A massive MIMO uplink semi-blind channel estimation method

Abstract

The application discloses a large-scale MIMO uplink semi-blind channel estimation method based on independent component analysis and compressed sensing, which comprises the following steps: using a windowed least square channel estimation of a transmission pilot to perform channel preliminary estimation; performing discrete Fourier transform matrix conversion on the channel preliminary estimation result to an angle domain; using a compressed sensing recovery method to acquire angle domain features and a user number of a target user channel according to the channel preliminary estimation result in the angle domain; performing blind source separation of independent component analysis according to the user number to obtain each user channel including the target user channel with ambiguity; and using the angle domain features of the target user channel to eliminate the ambiguity of the target user channel to obtain an estimation result of the target user. The method combines the advantages of the compressed sensing technology and the independent component analysis technology, combines the two together, and greatly improves the channel estimation performance.

Description

Technical Field

[0001] This invention belongs to the field of communication channel estimation technology, specifically relating to a large-scale MIMO uplink semi-blind channel estimation method based on independent component analysis and compressed sensing. Background Technology

[0002] For massive MIMO systems to achieve efficient downlink resource scheduling and beamforming, accurate channel state information is crucial. In time-division duplex systems, channel reciprocity can be leveraged to obtain downlink channel state information through uplink channel state estimation. However, due to the limited orthogonal pilot resources caused by channel correlation time, multi-cell massive MIMO systems only allocate orthogonal pilots to users within the same cell. Furthermore, the correlation between pilots from users in different cells leads to the target user's channel estimation results being significantly affected by inter-cell interference during uplink channel estimation, resulting in a substantial decrease in accuracy. Therefore, high-precision channel estimation under interference conditions has significant application value.

[0003] To address channel estimation under interference conditions, numerous methods have been proposed, including compressed sensing-based channel estimation and semi-blind channel estimation methods utilizing independent component analysis (ICA). The former leverages the sparsity of the angle domain to transform the channel estimation problem into the recovery of sparse coefficients in the angle domain, obtaining the channel estimation result through compressed sensing recovery. The latter uses ICA to perform blind source separation between target user data and interfering user data, and designs a special pilot signal from multiple cell users to eliminate ambiguity, thereby obtaining the channel estimation result for the target user.

[0004] Existing compressed sensing-based channel estimation relies on prior channel information, such as the range of the angle of arrival and the channel's covariance, to estimate the sparsity coefficients of the channel in the angle domain. However, in real-world communication systems, base stations cannot know the users' prior channel information. Existing semi-blind channel estimation methods using independent component analysis require, on the one hand, joint scheduling and pilot design by multiple cell base stations, and on the other hand, relatively long pilot lengths for ambiguity removal. However, in practical communication systems, coordinated scheduling between base stations in different cells is very difficult, and the available pilot lengths are also limited. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a large-scale MIMO uplink semi-blind channel estimation method based on independent component analysis and compressed sensing, aiming to combine the advantages of compressed sensing and independent component analysis techniques to achieve a significant improvement in channel estimation performance.

[0006] The technical solution of this invention is as follows:

[0007] A method for estimating the uplink semi-blind channel of large-scale MIMO based on independent component analysis and compressed sensing, comprising the following steps:

[0008] Preliminary channel estimation is performed using windowed least-squares channel estimation based on the transmission pilots;

[0009] The preliminary channel estimation results are transformed to the angular domain using a discrete Fourier transform matrix.

[0010] Using compressed sensing recovery methods, the angle domain characteristics and number of users of the target user channel are obtained based on the preliminary channel estimation results in the angle domain.

[0011] Blind source separation by performing independent component analysis based on the number of users is obtained to obtain each user channel, including the target user channel, which is ambiguous.

[0012] The ambiguity of the target user channel is eliminated by utilizing the angular domain features of the target user channel, thereby obtaining the estimation result of the target user.

