A Target Signal Estimation and Demodulation Method Based on Message Passing in Sudden Interference Scenarios

Abstract

This invention belongs to the field of information and communication technology, specifically relating to a receiver design method for burst interference scenarios in message passing. Considering that in burst interference scenarios, burst interference on some data symbols within a frame and background interference on pilot symbols are not from the same source, and the statistical characteristics of burst interference cannot be estimated by measuring pilot symbols, existing interference suppression frameworks have poor demodulation performance. This invention, based on a message passing framework, designs different noise reduction modules using target user constellation diagram information and the low-rank nature of the interference signal. Through iteration between different modules, burst interference is effectively suppressed, significantly improving receiver performance.

Description

Technical Field

[0001] This invention belongs to the field of information and communication technology, and specifically relates to a receiver design method for sudden interference scenarios based on message passing. Background Technology

[0002] In 5.5G uplink ultra-wideband communication (UCBC) scenarios, rearming the F, A, and E bands is required for uplink data transmission. However, these bands suffer from inter-standard interference, atmospheric waveguide interference, and electromagnetic pollution, which significantly limit the performance of the communication system. These interferences have the following characteristics: 1. The energy intensity of the interference is very high, with an interference-to-noise ratio (INR) exceeding 20 dB compared to the noise floor; 2. Within a transmission time interval (TTI), the sources of interference differ on different OFDM symbols. In other words, unlike the background interference that exists throughout the entire TTI from within the system, some OFDM symbols experience interference from the complex electromagnetic environment outside the system, and their interference characteristics are inconsistent with the background interference; 3. In sudden interference scenarios, the interference only occurs on some data symbols, while the pilot symbols are not affected by the sudden interference, making it impossible to measure the characteristics of sudden interference on data symbols through the pilot symbols.

[0003] Within the existing framework of the Minimum Mean Squared Error-based Interference Rejection Combining (MMSE-IRC) algorithm, accurately estimating the interference covariance matrix on data symbols is crucial. However, for data symbols subjected to non-homogeneous burst interference, general covariance estimation methods struggle to effectively eliminate the cross-terms between the target signal and the burst interference. The energy intensity of these cross-terms increases with the enhancement of the target signal, leading to an upper bound on the signal-to-interference-plus-noise ratio (SINR) of the data symbol estimation. This is particularly pronounced under higher-order modulation (64QAM and above), where the detection performance deteriorates sharply compared to the absence of cross-terms. Summary of the Invention

[0004] This invention improves the performance of target user signal estimation and demodulation through iterative interference cancellation within a Bayesian framework. The algorithm design is based on a Turbo-type Message Passing (TMP) framework. The receiver in this invention consists of three modules: an LMMSE linear estimation module A, a soft demodulation and noise reduction module B, and a low-rank noise reduction module C. In module A, an LMMSE linear estimator obtains coarse estimates of the target signal and interference signals, which are then input to modules B and C, respectively. Module B then uses constellation diagram information to reduce noise in the target signal estimate. Furthermore, since the number of antennas at the base station receiver far exceeds the number of interference sources, the interference signal exhibits low-rank characteristics in space. Therefore, module C can utilize the low-rank nature of the interference signal to achieve noise reduction. The three modules iteratively optimize the estimates until convergence.

[0005] The technical solution adopted in this invention is a target signal estimation and demodulation method based on message passing in sudden interference scenarios, which includes the following steps:

[0006] S1, System Modeling: There are T systems within a single frame. i One OFDM data symbol is affected by a burst of interference, and the location of the interference is known (assuming the interference does not occur on the pilot symbol). The total frequency domain bandwidth of the system is K = 12N. RB N subcarriers RB Let N be the number of resource blocks (RBs). Assume the number of receiving antennas at the base station is N. r The number of target signal streams is N s The number of interference signal streams is N i The signal received by the base station on the k-th subcarrier of the t-th data symbol can be represented as follows:

[0007]

[0008] in For the target user channel, To interfere with user channels, For target signal, This is an interference signal. It is white noise.

[0009] S2. Block Model: Considering that the channel changes of adjacent subcarriers in the frequency domain are relatively slow, adjacent R RBs are regarded as a sub-block, then the system is divided into Q = N blocks. RB / R sub-blocks, while T can be i Identical sub-blocks of OFDM symbols are concatenated together, meaning each sub-block has a dimension of N. r ×M, where M=KT i / Q, the received signal in the m-th column within the q-th sub-block is...

