Iterative signal detection method and system in a priori missing scenario

Abstract

The present invention discloses an iterative signal detection method and system in a prior missing scenario, and relates to the technical field of wireless communications. The method comprises the following steps: acquiring a communication signal in a multi-input multi-output system; extracting an initial estimate of the communication signal in a prior distribution scenario based on the communication signal; iteratively refining the communication signal and its initial estimate in the prior distribution scenario to obtain a final estimate of the communication signal and the prior distribution, and completing iterative signal detection. The method and system proposed by the present invention can effectively model a single probability distribution model, and the model is suitable for dense prior scenarios, the probability distribution range of the model is wider, and the accuracy of signal detection is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communications, and in particular to an iterative signal detection method and system in a priori missing scenario. Background Art

[0002] Modern communication systems often use Massive Multiple-Input Multiple-Output (Massive MIMO) technology and high-order Quadrature Amplitude Modulation (QAM) constellation diagrams to increase data transmission rate and improve signal communication quality. In MIMO systems, the design of the receiver is crucial because it is responsible for dealing with channel noise and interference between signals, and recovering the original data from the received signal, which usually involves complex signal processing algorithms. Receivers based on the Approximate Message Passing (AMP) algorithm have been proven to be very effective in massive MIMO detection. The AMP algorithm belongs to the category of Bayesian algorithms. It regards unknown parameters as random variables, introduces prior distributions to estimate their specific realizations, and iteratively solves and updates the posterior distribution of the signal, which has great advantages in solving speed and result accuracy. Although the AMP algorithm is very attractive in theory, in practice, it is rarely possible to know the prior distribution of the signal at this time. The prior distribution of the signal is usually related to the modulation method. For example, in the adaptive modulation scenario, the signal modulation method can be dynamically adjusted according to the real-time conditions of the communication channel to optimize the transmission performance of the signal, thereby maximizing the transmission rate while ensuring the transmission quality. This is an important technical means to improve the performance of wireless communication systems. Therefore, developing a signal detection method in the scenario of lack of prior is a difficult problem to be solved in the current communication field.

[0003] Existing technical solutions often use the Least Squares (LS) estimator or the Linear Minimum Mean Square Error (LMMSE) estimator based on Maximum Likelihood Estimation (MLE). The advantages are that the closed-form solution is easy to obtain, the principle is simple and easy to implement, but because the potential structure of the signal is ignored, their performance cannot be comparable to the Bayesian algorithm. The method based on the combination of Expectation Maximization (EM) of Empirical Bayes (EM) and AMP models the prior distribution of the signal as a Bernoulli-Gaussian mixture model with undetermined parameters. It can restore the signal according to the learned distribution and learn the distribution of the signal again when performing the noise reduction link of AMP. Although EM-AMP has shown excellent performance in terms of reconstruction error and running time of high-dimensional signals, its modeling range is limited. For example, the probability distribution of QAM constellation point pulse signals of different orders cannot be effectively modeled as a single probability distribution model. At the same time, it is not suitable for dense prior scenarios, resulting in low signal detection accuracy. Summary of the invention

[0004] In order to overcome the problem that the existing wireless communication technology cannot effectively model a single probability distribution model, and the existing model is not suitable for dense prior scenarios, resulting in low signal detection accuracy, the present invention provides an iterative signal detection method and system in a prior missing scenario, which can effectively model a single probability distribution model, and the model is suitable for dense prior scenarios, effectively improving the accuracy of signal detection.

[0005] To achieve the purpose of the present invention, the present invention adopts the following technical solutions:

[0006] A method for iterative signal detection in a priori missing scenario, the method comprising the following steps:

[0007] Acquiring communication signals in a multiple-input multiple-output system;

[0008] Based on the communication signal, extracting an initial estimate of the communication signal in a prior distribution scenario;

[0009] The communication signal and its initial estimate in the prior distribution scenario are iteratively refined to obtain a final estimate of the communication signal and the prior distribution, thereby completing iterative signal detection.

[0010] In the above technical scheme, by extracting the initial estimate of the acquired communication signal under the prior distribution scenario, the modulation mode of the signal can be further inferred, so as to iteratively refine the communication signal and its initial estimate under the prior distribution scenario, thereby improving the dimension of the signal and effectively improving the accuracy of signal estimation in dense prior scenarios.