[0013] In one embodiment, the preliminary channel estimation using windowed least-squares channel estimation with transmission pilots specifically includes:

[0014] Based on the correlation between the target user's pilot and the base station's received pilot, the least-squares channel estimation result H on the m-th antenna is obtained. LS,m for:

[0015]

[0016] Among them, Γ p Λ represents the correlation matrix between the target user pilot in the nth subcarrier and the interfering user pilot in the lth cell; p L represents the cyclic phase shift between the target user pilot in the nth subcarrier and the pth user pilot in the same cell, where L represents the number of cells and P represents the number of users in each cell. For the target user channel, Intra-cell interference in least-squares channel estimation This refers to inter-cell interference in least-squares channel estimation. This represents the noise estimated by the LS channel in the m-th antenna; This represents the channel coefficient of the p-th user in the l-th cell in the m-th antenna.

[0017] Target user channel in the frequency domain Interference channels within the cell There exists a set of cyclic phase shifts Λ p By utilizing the properties of the discrete Fourier transform, the cyclic shift between the two in the frequency domain is converted into a cyclic shift in the time delay domain;

[0018] Based on the sparsity of the channel in the time delay domain, the interfering channel within the cell after cyclic shifting is separated from the target user channel in the time delay domain. A window function is used to eliminate the interfering channel within the cell, and the preliminary least-squares channel estimation result for windowed interference removal is obtained as follows:

[0019]

[0020] Among them, F K Let represent the K-dimensional discrete Fourier transform matrix; W represents a rectangular window function of length W.

[0021] In one embodiment, the step of using compressed sensing recovery method to obtain the angle domain characteristics of the target user channel and the number of target users based on the preliminary channel estimation results in the angle domain specifically includes:

[0022] The channel of the p-th user in the l-th cell is represented in the angle domain using the discrete Fourier transform matrix. For the cluster delay line channel model, the angular domain coefficients of the channel are row sparse, i.e. In each column, only Q elements are non-zero, and the non-zero elements in each column are in the same position.

[0023] The expression for the angle domain coefficient of the p-th user channel within the l-th cell is: Among them, F M Represents an M-dimensional discrete Fourier transform matrix; For non-zero row indices in the row sparse angle coefficients;

[0024] The least squares channel estimation results of the target user By performing angle-domain representation and obtaining sparse angle-domain coefficients, we can model it as a standard compressed sensing recovery problem, with the specific expression as follows:

[0025]

[0026]

[0027] Where C represents the row sparsity coefficient in the angle domain. This represents the least squares channel estimation result;

[0028] The standard compressed sensing recovery problem is solved using the compressed sensing recovery method of orthogonal matching pursuit, and the non-zero row index is obtained. And the angular domain row sparsity coefficient C;

[0029] indexed by nonzero rows The difference yields an estimate of the number of users.

[0030] In one embodiment, the compressed sensing recovery method using orthogonal matching pursuit solves the standard compressed sensing recovery problem to obtain the non-zero row index. And the angular domain row sparsity coefficient C, specifically including:

[0031] Input: M-dimensional discrete Fourier transform matrix F M Target user LS channel estimation results The number of non-zero rows is Q;

[0032] Output: Non-zero row indices Angular domain row sparsity coefficient C;

[0033] Initialization: Residual Orthogonal projection transformation matrix P0 = 0, set of non-zero row indices Angular domain row sparsity coefficient C = 0;

[0034] Orthogonal matching: For i = 1, ..., Q, perform (1), (2), and (3) in sequence;

[0035] (1) From F M Find the column with the largest dot product with r0. and i max Add to the non-zero row index set

[0036] (2) Calculate orthogonal projection

[0037] (3) New residuals I M Represents an M-dimensional identity matrix;

[0038] Calculate the angular domain coefficient

[0039] In one embodiment, the blind source separation performed based on the number of users using independent component analysis to obtain ambiguity in each user channel, including the target user channel, specifically includes:

[0040] The received signal Y(n) on the nth subcarrier at the base station is represented as:

[0041]

