameters, linear transformation matrix, and the optimized sensing matrix includes: For the first sub-algorithm layer in the LAMP algorithm layer, the second estimated polar domain channel vector of the first sub-algorithm layer is calculated using the initial linear transformation parameters, the initial nonlinear transformation parameters, and the initial linear transformation matrix. For sub-algorithm layers that are not the first level in the LAMP algorithm layer, first fix the values ​​of the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrices of all sub-algorithm layers before the current layer. Then, based on the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrices of the previous sub-algorithm layer, calculate the linear transformation parameters, nonlinear transformation parameters, and linear transformation matrices of the current sub-algorithm layer. Finally, based on the linear transformation parameters, nonlinear transformation parameters, and linear transformation moments of the current sub-algorithm layer, calculate the estimated polar domain channel vector of the current sub-algorithm layer.

5. The estimation method for an XL-MIMO near-field compressed channel according to claim 4, characterized in that, Also includes: Based on the optimized sparse transformation matrix, optimized sensing matrix, optimized linear transformation parameters, optimized nonlinear transformation parameters, optimized linear transformation matrix, and the estimator of the near-field compressed channel to be estimated, the near-field compressed channel to be estimated is reconstructed.

6. An estimation device for an XL-MIMO near-field compressed channel, characterized in that, include: Channel vector acquisition module and estimation calculation module: The channel vector acquisition module is used to acquire the channel vector of the near-field compressed channel to be estimated; The estimation calculation module is used to input the channel vector into a channel estimation model constructed by a deep neural network, so that the channel estimation model converts the channel vector into a polar-domain channel vector through a built-in sparse transformation matrix; compresses the polar-domain channel vector into a signal vector through a built-in sensing matrix; adds noise to the signal vector to obtain a received signal vector; inputs the received signal vector into a built-in LAMP layer, so that the LAMP layer iteratively calculates the received signal vector through a built-in soft thresholding function to obtain the estimated polar-domain channel vector corresponding to the polar-domain channel vector; wherein, the soft thresholding function is calculated based on built-in linear transformation parameters, nonlinear transformation parameters, and a linear transformation matrix; and converts the estimated polar-domain channel vector into an estimate of the near-field compressed channel to be estimated through a built-in sparse transformation matrix. Specifically, the channel estimation model is trained in two stages during training. In the first stage, several first channel vectors with first true labels are obtained, and the values ​​of the perceptual matrix, sparse transformation matrix, linear transformation parameters, nonlinear transformation parameters, and linear transformation matrix in the channel estimation model to be trained are initialized; wherein, the first true label is used to represent the true value of the estimate of the first channel vector; The first channel vector is input to the channel estimation model to be trained; the channel estimation model to be trained transforms the first channel vector into a first polar domain channel vector through an initial sparse transformation matrix; the first polar domain channel vector is compressed into a first signal vector through an initial sensing matrix; first noise is added to the first signal vector to obtain a first received signal vector; the first received signal vector is input to the built-in LAMP algorithm layer to calculate the final estimated polar domain channel vector of the LAMP algorithm layer based on the initial linear transformation parameters, initial nonlinear transformation parameters, initial linear transformation matrix, and initial sensing matrix; Based on the initial sparse transformation matrix and the final estimated polar-domain channel vector, a first estimator corresponding to the first channel vector is obtained; based on the first channel vector, the first estimator, and the first loss function formula, the value of the first loss function is calculated; after each calculation of the first loss function value, it is determined whether the first loss function has converged; if not, the values ​​of the perceptual matrix and the sparse transformation matrix are adjusted, and the channel estimation model to be trained continues to be trained; if yes, it is determined that the first stage of training of the channel estimation model to be trained is completed, and the optimized perceptual matrix and the optimized sparse transformation matrix are obtained. In the second stage, the initial linear transformation parameters, initial nonlinear transformation parameters, and initial linear transformation matrix are adjusted according to the second loss function until the loss function value corresponding to the second loss function converges.

7. A terminal device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements an estimation method for an XL-MIMO near-field compressed channel as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the storage medium is located to perform an estimation method for an XL-MIMO near-field compressed channel as described in any one of claims 1 to 5.

Images