[0011] Furthermore, before extracting the initial estimate of the signal in the prior distribution scenario, the following steps are also included:

[0012] Construct a multi-input and multi-output system model, the expression is:

[0013] y=Hx+w;

[0014] The model includes a linear equalization module, an iterative noise reduction module and a priori estimation module;

[0015] The linear equalization module is used to perform linear equalization processing on the communication signal and extract an initial estimate of the communication signal in a priori distribution scenario;

[0016] The iterative noise reduction module is used to perform iterative noise reduction processing on the initial estimate of the communication signal in the prior distribution scenario, and the prior estimation module is used to refine the initial estimate after each iterative noise reduction processing and output a final estimate of the communication signal and the prior distribution;

[0017] Among them, x represents the communication signal, H represents the channel matrix, w represents the superimposed Gaussian white noise, and y represents the received signal.

[0018] In the above technical solution, the constructed multi-input and multi-output system model can directly cooperate with each other through the linear equalization module, the iterative noise reduction module and the prior estimation module, which is effectively applicable to dense prior scenarios and improves the accuracy of signal detection. At the same time, the probability distribution range of the model is wider.

[0019] Furthermore, the process of extracting an initial estimate of the communication signal in a prior distribution scenario includes:

[0020] The linear equalization module uses a linear equalizer to extract an initial estimate of the communication signal;

[0021] Based on the extracted initial estimate of the communication signal, the linear equalization module extracts an initial estimate of the communication signal in a prior distribution scenario.

[0022] Furthermore, the linear equalization module uses zero-forcing equalization to extract the initial estimate of the communication signal, which is expressed as:

[0023]

[0024] Among them, H *represents the matrix obtained by taking the conjugate transpose of the channel matrix H.

[0025] Furthermore, the linear equalization module uses the minimum mean square error equalization method to extract the initial estimate of the communication signal, which is expressed as:

[0026]

[0027] Among them, σ 2 represents the noise variance of the channel, I represents a vector whose elements are all 1, and H * represents the matrix obtained by taking the conjugate transpose of the channel matrix H.

[0028] Furthermore, the linear equalization module uses a binning modeling method to extract the initial estimate of the communication signal in the prior distribution scenario. The process includes:

[0029] Find vector Minimum value of and maximum value Defining the right spacing And create an isometric grid set, the expression is:

[0030]

[0031] Set The length is R, then the expression is:

[0032] Then set the vector of length R in Is from vector Elements Falling on The number of samples in the pane area centered is expressed as:

[0033]

[0034] Count vector The probability vector f is followed on R categories ZF , a multinomial distribution with N trials, is expressed as:

[0035]

[0036] Under the zero-forcing equilibrium method, f ZF The initial estimate under the prior distribution scenario is expressed as:

[0037]

[0038] Based on the initial estimate in the prior distribution scenario obtained under the zero-forcing equalization method, we get f under the minimum mean square error equalization method. MMSEThe initial estimate under the prior distribution scenario is expressed as:

[0039]

[0040] in, represents the initial estimate of the communication signal extracted based on the zero-forcing equalization method in the prior distribution scenario, It represents the initial estimate of the communication signal extracted based on the minimum mean square error equalization method in the prior distribution scenario, and N represents the signal length. represents the initial estimate under the zero-forcing equilibrium method, Represents the initial estimate under the minimum mean square error equalization method.

[0041] In the above technical solution, the acquired communication signal is subjected to initial estimation extraction in a priori distribution scenario, and the modulation mode of the signal can be further inferred.

[0042] Furthermore, the process of iteratively refining the communication signal and its initial estimate in the prior distribution scenario includes:

[0043] The communication signal and its initial estimate in the prior distribution scenario are input into the iterative noise reduction module and the prior estimation module;

[0044] The iterative denoising module iteratively denoises the communication signal and its initial estimate in the prior distribution scenario, and the prior estimation module deconvolves the denoised communication signal and its initial estimate in the prior distribution scenario, and iteratively learns a more accurate estimate of the signal prior distribution.