[0010] y q,m =H q,m s q,m +H I,q,m s I,q,m +n q,m

[0011] =H q,m s q,m +l q，m +n q,m

[0012] Subscript q，m This corresponds to the qK / Q+mod(m-1, K / Q)+1th subcarrier on the ceil(mQ / K)th OFDM data symbol subjected to burst interference in the system. ceil(x) represents the floor function of x, and mod(x, y) represents the modulo value of x with respect to y. q，m =H I，q，m s I，q，m This represents the interference signal. The target symbol s... q，m and interference signal l q，m Represented in matrix form S q =[s q，1 , ..., s q，M ] and L q =[l q，1 , ..., l q，M In subsequent steps, each sub-block is processed independently with identical processing steps. For the sake of simplicity in notation, the sub-block subscript q will be omitted in the following content; that is, the symbol y will be used instead. q，m H q，m s q，m , l q，m n q，m S q and L q Remove the subscript q and abbreviate it to the corresponding symbol y. m H m s m , l m n m , S and L.

[0013] S3. Receiver parameter initialization: in This represents the prior information of the target symbol S in module A. for The prior variance, which is an N s A vector of ×1 represents the different variances corresponding to the prior information of different user target symbols within the same sub-block. This represents the prior information of the interference signal L in module A. Prior value The prior variance is a scalar, meaning that different interference signal estimates within the same sub-block use the same variance value. The energy intensity of the interference signal. In subsequent steps, the abbreviation "pri" for "prior" indicates prior information, the abbreviation "post" for "posterior" indicates posterior information, and the abbreviation "ext" for "extrinsic" indicates extrinsic information. Different superscripts for the same symbol are used to distinguish the information category to which it belongs. The subscripts A, B, and C indicate the modules A, B, and C to which the corresponding symbols belong, respectively.

[0014] S4, LMMSE linear estimation module A: Based on the block model in step S2, it performs LMMSE estimation for the target symbol and interference signal in each column of the sub-block. When the base station uses a planar antenna array for reception, the interference signal l m The covariance matrix exhibits off-diagonal properties but manifests as a specific pattern, which is represented by the normalized disturbance covariance matrix. express.

[0015] Let the residual received signal of module A be... Its covariance matrix is,

[0016]

[0017] in This represents a diagonal matrix with vector x as its diagonal elements, where all off-diagonal elements are zero. Let be the white noise energy intensity, and I be the identity matrix. Target symbol s m The posterior estimate is,

[0018]

[0019] Where diag(X) represents taking the diagonal elements of matrix X as column vectors. Interference signal l m The posterior estimate is,

[0020]

[0021] The function trace(X) calculates the trace of matrix X.

[0022] S5. Calculation of external information for module A: First, average the posterior variances across different columns within the sub-block.

[0023]

[0024] The formula for calculating external information is:

[0025]

[0026] Where ⊙ represents the Hadamard product. Then, the external information of the target symbol and the interference signal is input into modules B and C respectively.

[0027] S6. Soft demodulation and noise reduction module B: In module B, the constellation diagram information of the target symbols is used to estimate the value. Noise reduction, its n rows and m columns are Its variance is Right now The nth element. Assume the target user uses 2. J In QAM modulation, one constellation point represents J bits, and the symbol... The probability of corresponding to the kth constellation point for,

[0028]

[0029] Where c k Let represent the k-th constellation point. The posterior mean and variance are respectively,

[0030]

[0031] Where |x| represents finding the modulus of the complex number x.

[0032] S7. Module B External Information Calculation: First, calculate the average of the posterior variances of different users within a sub-block.

[0033]

[0034] The posterior variance is The posterior means are concatenated into a matrix form. The element in its nth row m is The formula for calculating external information in module B is as follows:

[0035]

[0036] Where ⊙ represents the Hadamard product. Then, the external information mean and variance of module B are returned to module A.