[0042] Where L represents the number of cells, and P represents the number of users in each cell. Let Z(n) represent the channel coefficient of the p-th user channel in the l-th cell on the n-th subcarrier, and Z(n) represent the received noise on the n-th subcarrier. This represents the transmitted data of the p-th user in the l-th cell on the n-th subcarrier. Orthogonal frequency division multiplexing (OFDM) is used to split the frequency-selective fading channel into N independent flat fading sub-channels, and independent component analysis is performed on each subcarrier. The independent component analysis is jointly implemented on each subcarrier by approximating the diagonalized eigenma matrix. The main steps include:

[0043] The received signal Y(n) from the base station is pre-whitened to obtain whitened data Y. W (n)=W(n)Y(n), W(n)=D y -1 / 2 E y H Let D represent the whitening matrix, where D is the whitening matrix. y E is a diagonal matrix composed of the eigenvalues ​​of the received signal covariance matrix. y The corresponding feature matrix;

[0044] Calculate whitening data Y W The fourth-order cumulative tensor F4(M) of (n) is obtained by performing eigenvalue decomposition to obtain the set of n eigenmatrices with the largest eigenvalues.

[0045] For the set of characteristic matrices Perform joint approximate diagonalization to obtain the joint approximate diagonalized unitary matrix V. JADE ;

[0046] Obtain channel estimation results for all users.

[0047] In one embodiment, the step of using the angular domain features of the target user channel to perform ambiguity cancellation on the target user channel to obtain the estimation result of the target user specifically includes:

[0048] Target user channel Ω(n) represents the phase shift caused by phase ambiguity; The selection matrix for eliminating order ambiguity is represented; ω(n) represents the phase shift caused by quadrant ambiguity.

[0049] in,

[0050]

[0051]

[0052] It is the detection data of all users output by independent component analysis on the nth subcarrier, and it is a U*K matrix. This represents the fourth power of the (n,k)th element;

[0053] Based on the channel's angular domain sparsity and energy differences, the selection matrix for eliminating order ambiguity is as follows:

[0054]

[0055]

[0056]

[0057] Among them, I U Represents a U-dimensional identity matrix; This represents the normalized channel energy of the u-th channel. This represents the difference in the normalized angle domain main direction index, where 'o' stands for optimal, indicating that it is... The optimal solution;

[0058] Based on the correlation between the transmission pilot and the reception pilot, the quadrant phase shift corresponding to quadrant ambiguity is obtained as follows:

[0059]

[0060]

[0061]

[0062] in, This indicates the extraction of the real part; This represents four possible quadrant phase shifts; This represents the pilot symbol for the target user on the nth subcarrier;

[0063] This indicates the received pilot signal after phase ambiguity cancellation and sequence ambiguity cancellation. q represents the index of the pilot signal in the transmission symbol. o (n) represents the quadrant phase shift type of quadrant ambiguity.

[0064] In one embodiment, Further expressed as:

[0065]

[0066]

[0067] Where, m n,u This represents the angular domain feature index of the u-th channel in the independent component analysis estimation results; This represents the angle domain feature index of the p-th user within the l-th cell.

[0068] The u-th channel is transformed into its angular domain representation using the discrete Fourier transform matrix. The largest angle domain coefficient index corresponds to the angle domain feature index, represented as:

[0069]

[0070] The angle domain coefficients of the least-squares channel estimation for the p-th user in the l-th cell are expressed as follows: The angular domain feature index of the p-th user is represented as:

[0071] The technical solution of this disclosure combines the advantages of compressed sensing technology and independent component analysis technology. By combining the two, a semi-blind channel estimation method with low pilot overhead, which does not rely on inter-cell base station cooperation, channel prior information, or inter-cell information, is proposed. Simulation results and complexity analysis show that the proposed method achieves a significant improvement in channel estimation performance by sacrificing only a small amount of complexity compared to traditional methods.