[0045] Furthermore, the iterative denoising module performs several rounds of iterative denoising on the communication signal and its initial estimate in the prior distribution scenario; the prior estimation module uses the empirical Bayesian deconvolution method to perform a priori estimation on the denoised communication signal, and the expression is:

[0046]

[0047] in, Both represent intermediate variables of linear operations in the iteration process. Subscript a represents the ath element of a vector of length M, and subscript i represents the ith element of a vector of length N. ai represents the element in the ath row and ith column of the matrix H, t represents the number of iterations, and the functions η(·) and θ(·) mean to take the posterior distribution The mean and variance of and They represent the posterior distribution The mean and variance of Indicates that a priori estimation of the degrees of freedom dof is performed based on Bayesian deconvolution.

[0048] A loss function is set to optimize the multi-input multi-output system model in the iterative refinement process, and when the loss function converges or when a preset iteration round ends, a multi-input multi-output system model capable of iterative signal detection in a priori missing scenario is obtained;

[0049] Model outputs of MIMO systems and obtain final estimates of the signals and the final estimate of the prior distribution p T (x).

[0050] Furthermore, the process of the prior estimation module using the empirical Bayesian deconvolution method to perform a priori estimation on the noise-reduced communication signal includes:

[0051] The scalar state evolution equation in a large dimensional system is expressed as:

[0052]

[0053] Assume the convolution expression is:

[0054]

[0055] The imaginary part and the real part of the signal in the deconvolution process are processed separately. The specific process is:

[0056] A: Given a fixed interval constellation point value space

[0057] B: Yes Perform binning modeling to obtain the observation space Count vector The probability vector for S categories is For a multinomial distribution with N trials, the expression is:

[0058]

[0059] C: Prior estimation module Assume g(x) is The exponential family distribution on is expressed by the discrete formula:

[0060]

[0061] Construct a transfer matrix P with a dimension of S×L. The matrix elements are the transition probabilities, and the expression is:

[0062]

[0063] The convolution relationship can be simplified to matrix multiplication make It can be easily expressed as a function of α;

[0064] get The log-likelihood function of is:

[0065]

[0066] in definition:

[0067]

[0068] Let W s is an L-dimensional vector, that is, W s (α)=[w s1 (α),w s2 (α),...,w sL (α)] T ;

[0069] The scoring function for α is:

[0070]

[0071] in,

[0072] Maximum likelihood estimate of α By numerically maximizing l(α) or solving the matrix equation Get, substitute back Get an estimate of the prior distribution p(x);

[0073] Among them, Q represents the structure matrix with dimension L×dof, using natural spline basis function generate, represents the j-th row vector of Q; dof represents the dimension of the unknown vector α; 1 represents a full-1 vector of length L, and 1v(α) is used to ensure probability normalization.

[0074] In the above technical scheme, the iterative denoising module is combined with the prior estimation module to iteratively denoise the communication signal and its initial estimate in the prior distribution scenario and iteratively learn a more accurate estimate of the signal prior distribution, and set a loss function and several rounds of iterations to iteratively optimize the multi-input multi-output system model to obtain a multi-input multi-output system model that can perform iterative signal detection in the scenario where prior is missing, so that the model can be effectively applied to the scenario of dense prior and improve the accuracy of signal detection. At the same time, the probability distribution range of the model is wider.

[0075] An iterative signal detection system in a priori missing scenario, the system comprising:

[0076] An acquisition module, used for acquiring communication signals in a multiple-input multiple-output system;

[0077] A processing module, configured to extract an initial estimate of the communication signal in a prior distribution scenario based on the communication signal;

[0078] The iterative signal detection module is used to iteratively refine the communication signal and its initial estimate in the prior distribution scenario, obtain the final estimate of the communication signal and the prior distribution, and complete the iterative signal detection.

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] The present invention provides an iterative signal detection method and system in a prior-missing scenario. Firstly, by extracting the initial estimate of the acquired communication signal in the prior distribution scenario, the modulation mode of the signal can be further inferred, so as to iteratively refine the communication signal and its initial estimate in the prior distribution scenario, thereby improving the dimension of the signal, and then effectively improving the accuracy of signal estimation in dense prior scenarios; secondly, the constructed multi-input and multi-output system model can be directly coordinated with each other through a linear equalization module, an iterative noise reduction module and a prior estimation module, so as to be effectively applicable to dense prior scenarios and improve the accuracy of signal detection. At the same time, the probability distribution range of the model is wider. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 A flowchart of the steps of an iterative signal detection method in a priori missing scenario provided by an embodiment of the present application;