[0037] S8, Interference Low-Rank Noise Reduction Module C: Assuming the channel is flat within a sub-block, i.e., the channel on different carriers within the sub-block... The interference signal is the same, L = [l1, ..., l2]. M ], l m =H I，m s I，m The rank is N i However, due to the existence of channel frequency selectivity, the channel on different carriers has small variations. After performing Singular Value Decomposition (SVD) on the interference signal, except for the first N... iBesides the larger singular values, there are also some smaller singular values. During noise reduction, directly setting these smaller singular values ​​to zero will result in the loss of some useful information; completely absorbing them will introduce excessive background noise. This invention uses Optimal Shrinker to process singular values, and the method for calculating the posterior information of the interference is as follows:

[0038] S81, First, the input of module C. Perform SVD decomposition.

[0039]

[0040] in Singular values ​​σ in i i = 1, 2, ..., N r Arranged in descending order, that is Normalize the singular values ​​to,

[0041]

[0042] in Let V be the input variance of module C.

[0043] S82, Denoising the singular values ​​yields...

[0044]

[0045] Where β = N r / M. Further processor singular values ​​are,

[0046]

[0047] Where x = x(σ) i The resulting singular value matrix is ​​denoted as...

[0048] S83, the posterior mean of the interference in the reconstruction module C is in This represents a low-rank matrix noise reducer.

[0049] S9. Calculation of extrinsic information for module C: The mean of extrinsic information for module C can be represented as a combination of prior and posterior information. The formula for calculating the linear parameter is as follows:

[0050]

[0051] Where div(X) represents the divergence of matrix X.<X,Y> This represents calculating the inner product of matrices. The method for calculating the matrix divergence div(·) above involves first adding white noise N of intensity ε to the prior information. div get Noise reduction The calculated divergence is

[0052] The external information variance calculation method is as follows Among them ||·|| F Let F be the matrix norm. Finally, the mean and variance of the external information from module C are input into module A for iteration.

[0053] S10. If the algorithm converges or reaches the preset maximum number of iterations, the process ends; otherwise, proceed to step S4.

[0054] The beneficial effects of this invention are as follows: It utilizes the constellation diagram information of the target symbol and the low-rank nature of the interference signal to denoise the estimated value of the linear module. Through iterative processing of different modules within the algorithm, the interference can be gradually eliminated. More specifically, a special linear estimation module and a low-rank denoising module are designed specifically for the characteristics of the interference signal, making the conditional assumptions in the receiver algorithm closer to reality, effectively achieving interference elimination, and significantly improving data demodulation performance. Attached Figure Description

[0055] Figure 1 This is a block diagram of the receiver structure of the present invention;

[0056] Figure 2 It is a mesh plot of the normalized disturbance covariance matrix;

[0057] Figure 3 It is a heatmap of the normalized interference covariance matrix;

[0058] Figure 4 This is a comparison chart of singular values ​​of the interference signal matrix;

[0059] Figure 5 This is a graph showing the BLER demodulation performance of the algorithm under 16QAM;

[0060] Figure 6 This is a graph showing the performance of the BLER demodulation algorithm under 64QAM. Detailed Implementation

[0061] The present invention will now be described in further detail with reference to the accompanying drawings. A block diagram of the receiver structure of the present invention is shown below. Figure 1 As shown, the algorithm includes a linear estimation module A, a soft demodulation and noise reduction module B, and a low-rank noise reduction module C. The simulation channel is the Urban Macro (UMa) scenario channel described in Section 7.5 of the 3GPP 38.901 standard, and the parameter settings refer to the UMA scenario NLOS column in Table 7.5-6 Part-1 of that standard. In the simulation, the number of base station receiving antennas ε and the number of target user flows N are... s =8, Number of sudden interference flows N i=8, burst interference for IoT is 10dB / stream, the total number of OFDM data symbols in one TTI is T=12, and the number of data symbols affected by burst interference is T i =2, system bandwidth is N RB =24 RBs, a total of K=288 subcarriers, R=4 RBs form a sub-block, the system is divided into Q=6 sub-blocks, the modulation scheme is 16QAM or 64QAM, the channel coding adopts LDPC coding, the coding code rate is 3 / 4, and all OFDM data symbols in a single TTI are encoded into a codeword. In addition, since the location of burst interference is known, data symbols not affected by burst interference are directly demodulated using the MMSE-IRC algorithm, while data symbols affected by burst interference are demodulated using the algorithm designed in this invention (named C-TMP-Opt), with the maximum number of iterations set to 5 or 10. The specific implementation under the current configuration is as follows:

[0062] S1. System Modeling: Within a frame, two OFDM data symbols are affected by burst interference, and the location of the interference is known (assuming the interference does not occur on the pilot symbol). The total frequency domain bandwidth of the system is K = 12N. RB =288 subcarriers, N RB =24 represents the number of resource blocks (RBs). Let N be the number of receiving antennas at the base station. r =64, target signal stream number is N s =8, the number of interference signal streams is N i =8, the base station received signal on the k-th subcarrier of the t-th data symbol can be represented as,

[0063]

[0064] in For the target user channel, To interfere with user channels, For target signal, This is an interference signal. It is white noise.

[0065] S2. Block Model: Considering that the channel changes of adjacent subcarriers in the frequency domain are relatively slow, adjacent R = 4 RBs are regarded as a sub-block, so the system is divided into q = N blocks. RB / R = 6 sub-blocks, and T can be... i = Two identical sub-blocks of OFDM symbols are concatenated together, that is, each sub-block has a dimension of 64×96, and the received signal in the m-th column of the q-th sub-block is,

[0066] y q,m =H q,m s q，m +H I，q，m sI,q,m +n q，m

[0067] =H q，m s q，m +l q，m +n q，m

[0068] Where the subscripts q and m correspond to the 48q+mod(m-1,48)+1th subcarrier on the ceil(m / 48)th OFDM data symbol subjected to burst interference, ceil(x) represents the floor function of x, and mod(x,y) represents the modulo value of x. q，m =H I，q，m s I，q，m This represents the interference signal. The target symbol s... q，m and interference signal l q，m Represented in matrix form S q =[s q，1 , ..., s q，M ] and L q =[l q，1 , ..., l q，M In subsequent steps, each sub-block is processed independently with identical processing steps. For the sake of simplicity in notation, the sub-block subscript q will be omitted in the following content; that is, the symbol y will be used instead. q，m H q，m s q，m , l q，m n q，m S q and L q Remove the subscript q and abbreviate it to the corresponding symbol y. m H m s m , l m n m , S and L.

[0069] S3. Receiver parameter initialization: in This represents the prior information of the target symbol S in module A. for The prior variance is an 8×1 vector, meaning that the prior information of different user target symbols within the same sub-block corresponds to different variances. This represents the prior information of the interference signal L in module A. Prior value The prior variance is a scalar, meaning that different interference signal estimates within the same sub-block use the same variance value. The energy intensity of the interference signal. In subsequent steps, the abbreviation "pri" for "prior" indicates prior information, the abbreviation "post" for "posterior" indicates posterior information, and the abbreviation "ext" for "extrinsic" indicates extrinsic information. Different superscripts for the same symbol are used to distinguish its information category. The subscripts A, B, and C represent the modules A, B, and C where the corresponding symbol is located, respectively.

[0070] S4, LMMSE linear estimation module A: Based on the block model in step S2, it performs LMMSE estimation for the target symbol and interference signal in each column of the sub-block. When the base station uses a planar antenna array for reception, the interference signal l m The covariance matrix exhibits off-diagonal properties but manifests as a specific pattern, which is represented by the normalized disturbance covariance matrix. Representation. Matrix The value can be obtained by statistically averaging the P interference signals received in the base station's history, and the matrix of the received p-th interference signal is denoted as L. p First, calculate the statistical average of historical interference. The normalized covariance matrix is Where `diag(·)` extracts the diagonal elements of the matrix, and `mean(·)` extracts the average. The matrix obtained by statistical averaging is as follows: Figure 1 (mesh diagram) and Figure 2 As shown in the heatmap.

[0071] Let the residual received signal of module A be... Its covariance matrix is,

[0072]

[0073] in This represents a diagonal matrix with vector x as its diagonal elements, where all off-diagonal elements are zero. Let be the white noise energy intensity, and I be the identity matrix. Target symbol s m The posterior estimate is,

[0074]

[0075] Where diag(X) represents taking the diagonal elements of matrix X as column vectors. Interference signal l m The posterior estimate is,

[0076]

[0077] The function trace(X) calculates the trace of matrix X.