[0072] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0073] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0074] Figure 1 This is a schematic diagram of the uplink semi-blind channel estimation method for large-scale MIMO in an embodiment of the present invention;

[0075] Figure 2 This is a comparison chart of the channel estimation NMSE performance of the CS-ICA method under different signal-to-noise ratio settings in the embodiments of the present invention;

[0076] Figure 3 This is a comparison chart of the output SINR performance of the CS-ICA method under different numbers of users in the embodiments of the present invention. Detailed Implementation

[0077] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present invention are shown in the drawings, not the entire structure.

[0078] Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the steps as sequential processes, many of these steps can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the steps can be rearranged. The process can be terminated when its operation is complete, but may also have additional steps not included in the figures. The process can correspond to a method, function, procedure, subroutine, subroutine, etc.

[0079] like Figure 1 As shown in the embodiment, a large-scale MIMO uplink semi-blind channel estimation method based on Independent Component Analysis (ICA) and Compressed Sensing (CS), also known as the CS-ICA method, includes the following steps:

[0080] Preliminary channel estimation is performed using windowed least square (LS) channel estimation based on the transmission pilots;

[0081] The preliminary channel estimation results are transformed to the angular domain using a Discrete Fourier Transform (DFT) matrix.

[0082] Using compressed sensing recovery methods, the angle domain characteristics and number of users of the target user channel are obtained based on the preliminary channel estimation results in the angle domain.

[0083] Blind source separation by performing independent component analysis based on the number of users is obtained to obtain each user channel, including the target user channel, which is ambiguous.

[0084] The ambiguity of the target user channel is eliminated by utilizing the angular domain features of the target user channel, thereby obtaining the estimation result of the target user.

[0085] The specific implementation process is as follows:

[0086] Windowed LS channel coarse estimation:

[0087] The base station's receive pilot on the nth subcarrier is:

[0088]

[0089] in, Z(n) represents the pilot signal transmitted by the p-th user in the l-th cell on the n-th subcarrier; Z(n) represents the received noise on the n-th subcarrier. This represents the channel coefficient of the p-th user channel in the l-th cell on the n-th subcarrier.

[0090] By correlating the target user's pilot signals with the base station's received pilot signals, the LS channel estimation result on the m-th antenna can be obtained as follows:

[0091]

[0092] Among them, Γ p Λ represents the correlation matrix between the target user pilot in the nth subcarrier and the interfering user pilot in the lth cell; p This represents the cyclic phase shift between the target user pilot in the nth subcarrier and the pth user pilot in the cell; This represents the noise estimated by the LS channel in the m-th antenna; Let Hinter represent the channel coefficient of the p-th user in the l-th cell at the m-th antenna. (2) In equation (2), the first term is the target user channel, the second term is the intra-cell interference in LS channel estimation, and the third term is the inter-cell interference in LS channel estimation, denoted as Hinter. It can be found that there is a set of cyclic phase shifts Λ between the target user channel and the intra-cell interference channel in the frequency domain. p By utilizing the properties of the DFT, the cyclic shift between the two in the frequency domain can be converted into a cyclic shift in the time delay domain. Furthermore, due to the sparsity of the channel in the time delay domain, the intra-cell interference channel, after cyclic shifting, will be separated from the target user channel in the time delay domain. Therefore, a window function can be used to eliminate the intra-cell interference channel. Thus, the preliminary estimation result of the windowed LS channel for interference removal can be obtained as follows:

[0093]

[0094] Among them, F K Let represent a K-dimensional DFT matrix; W represents a rectangular window function of length W.

[0095] User number estimation based on compressed sensing:

[0096] For a uniform linear array (ULA) in a large-scale MIMO system, the channel of the p-th user in the l-th cell can be represented in the angle domain using the DFT matrix. For the Clustered Delay Line (CDL) channel model, the angular domain coefficients of the channel are row-sparse, i.e. In each column, only Q elements are non-zero, and the non-zero elements in each column are in the same position. Therefore, the angle domain coefficient of the p-th user channel in the l-th cell can be expressed as:

[0097]

[0098] In the formula, FM It is an M-dimensional DFT matrix; The index of the non-zero row in the sparse angle coefficient.