[0082] Figure 2 A schematic diagram of the multi-input multi-output system model structure provided in an embodiment of the present application;

[0083] Figure 3 The convergence of the EB-AMP algorithm under the condition of the transmission signal length N=512 provided in the embodiment of the present application;

[0084] Figure 4 A Gaussian verification diagram of the state evolution equation of the ZF-initialized EB-AMP algorithm provided in an embodiment of the present application during the second iteration;

[0085] Figure 5 Performance comparison diagram of Oracle-AMP, EB-AMP, LS and LMMSE methods under different signal-to-noise ratios with a transmission signal length N=256 provided in an embodiment of the present application;

[0086] Figure 6 A performance comparison chart of the Oracle-AMP, EB-AMP, LS and LMMSE methods under different signal-to-noise ratios with a transmission signal length N=2048 provided in an embodiment of the present application;

[0087] Figure 7A diagram showing the recovery of a priori distribution of a signal provided in an embodiment of the present application under the condition that the length of the transmitted signal N=512;

[0088] Figure 8 A schematic diagram of the structure of an iterative signal detection system in a priori missing scenario provided in an embodiment of the present application. DETAILED DESCRIPTION

[0089] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Preferred embodiments of the present invention are provided in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.

[0090] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0091] Embodiment 1:

[0092] This embodiment provides an iterative signal detection method in a scenario where a priori is missing. Figure 1 , the method comprises the following steps:

[0093] Step S1: Acquire a communication signal in a multiple-input multiple-output system;

[0094] Step S2: extracting an initial estimate of the communication signal in a prior distribution scenario based on the communication signal;

[0095] Step S3: Iteratively refine the communication signal and its initial estimate in the prior distribution scenario to obtain a final estimate of the communication signal and the prior distribution, thereby completing iterative signal detection.

[0096] It can be understood that by extracting the initial estimate of the acquired communication signal under the prior distribution scenario, the modulation mode of the signal can be further inferred, so as to iteratively refine the communication signal and its initial estimate under the prior distribution scenario, thereby improving the dimension of the signal and effectively improving the accuracy of signal estimation in dense prior scenarios.

[0097] As a preferred embodiment, before extracting the initial estimation of the signal in the prior distribution scenario, the following steps are also included:

[0098] Construct a multi-input and multi-output system model, the expression is:

[0099] y=Hx+w;

[0100] See also Figure 2 , the model includes a linear equalization module, an iterative noise reduction module and a priori estimation module;

[0101] The linear equalization module is used to perform linear equalization processing on the communication signal and extract an initial estimate of the communication signal in a priori distribution scenario;

[0102] The iterative noise reduction module is used to perform iterative noise reduction processing on the initial estimate of the communication signal in the prior distribution scenario, and the prior estimation module is used to refine the initial estimate after each iterative noise reduction processing and output a final estimate of the communication signal and the prior distribution;

[0103] Among them, x represents the communication signal, H represents the channel matrix, w represents the superimposed Gaussian white noise, and y represents the received signal.

[0104] Specifically, in a massive MIMO communication system, data is transmitted Through the channel matrix Then, Gaussian white noise is superimposed Get the received signal This process is abstracted into a mathematical model (multi-input multi-output system model) and expressed as: y=Hx+w, where the transmitted signal x is generated through high-order QAM modulation.

[0105] It can be understood that the constructed multi-input and multi-output system model can directly cooperate with each other through the linear equalization module, the iterative noise reduction module and the prior estimation module, which is effectively applicable to dense prior scenarios and improves the accuracy of signal detection. At the same time, the probability distribution range of the model is wider.

[0106] In step S2, the process of extracting an initial estimate of the communication signal in a prior distribution scenario includes:

[0107] The linear equalization module uses a linear equalizer to extract an initial estimate of the communication signal;

[0108] Based on the extracted initial estimate of the communication signal, the linear equalization module extracts an initial estimate of the communication signal in a prior distribution scenario.

[0109] Furthermore, the linear equalization module uses zero-forcing equalization to extract the initial estimate of the communication signal, which is expressed as:

[0110]

[0111] Among them, H * represents the matrix obtained by taking the conjugate transpose of the channel matrix H.