[0078] S5. Calculation of external information for module A: First, average the posterior variances across different columns within the sub-block.

[0079]

[0080] The formula for calculating external information is:

[0081]

[0082] Where ⊙ represents the Hadamard product. Then, the external information of the target symbol and the interference signal is input into modules B and C respectively.

[0083] S6. Soft demodulation and noise reduction module B: In module B, the constellation diagram information of the target symbols is used to estimate the value. Noise reduction, its n rows and m columns are

[0084] Its variance is Right now The nth element. The target user uses 2. J =64-order QAM modulation, one constellation represents J = 6 bits, The probability of corresponding to the kth constellation point for,

[0085]

[0086] Where c k Let represent the k-th constellation point. The posterior mean and variance are respectively,

[0087]

[0088] Where |x| represents finding the modulus of the complex number x.

[0089] S7. Module B External Information Calculation: First, calculate the average of the posterior variances of different users within a sub-block.

[0090]

[0091] The posterior variance is The posterior means are concatenated into a matrix form. The element in its nth row m is The formula for calculating external information in module B is as follows:

[0092]

[0093] Where ⊙ represents the Hadamard product. Then, the external information mean and variance of module B are returned to module A.

[0094] S8, Interference Low-Rank Noise Reduction Module C: Assuming the channel is flat within a sub-block, i.e., the channel on different carriers within the sub-block... The interference signal is the same, L = [l1,...,l 96],l m =H I,m s I,m The rank is N i =8, but due to the existence of channel frequency selectivity, the channel on different carriers has small variations. After performing Singular Value Decomposition (SVD) on the interference signal, except for the first N... i =In addition to the eight larger singular values, there are also some relatively smaller singular values, such as Figure 4 The 8th to 16th singular values ​​are shown in the diagram. During noise reduction, directly setting the smaller leaked singular values ​​to zero will result in the loss of some useful information; conversely, completely absorbing them will introduce excessive background noise. This invention uses an Optimal Shrinker to process singular values, and the method for calculating the posterior information of the interference is as follows:

[0095] S81, First, the input of module C. Perform SVD decomposition.

[0096] ∑=diag{σ1,σ2,...,σ 64},

[0097] in Singular values ​​σ in i The numbers i = 1, 2, ..., 64 are arranged in descending order, i.e., σ1 > σ2 > ... > σ 64 Normalize the singular values ​​to,

[0098]

[0099] in Let V be the input variance of module C.

[0100] S82, Denoising the singular values ​​yields...

[0101]

[0102] Where β = N r / M = 64 / 96 = 2 / 3. Further, the processor's singular values ​​are...

[0103]

[0104] Where x = x(σ) i The processed singular value matrix is ​​denoted as ∑. opt =diag{η(σ1), ...,η(σ)} 64 OptimalShrinker can handle some leaked singularities. Figure 4 The 8th to 16th smaller singular values ​​in the middle achieve partial recovery.

[0105] S83, the posterior mean of the interference in the reconstruction module C is in This represents a low-rank matrix noise reducer.

[0106] S9. Calculation of extrinsic information for module C: The mean of extrinsic information for module C can be represented as a combination of prior and posterior information. The formula for calculating the linear parameter is as follows:

[0107]

[0108] Where div(X) represents the divergence of matrix X.<X,Y> This represents calculating the inner product of matrices. The method for calculating the matrix divergence div(·) above involves first adding a strength of ε = 10 to the prior information. -3 White noise N div get Noise reduction Calculate divergence

[0109] The external information variance calculation method is as follows Among them ||·|| F Let F be the matrix norm. Finally, input the mean and variance of the external information from module C into module A for iteration. S10: If the algorithm converges or reaches the maximum number of iterations (5 or 10), the process ends; otherwise, proceed to S4.

[0110] Figure 5 and Figure 6 The BLER performance curves of different algorithms for data demodulation under 16QAM and 64QAM modulation are shown respectively. The horizontal axis represents the strength of the target user signal relative to the noise floor, and the vertical axis represents the BLER of data demodulation. The curve Interference-free corresponds to the case where no data symbols are subjected to sudden interference. The curve MMSE-IRC-Bound corresponds to a performance bound of the MMSE-IRC algorithm. The interference covariance matrix is ​​obtained by averaging the interference signals on all subcarriers within a single OFDM symbol subblock (in reality, the interference signals are unknown, and it is only used as a reference performance bound). The curve C-TMP-Opt-10 corresponds to the result of 10 iterations of the C-TMP-Opt algorithm in this invention. The curve C-TMP-Opt-5 is the result of 5 iterations of the algorithm. The curve LMMSE corresponds to the case where sudden interference is treated as white noise.