[0099] Based on the above characteristics, the LS channel estimation results of the target user will be used. By performing angle-domain representation and obtaining sparse angle-domain coefficients, the problem can be modeled as a standard compressed sensing recovery problem as follows:

[0100]

[0101] This represents the channel estimation results for all antennas of the target user, which are pieced together from the estimation results of each antenna. The compressed sensing recovery problem described above can be solved using compressed sensing recovery methods such as Orthogonal Matching Pursuit (OMP). The specific process is as follows:

[0102] Input: M-dimensional DFT matrix F M LS channel estimation results for target users The number of non-zero rows is Q.

[0103] Output: Non-zero row indices Angular domain row sparsity coefficient C.

[0104] ① Initialization: Residual Orthogonal projection transformation matrix P0 = 0, set of non-zero row indices The angular domain row sparsity coefficient C = 0.

[0105] ② Orthogonal matching: for i = 1, ..., Q

[0106] I. From F M Find the column with the largest dot product with r0. and i max Add to the non-zero row index set

[0107] II. Calculate orthogonal projection

[0108] III. Subtract the impact of this tracking from the preliminary channel estimation results to obtain the new residual. I M Represents an M-dimensional identity matrix. Or written as...

[0109] ③ Calculate the angular domain coefficient

[0110] Finally, through non-zero row indexes The difference can be used to obtain an estimate of the number of target users.

[0111]

[0112] Blind source separation using independent component analysis:

[0113] The received signal on the nth subcarrier at the base station can be represented as:

[0114]

[0115] Where Z(n) represents the received noise on the nth subcarrier; This represents the transmitted data of the p-th user within the l-th cell on the n-th subcarrier. Orthogonal Frequency Division Multiplexing (OFDM) splits the frequency-selective fading channel into N independent flat fading sub-channels and performs independent component analysis on each subcarrier. Here, Joint Approximation Diagonalization via Eigenmatrix (JADE) is implemented on each subcarrier, and its main steps are as follows:

[0116] ① Pre-whitening the received signal from the base station reduces the complexity of subsequent methods. The whitened signal can be represented as Y. W (n)=W(n)Y(n), W(n)=D y -1 / 2 E y H Let D represent the whitening matrix, where D is the whitening matrix. y E is a diagonal matrix composed of the eigenvalues ​​of the received signal covariance matrix. y This is the corresponding characteristic matrix.

[0117] ② Calculate the whitening data Y W The fourth-order cumulative tensor F4(M) of (n) is used, and eigenvalue decomposition is performed to obtain the set of n eigenmatrices with the largest eigenvalues.

[0118] ③ For the set of characteristic matrices Perform joint approximate diagonalization to obtain the joint approximate diagonalized unitary matrix V. JADE .

[0119] ④ Obtain the channel estimation results for all users. Detection data from all users

[0120] The channel estimation results obtained directly from independent component analysis (ICA) still exhibit three types of ambiguity compared to the channel estimation results for the target user. First, there is phase ambiguity and quadrant ambiguity, which pertain to the blind source separation results of ICA. Compared to the actual target user channel Some phase shift still exists. Secondly, there is order ambiguity, meaning that the blind source separation results of independent component analysis contain channel estimation results for multiple users, of which only one is the result for the target user, which needs to be identified and selected. Therefore, the target user channel can be represented as:

[0121]

[0122] Where Ω(n) represents the phase shift caused by phase ambiguity; ω(n) represents the selection matrix for eliminating order ambiguity; ω(n) represents the phase shift caused by quadrant ambiguity.

[0123] Blur removal based on angle domain sparsity:

[0124] For users of 4-QAM modulation, phase ambiguity can be calculated from statistics of the received signal:

[0125]

[0126] It represents the detection data of all users output by the ICA on the nth subcarrier. It is a U*K matrix. This represents the fourth power of the (n,k)th element.

[0127] Then, by utilizing the angular domain sparsity and energy differences of the channel, the selection matrix for eliminating order ambiguity can be obtained as follows:

[0128]

[0129] Among them, I U Represents a U-dimensional identity matrix; u n,o This represents an index. The subscript 'o' indicates optimal, meaning it is... The optimal solution.