[0112] Furthermore, the linear equalization module uses the minimum mean square error equalization method to extract the initial estimate of the communication signal, which is expressed as:

[0113]

[0114] Among them, σ 2 represents the noise variance of the channel, I represents a vector whose elements are all 1, and H * represents the matrix obtained by taking the conjugate transpose of the channel matrix H.

[0115] Furthermore, the linear equalization module uses the binning modeling method to extract the initial estimate of the communication signal in the prior distribution scenario, which is expressed as:

[0116]

[0117] in, represents the initial estimate of the communication signal extracted based on the zero-forcing equalization method in the prior distribution scenario, It represents the initial estimate of the communication signal extracted based on the minimum mean square error equalization method in the prior distribution scenario, and N represents the signal length. represents the initial estimate under the zero-forcing equilibrium method, Represents the initial estimate under the minimum mean square error equalization method.

[0118] Specifically, the initial estimation of the signal and its prior distribution is obtained using the zero-forcing equalization method (ZF) or the minimum mean square error equalization method (MMSE).

[0119] Step 2.1: Apply the formula or The initial estimation of the signal is obtained, the former is the result of the ZF equalization method, and the latter is the result of the MMSE equalization method.

[0120] Step 2.2: Results of ZF equilibrium method As an example, the binning method is used to model to obtain an initial estimate of the prior distribution. The specific process includes:

[0121] Find vector The minimum and maximum values ​​of and Defining the right spacing (The spacing length is generally less than 0.1, depending on the specific situation), which enables the creation of an equidistant grid set. Assume that the length of this set is R, recorded as

[0122] Then set the vector of length R in Is from vector Elements Falling on The number of samples for the center pane area, for example:

[0123]

[0124] Count vector The probability vector f is followed on R categories ZF , a multinomial distribution with N trials can be expressed as:

[0125]

[0126] f ZF An unbiased estimate of is:

[0127]

[0128] That is, the initial estimate of the prior distribution modeled by the ZF equilibrium method, which contains noise and has low accuracy.

[0129] Similarly, we can also get

[0130] It can be understood that by performing initial estimation extraction on the acquired communication signal under a prior distribution scenario, the modulation mode of the signal can be further inferred.

[0131] In step S3, the process of iteratively refining the communication signal and its initial estimate in the prior distribution scenario includes:

[0132] The communication signal and its initial estimate in the prior distribution scenario are input into the iterative noise reduction module and the prior estimation module;

[0133] The iterative denoising module iteratively denoises the communication signal and its initial estimate in the prior distribution scenario, and the prior estimation module deconvolves the denoised communication signal and its initial estimate in the prior distribution scenario, and iteratively learns a more accurate estimate of the signal prior distribution.

[0134] Further, let or and For discrete distribution p 1 The mean and variance of (x), The expression of the first denoising of the communication signal and its initial estimate in the prior distribution scenario by the iterative denoising module is:

[0135]

[0136]

[0137]

[0138] in, Both represent intermediate variables of linear operations in the iteration process. Subscript a represents the ath element of a vector of length M, and subscript i represents the ith element of a vector of length N. ai represents the element in the ath row and ith column of the matrix H. The superscript 1 or 2 represents the count value of the number of iterations. The functions η(·) and θ(·) mean to take the posterior distribution The mean and variance of and They represent the posterior distribution The mean and variance of .

[0139] Furthermore, based on the first iterative denoising, the iterative denoising module performs T-1 iterative denoising on the communication signal and its initial estimate in the prior distribution scenario; the prior estimation module uses the empirical Bayesian deconvolution method to perform a priori estimation on the denoised communication signal, and the expression is:

[0140]

[0141] in, Both represent intermediate variables of linear operations in the iteration process. Subscript a represents the ath element of a vector of length M, and subscript i represents the ith element of a vector of length N. ai represents the element in the ath row and ith column of the matrix H, t represents the number of iterations, and the functions η(·) and θ(·) mean to take the posterior distribution The mean and variance of and They represent the posterior distribution The mean and variance of Indicates that a priori estimation of the degrees of freedom dof is performed based on Bayesian deconvolution.