[0111] When BLER = 0.1, the performance of the C-TMP-Opt algorithm after 10 and 5 iterations is basically the same under 16QAM modulation, with a difference of about 1.4 dB compared to Interference-free. Under 64QAM modulation, the difference between the C-TMP-Opt algorithm after 10 and 5 iterations is still small, slightly higher than that under 16QAM modulation, with a distance of 1.8 dB from Interference-free at 5 iterations. As the QAM modulation order increases, the increased constellation density weakens the noise reduction capability of module B, resulting in a decrease in the performance of the C-TMP-Opt algorithm. Therefore, the difference between it and Interference-free is relatively larger, while the difference with LMMSE is reduced. In addition, it is noted that as the QAM modulation order increases, the gap between the interference-free case and the LMMSE algorithm gradually decreases. This phenomenon is because the overall error rate within the codeword is lower when only two symbols are interfered with, and all symbols within a TTI are encoded into a single codeword. Under higher-order modulation, the code length is much larger than that under lower-order modulation, which can better achieve error correction.

[0112] In summary, the C-TMP-Opt algorithm proposed in this invention has significant gains under high-order modulation (64QAM), achieving a 3dB gain compared to the LMMSE baseline at BLER=0.1, and a loss of less than 2dB compared to the interference-free case; the algorithm gains are even more significant under low-order modulation (16QAM), with a 6dB gain compared to the LMMSE baseline and a loss of only 1.4dB compared to the interference-free case.