[0130] The normalized channel energy of the u-th channel can be further expressed as:

[0131]

[0132] in, This represents the channel energy of the u-th channel.

[0133] The difference in the normalized angle domain principal direction index can be further expressed as:

[0134]

[0135] Where, m n,u This represents the angular domain feature index of the u-th channel in the independent component analysis estimation results; This represents the angle domain feature index of the p-th user within the first cell.

[0136] The u-th channel is transformed into the angle domain by the DFT matrix and represented as follows: The largest angle domain coefficient index corresponds to the angle domain feature index, which can be represented as:

[0137]

[0138] The angle domain coefficients of the LS channel estimation for the p-th user in the first cell can be expressed as: Therefore, the angular domain feature index of the p-th user can be represented as:

[0139]

[0140] Finally, by utilizing the correlation between the transmitted pilot and the received pilot, the quadrant phase shift corresponding to quadrant ambiguity is obtained as follows:

[0141]

[0142] in, This indicates the extraction of the real part; Let q represent the qth possible quadrant phase shift. o (n) represents the quadrant phase shift type of quadrant ambiguity. For 4QAM, there are a total of 4 quadrant phase shifts. This represents the pilot symbol for the target user on the nth subcarrier; This indicates the received pilot signal after phase ambiguity cancellation and sequence ambiguity cancellation. This indicates the index of the pilot signal in the transmission symbol.

[0143] To better demonstrate the effects of the present invention, Monte Carlo simulations were performed in the embodiments. The system simulation parameters were set as follows: number of base station antennas M = 128, number of subcarriers N = 256, and pilot length K. p =1, number of transmitted symbols K=200, number of cells L=7, number of users per cell P=2. The simulation uses the CDL-B model in 3GPP TR38.901, selecting the first 3 clusters, and the channel experimental extension is set to 1000ns.

[0144] Figure 2The paper demonstrates the NMSE performance of the proposed CS-ICA method under different signal-to-noise ratio (SNR) settings, with the large-scale fading parameter for in-cell users set to 1 and the large-scale fading parameter for out-of-cell users set to 0.7. The performance is compared with that of windowed LS channel estimation methods, SVD-based channel estimation methods, ICA-based channel estimation methods, and MMSE channel estimation methods with known channel covariance information. Within the SNR range of -5dB to 15dB, the proposed CS-ICA method shows a 10dB gain compared to the windowed LS channel estimation method. All of these methods assume that the base station does not know the prior channel state information of the users. However, the LMMSE method assumes that the covariance information of all users is known and achieves strong channel estimation accuracy, yet the proposed method suffers only a 1dB performance loss compared to these methods.

[0145] The method complexity analysis compares the normalized numerical and symbolic complexity of the proposed CS-ICA method with windowed LS channel estimation, SVD-based channel estimation, ICA-based channel estimation, and MMSE channel estimation with known channel covariance information. The complexity analysis shows that the proposed CS-ICA method only increases complexity by 15% compared to ICA-based channel estimation methods, but achieves a performance gain of nearly 15 dB. Furthermore, compared to LMMSE methods with known perfect channel information, the complexity of the proposed CS-ICA method decreases to 1 / 600, while achieving similar NMSE channel estimation performance. Compared to SVD-based channel estimation methods, the proposed CS-ICA method increases complexity by 11%, but achieves a performance gain of nearly 8 dB.

[0146] Table 1 Comparison of the method in this embodiment with other methods

[0147]

[0148] Figure 3 The output SINR performance of the proposed CS-ICA method is demonstrated under different numbers of users, with the large-scale fading parameter set to 1 for in-cell users and 0.7 for out-of-cell users. It can be observed that the method achieves optimal performance only when the number of users used is exactly equal to the actual number of users. The output SINR performance achieved using the CS-based user number estimation proposed in this invention suffers only a 1 dB loss compared to the optimal performance.