[0142] Specifically, in a large-dimensional system, the following scalar state evolution equation holds true, expressed as:

[0143]

[0144] in, for In the equation The corresponding random variable; ε is a random variable with a mean of 0 and a variance of where X is a Gaussian random variable and X is a random variable that obeys the true distribution g(x). This process can be described by convolution:

[0145]

[0146] in, is the empirical distribution of the observed values, is the transition probability of Gaussian form. Detailed deconvolution requires processing the imaginary part and the real part of the signal separately. The specific process is:

[0147] A. Given a fixed interval constellation point value space Considering that the modulus of a single constellation point vector in the QAM constellation diagram is limited after power normalization, in practice, an equidistant grid with an interval of 0.02 can be given.

[0148] B.Yes Perform binning modeling to obtain the observation space Count vector The probability vector for S categories is Multinomial distribution with N trials, that is:

[0149] C. This prior estimation module assumes that g(x) is The exponential family distribution on is expressed as follows using the discrete formula:

[0150]

[0151] Where Q is the structure matrix of dimension L×dof, using the natural spline basis function generate, is the jth row vector of Q; dof is the degree of freedom, which is also the dimension of the unknown vector α; 1 is a vector of all 1s of length L, and 1v(α) is used to ensure probability normalization. The purpose of deconvolution is to find an estimate of α

[0152] Construct a transfer matrix P of dimension S×L, where the matrix elements are the transition probabilities: The convolution relationship can be simplified to matrix multiplication therefore It can be easily expressed as a function of α.

[0153] The log-likelihood function of is:

[0154]

[0155] in Then define:

[0156]

[0157] Let W s is an L-dimensional vector, that is, W s (α)=[ws1 (α),w s2 (α),...,w sL (α)] T .

[0158] Then the scoring function for α is:

[0159]

[0160] in,

[0161] Maximum likelihood estimate of α This can be done by numerically maximizing l(α) or solving the matrix equation Get, substitute back You can get the estimate of the prior distribution p(x), and this operation can be run in the R environment.

[0162] Furthermore, a loss function is set to optimize the multi-input multi-output system model in the iterative refinement process, and when the loss function converges or when a preset iteration round ends, a multi-input multi-output system model capable of iterative signal detection in a priori missing scenario is obtained;

[0163] Model outputs of MIMO systems and obtain final estimates of the signals and the final estimate of the prior distribution p T (x).

[0164] It can be understood that the iterative denoising module combines with the prior estimation module to iteratively denoise the communication signal and its initial estimate in the prior distribution scenario and iteratively learn a more accurate estimate of the signal prior distribution, and sets the loss function and several rounds of iterations to iteratively optimize the multi-input multi-output system model to obtain a multi-input multi-output system model that can perform iterative signal detection in the scenario where the prior is missing, so that the model can be effectively applied to the scenario of dense prior, and improve the accuracy of signal detection, while the probability distribution range of the model is wider.

[0165] In this embodiment, firstly, by performing initial estimation extraction on the acquired communication signal under the prior distribution scenario, the modulation mode of the signal can be further inferred, so as to iteratively refine the communication signal and its initial estimation under the prior distribution scenario, thereby improving the dimension of the signal, and then effectively improving the accuracy of signal estimation in dense prior scenarios; secondly, the constructed multi-input and multi-output system model can be effectively applied to dense prior scenarios through direct cooperation among the linear equalization module, the iterative noise reduction module and the prior estimation module, and improve the accuracy of signal detection, while the probability distribution range of the model is wider.

[0166] Embodiment 2:

[0167] In order to verify the feasibility and effectiveness of the method of the present invention, this embodiment conducts experiments on the present invention, as follows:

[0168] like Figure 3 As shown in the figure, the convergence of the EB-AMP algorithm is shown under the conditions of a signal-to-noise ratio of 10dB, an 8-order QAM constellation modulation mode, and a transmission signal length of N = 512. As can be seen from the figure, with the increase in the number of iterations, the MSE gradually decreases to a basically unchanged level. It can be seen that the EB-AMP algorithm initialized with the ZF or MMSE equalization method can converge normally, and the performance of the two is basically the same, close to the known prior Qracle-AMP algorithm.

[0169] like Figure 4 As shown, it is verified that the state evolution equation of the EB-AMP algorithm initialized by the ZF equilibrium method during the second iteration is In the second iteration, Subtract the real X, separate the real part from the imaginary part, perform Poisson regression after binning modeling, and then compare it with the Gaussian variable Comparing the distribution of Figure 4 The quantile-quantile plot shown in . Figure 4 It can be seen that the sample fits well, indicating that the Gaussianity of the iterative process is guaranteed. In addition, the Gaussianity of the state evolution equation of the EB-AMP algorithm initialized by the MMSE equalization method can also be guaranteed during the second iteration.