Claims

1. A method for target signal estimation and demodulation in sudden interference scenarios based on message passing, characterized in that, Includes the following steps: S1, System Modeling: There are T systems within a single frame. i Several OFDM data symbols are affected by sudden interference, and the location of the interference is known. The total frequency domain bandwidth of the system is K = 12N. RB N subcarriers RB Let N be the number of Resource Blocks (RBs). r The number of target signal streams is N s The number of interference signal streams is N i The signal received by the base station on the k-th subcarrier of the t-th data symbol can be represented as follows: in For the target user channel, To interfere with user channels, For target signal, This is an interference signal. It is white noise; S2. Block Model: Considering that the channel changes of adjacent subcarriers in the frequency domain are relatively slow, adjacent R RBs are regarded as a sub-block, then the system is divided into Q = N blocks. RB / R sub-blocks, while T can be i Identical sub-blocks of OFDM symbols are concatenated together, meaning each sub-block has a dimension of N. r ×M, where M=KT i / Q, the received signal in the m-th column within the q-th sub-block is... y q,m =H q，m s q，m +H I，q,m s I，q，m +n q,m =H q，m s q,m +l q，m +n q，m Where the subscripts q and m correspond to the qK / Q+mod(m-1, K / Q)+1th subcarrier on the ceil(mQ / K)th OFDM data symbol subjected to burst interference, ceil(x) represents the floor function of x, and mod(x,y) represents the modulo value of x with respect to y. q,m =H I,q, m s I,q,m Indicating interference signal, target symbol s q,m and interference signal l q,m Represented in matrix form S q =[s q,1 ,...,s q,M ] and L q =[l q,1 ,...,l q,M In subsequent steps, each sub-block is processed independently with identical processing steps. For the sake of simplicity in notation, the sub-block subscript q will be omitted in the following content; that is, the symbol y will be used instead. q,m H q,m ,s q,m ,l q,m ,n q,m ,S q and L q Remove the subscript q and abbreviate it to the corresponding symbol y. m H m s m , l m n m S and L; S3. Receiver parameter initialization: in This represents the prior information of the target symbol S in module A. for The prior variance, which is an N s A vector of ×1 represents the different variances corresponding to the prior information of different user target symbols within the same sub-block. This represents the prior information of the interference signal L in module A. Prior value The prior variance is a scalar, meaning that different interference signal estimates within the same sub-block use the same variance value. To determine the energy intensity of the interference signal, the superscript "prior" is used in subsequent steps to represent prior information, the superscript "posterior" is used to represent posterior information, and the superscript "extrinsic" is used to represent extrinsic information. Different superscripts for the same symbol are used to distinguish the information category to which it belongs. The subscripts A, B, and C represent the modules A, B, and C where the corresponding symbol is located, respectively. S4, LMMSE linear estimation module A: Based on the block model in step S2, it performs LMMSE estimation for the target symbol and interference signal in each column of the sub-block. When the base station uses a planar antenna array for reception, the interference signal l m The covariance matrix exhibits off-diagonal properties but manifests as a specific pattern, which is represented by the normalized disturbance covariance matrix. express; Let the residual received signal of module A be... Its covariance matrix is, in This represents a diagonal matrix with vector x as its diagonal elements, where all off-diagonal elements are zero. Let I be the white noise energy intensity, I be the identity matrix, and s be the target symbol. m The posterior estimate is, Where diag(X) represents taking the diagonal elements of matrix X as column vectors, and the interference signal l m The posterior estimate is, Where trace(X) represents finding the trace of matrix X; S5. Calculation of external information for module A: First, average the posterior variances across different columns within the sub-block. The formula for calculating external information is: Where ⊙ is the Hadamard product, and then the external information of the target symbol and the interference signal are respectively input into modules B and C; S6. Soft demodulation and noise reduction module B: In module B, the constellation diagram information of the target symbols is used to estimate the value. Noise reduction, its n rows and m columns are n = 1, ..., N s , m = 1, ..., M, and its variance is Right now The nth element, assuming the target user uses 2 J In QAM modulation, one constellation point represents J bits, and the symbol... The probability of corresponding to the kth constellation point for, Where c k Let the k-th constellation point be represented by its posterior mean and variance, respectively. Where |x| represents finding the modulus of the complex number x; S7. Module B External Information Calculation: First, calculate the average of the posterior variances of different users within a sub-block. The posterior variance is The posterior means are concatenated into a matrix form. The element in its nth row m is The formula for calculating external information in module B is as follows: Where ⊙ is the Hadamard product, the mean and variance of the external information of module B are then returned to module A; S8, Interference Low-Rank Noise Reduction Module C: Assuming the channel is flat within a sub-block, i.e., the channel on different carriers within the sub-block... The interference signal is the same, L = [l1, ..., l2]. M ], l m =H I，m s I，m The rank is N i However, due to the existence of channel frequency selectivity, the channel on different carriers has small variations. After performing Singular Value Decomposition (SVD) on the interference signal, except for the first N... i Besides the larger singular values, there are also some smaller singular values. During noise reduction, directly setting these smaller singular values ​​to zero will result in the loss of some useful information; conversely, completely absorbing them will introduce excessive background noise. Therefore, Optimal Shrinker is used to process the singular values. The method for calculating the posterior information of the interference is as follows. S81, First, the input of module C. Perform SVD decomposition. in Singular values ​​σ in i i = 1, 2, ..., N r Arranged in descending order, that is Normalize the singular values ​​to, in Let Variance be the input variance of module C; S82, Denoising the singular values ​​yields... Where β = N r / M, further processor singular values, Where x = x(σ) i The resulting singular value matrix is ​​denoted as . S83, the posterior mean of the interference in the reconstruction module C is in This represents a low-rank matrix noise reducer. S9. Calculation of extrinsic information for module C: The mean of extrinsic information for module C can be represented as a combination of prior and posterior information. The formula for calculating the linear parameter is as follows: Where div(X) represents the divergence of matrix X.<X,Y> This represents calculating the inner product of matrices. The method for calculating the matrix divergence div(·) above involves first adding white noise N of intensity ε to the prior information. div get Noise reduction The calculated divergence is The external information variance calculation method is as follows Among them ||·|| F The F-norm of the matrix is ​​used. Finally, the mean and variance of the external information from module C are input into module A for iteration. S10. If the algorithm converges or reaches the preset maximum number of iterations, the process ends; otherwise, proceed to step S4.

2. The target signal estimation and demodulation method based on message passing in a sudden interference scenario according to claim 1, characterized in that: In step S1, it is assumed that the interference does not occur on the pilot symbol.

Images