[0149] The technical solution of this disclosure combines the advantages of compressed sensing technology and independent component analysis technology. By combining the two, a semi-blind channel estimation method with low pilot overhead, which does not rely on inter-cell base station cooperation, channel prior information, or inter-cell information, is proposed. Simulation results and complexity analysis show that the proposed method achieves a significant improvement in channel estimation performance by sacrificing only a small amount of complexity compared to traditional methods.

[0150] In this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a step or method that comprises a list of elements includes not only those elements but also other elements not expressly listed or inherent to such a step or method.

[0151] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for estimating the uplink semi-blind channel of large-scale MIMO based on independent component analysis and compressed sensing, characterized in that, The method includes the following steps: Preliminary channel estimation is performed using windowed least-squares channel estimation based on the transmission pilots; The preliminary channel estimation results are transformed to the angular domain using a discrete Fourier transform matrix. Using compressed sensing recovery methods, the angle domain characteristics and number of users of the target user channel are obtained based on the preliminary channel estimation results in the angle domain. Blind source separation by performing independent component analysis based on the number of users is obtained to obtain each user channel, including the target user channel, which is ambiguous. The ambiguity of the target user channel is eliminated by utilizing the angular domain features of the target user channel to obtain the estimation result of the target user; The method of utilizing compressed sensing to recover information, based on the preliminary channel estimation results in the angle domain, obtains the angle domain characteristics of the target user channel and the number of target users, specifically including: Using the discrete Fourier transform matrix to the first l The first in the community p The channel for each user is represented in the angle domain. For the cluster delay line channel model, the angular domain coefficients of the channel are row sparse, i.e. In each column, only Q All elements are non-zero, and the non-zero elements in each column are in the same position; No. l The first in the community p The expression for the angle domain coefficients of a user channel is: ,in, express M 3D discrete Fourier transform matrix; For non-zero row indices in the row sparse angle coefficients; The least squares channel estimation results for the target user By performing angle-domain representation and obtaining sparse angle-domain coefficients, we can model it as a standard compressed sensing recovery problem, with the specific expression as follows: in, Represents the sparse coefficients in the angle domain. This represents the least squares channel estimation result; The standard compressed sensing recovery problem is solved using the compressed sensing recovery method of orthogonal matching pursuit, and the non-zero row index is obtained. and angular domain row sparsity coefficient ; indexed by nonzero rows The difference yields an estimate of the number of users. ; The blind source separation, which involves performing independent component analysis based on the number of users, to obtain ambiguity in each user channel, including the target user channel, specifically includes: Base station end n Received signal on each subcarrier Represented as: in, L Indicates the number of cells. P This indicates the number of users in each community. Indicates the first l The first in the community p The user in the first n Data transmitted on each subcarrier Indicates the first l The first in the community p The channel coefficients of a user channel on the nth subcarrier. Indicates the first n Noise is received on each subcarrier, and the frequency-selective fading channel is split into multiple subcarriers using orthogonal frequency division multiplexing. N Each subcarrier has an independent flat fading sub-channel, and independent component analysis is performed on each subcarrier. The independent component analysis is jointly implemented on each subcarrier by approximating the diagonalized eigenma matrix. The main steps include: Receiving signals from base stations Pre-whitening is performed to obtain whitening data. , Let represent the whitening matrix, where The diagonal matrix formed by the eigenvalues ​​of the received signal covariance matrix. The corresponding feature matrix; Calculate whitening data The fourth-order cumulative tensor Then perform eigenvalue decomposition to obtain the eigenvalue with the largest value. n A set consisting of characteristic matrices ; For the set of characteristic matrices Perform joint approximate diagonalization to obtain the joint approximate diagonalized unitary matrix. ; Obtain channel estimation results for all users. .