[0170] like Figure 5 and 6 As shown in the figure, the performance comparison of Oracle-AMP, EB-AMP, LS and LMMSE methods under different signal-to-noise ratios with an 8-order constellation modulation mode is shown. As the signal-to-noise ratio gradually increases, the MSE of the four methods gradually decreases. Obviously, the performance of Oracle-AMP and EB-AMP is better than LS and LMMSE. Figure 5 The signal length N = 512, Figure 6 The signal length N=2048. When the signal length increases, the performance of the EB-AMP method will approach the real Oracle-AMP algorithm.

[0171] like Figure 7 As shown in the figure, the recovery of the signal probability distribution is demonstrated when the signal-to-noise ratio is 10dB, the modulation mode is 8-order QAM constellation, and the transmission signal length N=512. (a) and (c) are the real and imaginary parts of the initial signal probability distribution, respectively, and (b) and (c) are the real and imaginary parts of the final signal probability distribution, respectively. It can be seen that the noise component is significantly reduced, and the original modulation mode can be inferred after extracting the peak information.

[0172] Embodiment three:

[0173] This embodiment provides an iterative signal detection system in a scenario where a priori is missing. Figure 8 , the system comprising:

[0174] An acquisition module, used for acquiring communication signals in a multiple-input multiple-output system;

[0175] A processing module, configured to extract an initial estimate of the communication signal in a prior distribution scenario based on the communication signal;

[0176] The iterative signal detection module is used to iteratively refine the communication signal and its initial estimate in the prior distribution scenario, obtain the final estimate of the communication signal and the prior distribution, and complete the iterative signal detection.