2. The method for estimating the uplink semi-blind channel of large-scale MIMO according to claim 1, characterized in that, The preliminary channel estimation using windowed least-squares channel estimation with transmission pilots specifically includes: Based on the correlation between the pilot signals received by the target user and the pilot signals received by the base station, the first... m Least-squared channel estimation results on each antenna for: in, Indicates the first n The target user pilot in the first subcarrier and the first l The first community p Correlation matrix of pilot signals for interfering users; Indicates the first n The target user pilot in each subcarrier is the same as the first one in the same cell. p Cyclic phase shift between user pilots L Indicates the number of cells. P This indicates the number of users in each community; For the target user channel, Intra-cell interference in least-squares channel estimation This refers to inter-cell interference in least-squares channel estimation. Indicates the first m Noise in least-squares channel estimation for each antenna Indicates the first l The first community p The user in the first m Channel coefficients in each antenna; Target user channel in the frequency domain Interference channels within the cell There exists a set of cyclic phase shifts By utilizing the properties of the discrete Fourier transform, the cyclic shift between the two in the frequency domain is converted into a cyclic shift in the time delay domain; Based on the sparsity of the channel in the time delay domain, the interfering channel within the cell after cyclic shifting is separated from the target user channel in the time delay domain. A window function is used to eliminate the interfering channel within the cell, and the preliminary least-squares channel estimation result for windowed interference removal is obtained as follows: in, express K 3D discrete Fourier transform matrix; Indicates length is W The rectangular window function.

3. The method for estimating the uplink semi-blind channel of large-scale MIMO according to claim 1, characterized in that, The compressed sensing recovery method using orthogonal matching pursuit solves the standard compressed sensing recovery problem, obtaining non-zero row indices. and angular domain row sparsity coefficient Specifically, it includes: Input: M-dimensional discrete Fourier transform matrix Target user least squares channel estimation results The number of non-zero rows is Q; Output: Non-zero row indices ; Angular domain row sparsity coefficient ; Initialization: Residual Orthogonal projection transformation matrix non-zero row index set Angular domain row sparsity coefficient ; Orthogonal matching: for Then proceed with (1), (2), and (3) in sequence. (1) From Searching for and The column with the largest inner product and will Add to the non-zero row index set ; (2) Calculate the orthogonal projection ; (3) New residuals , Represents an M-dimensional identity matrix; Calculate the angular domain coefficient .

4. The method for estimating the uplink semi-blind channel of large-scale MIMO according to claim 1, characterized in that, The step of using the angular domain features of the target user channel to perform ambiguity removal on the target user channel to obtain the estimation result of the target user specifically includes: Target user channel , This indicates the phase shift caused by phase ambiguity; This represents the selection matrix for eliminating order ambiguity. This indicates the phase shift caused by quadrant blurring; in, in, Indicates the first n The detection data of all users, output by independent component analysis on each subcarrier, is a matrix. Then it means that the (th) n,k ) elements raised to the power of 4; Based on the channel's angular domain sparsity and energy differences, the selection matrix for eliminating order ambiguity is as follows: in, This represents an index. express U 1D identity matrix Indicates the first u Normalized channel energy of the path channel Indicates the difference in the normalized angle domain main direction index, subscript o Indicates the optimal solution; Based on the correlation between the transmission pilot and the reception pilot, the quadrant phase shift corresponding to quadrant ambiguity is obtained as follows: in, This indicates the extraction of the real part; This represents four possible quadrant phase shifts; This represents the pilot symbol for the target user on the nth subcarrier; This indicates the received pilot signal after phase ambiguity cancellation and sequence ambiguity cancellation. This indicates the index of the pilot signal in the transmission symbol. This indicates a quadrant phase shift type where the quadrant is blurred.

5. The large-scale MIMO uplink semi-blind channel estimation method according to claim 4, characterized in that, Further expressed as: in, This indicates that in the independent component analysis estimation results, the first... u Angular domain feature index of the road channel; Indicates the first l The first in the community p Angle domain feature index for each user; No. u The path channel is transformed into the angular domain using the discrete Fourier transform matrix as follows: The largest angle domain coefficient index corresponds to the angle domain feature index, which is represented as: No. l The angle domain coefficients of the least-squares channel estimation for the p-th user within a cell are expressed as follows: The angular domain feature index of the p-th user is represented as: .

Images