[0177] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An iterative signal detection method in a priori missing scenario, characterized in that: The method comprises the following steps: Acquiring communication signals in a multiple-input multiple-output system; Based on the communication signal, extracting an initial estimate of the communication signal in a prior distribution scenario; Iteratively refine the communication signal and its initial estimate in the prior distribution scenario to obtain the final estimate of the communication signal and the prior distribution, and complete iterative signal detection; The process of extracting an initial estimate of the communication signal under the prior distribution scenario includes: The linear equalization module uses a linear equalizer to extract an initial estimate of the communication signal; Based on the extracted initial estimate of the communication signal, the linear equalization module extracts an initial estimate of the communication signal in a prior distribution scenario; The linear equalization module uses zero-forcing equalization to extract the initial estimate of the communication signal, which is expressed as: Among them, H * represents the matrix obtained by taking the conjugate transpose of the channel matrix H, and y represents the received signal; The linear equalization module uses the minimum mean square error equalization method to extract the initial estimate of the communication signal, which is expressed as: Among them, σ 2 represents the noise variance of the channel, I represents a vector whose elements are all 1, and H * represents the matrix obtained by taking the conjugate transpose of the channel matrix H, and y represents the received signal; The linear equalization module uses binning modeling to extract the initial estimate of the communication signal in the prior distribution scenario. The process includes: Find vector Minimum value of and maximum value Defining the right spacing And create an isometric grid set, the expression is: Set The length is R, then the expression is: Then set the vector of length R in Is from vector Elements Falling on The number of samples in the pane area centered is expressed as: Count vector The probability vector f is followed on U categories ZF , a multinomial distribution with N trials, is expressed as: Under the zero-forcing equilibrium method, f ZF The initial estimate under the prior distribution scenario is expressed as: Based on the initial estimate in the prior distribution scenario obtained under the zero-forcing equalization method, we get f under the minimum mean square error equalization method. MMSE The initial estimate under the prior distribution scenario is expressed as: in, represents the initial estimate of the communication signal extracted based on the zero-forcing equalization method in the prior distribution scenario, It represents the initial estimate of the communication signal extracted based on the minimum mean square error equalization method in the prior distribution scenario. represents the initial estimate under the zero-forcing equilibrium method, represents the initial estimate under the minimum mean square error equalization method; The process of iteratively refining the communication signal and its initial estimate in the prior distribution scenario includes: The communication signal and its initial estimate in the prior distribution scenario are input into the iterative noise reduction module and the prior estimation module; The iterative noise reduction module iteratively performs noise reduction on the communication signal and its initial estimate in the prior distribution scenario, and the prior estimation module deconvolves the noise-reduced communication signal and its initial estimate in the prior distribution scenario, and iteratively learns a more accurate estimate of the signal prior distribution; The iterative denoising module performs several rounds of iterative denoising on the communication signal and its initial estimate in the prior distribution scenario; the prior estimation module uses the empirical Bayesian deconvolution method to perform a priori estimation on the denoised communication signal, and the expression is: A loss function is set to optimize the multi-input multi-output system model in the iterative refinement process, and when the loss function converges or when a preset iteration round ends, a multi-input multi-output system model capable of iterative signal detection in a priori missing scenario is obtained; Model outputs of MIMO systems and obtain final estimates of the signals and the final estimate of the prior distribution p T (x); in, Both represent intermediate variables of linear operations in the iteration process. Subscript a represents the ath element of a vector of length M, and subscript i represents the ith element of a vector of length n. ai represents the element in the ath row and ith column of the matrix H, t=1...T represents the number of iterations, and the functions η(·) and θ(·) mean to take the posterior distribution The mean and variance of and They represent the posterior distribution The mean and variance of Indicates the prior estimation of degrees of freedom dof based on Bayesian deconvolution, y a represents the ath element of the received signal y; The process of the prior estimation module using the empirical Bayesian deconvolution method to perform a priori estimation on the noise-reduced communication signal includes: The scalar state evolution equation in a large dimensional system is expressed as: in, for In the equation The corresponding random variable; ε is a random variable with a mean of 0 and a variance of is a Gaussian random variable, X is a random variable that obeys the true distribution g(x); Assume the convolution expression is: in, is the empirical distribution of the observed values, is the transition probability in Gaussian form; The imaginary part and the real part of the signal in the deconvolution process are processed separately. The specific process is: A: Given a fixed interval constellation point value space B: Yes Perform binning modeling to obtain the observation space Count vector The probability vector for S categories is For a multinomial distribution with N trials, the expression is: C: Prior estimation module Assume g(x) is The exponential family distribution on is expressed by the discrete formula: Construct a transfer matrix P with a dimension of S×L. The matrix elements are the transition probabilities, and the expression is: The convolution relationship can be simplified to matrix multiplication make It can be easily expressed as a function of α; get The log-likelihood function of is: in definition: Let W s is an L-dimensional vector, that is, W s (α)=[w s1 (α),w s2 (α),...,w sL (α)] T ; The scoring function for α is: in, Maximum likelihood estimate of α By numerically maximizing l(α) or solving the matrix equation Get, substitute back Get an estimate of the prior distribution p(x); Among them, Q represents the structure matrix with dimension L×dof, using natural spline basis function generate, represents the j-th row vector of Q; dof represents the dimension of the unknown vector α; 1 represents a full-1 vector of length L, and 1v(α) is used to ensure probability normalization.

2. The iterative signal detection method in the scenario of prior missing according to claim 1, characterized in that: Before extracting the initial estimate of the signal under the prior distribution scenario, the following steps are also included: Construct a multi-input and multi-output system model, the expression is: y=Hx+w; The model includes a linear equalization module, an iterative noise reduction module and a priori estimation module; The linear equalization module is used to perform linear equalization processing on the communication signal and extract an initial estimate of the communication signal in a priori distribution scenario; The iterative noise reduction module is used to perform iterative noise reduction processing on the initial estimate of the communication signal in the prior distribution scenario, and the prior estimation module is used to refine the initial estimate after each iterative noise reduction processing and output a final estimate of the communication signal and the prior distribution; Among them, x represents the communication signal, H represents the channel matrix, w represents the superimposed Gaussian white noise, and y represents the received signal.

3. An iterative signal detection system in a priori missing scenario, the system is used to implement the steps of the method according to any one of claims 1-2, characterized in that: The system comprises: An acquisition module, used for acquiring communication signals in a multiple-input multiple-output system; A processing module, configured to extract an initial estimate of the communication signal in a prior distribution scenario based on the communication signal; The iterative signal detection module is used to iteratively refine the communication signal and its initial estimate in the prior distribution scenario, obtain the final estimate of the communication signal and the prior distribution, and complete the iterative signal detection.

Images