Super network assisted data model dual-drive MIMO-OFDM signal detection method

Abstract

The invention discloses a super network assisted data model dual-drive MIMO-OFDM (Multiple Input Multiple Output-Orthogonal Frequency Division Multiplexing) signal detection method, which comprises a signal detection network with a group of adjustable parameters and a trainable super network. The method comprises the following steps of: firstly, acquiring channel estimation by utilizing a linear minimum mean square error algorithm, taking an estimated channel matrix HLMMSE as input of a super network, and outputting an adjustable parameter omega hyper to be provided for a signal detection network; next, real value decomposition is carried out on a channel matrix HLMMSE and a received signal vector y, the real value decomposition and deduced equivalent noise variance information # imgabs0 # are input into a signal detection network together, and the signal detection network is formed by expanding a graph neural network enhanced expected propagation algorithm; in combination with a real number field channel matrix # imgabs1 #, a real number field receiving signal vector # imgabs2 # and environment information # imgabs3 # and a parameter omega hyper provided by a super network, an estimation # imgabs4 # of a sending symbol vector is given, and finally, the estimated symbol vector # imgabs5 # is demapped to obtain an estimation # imgabs6 # of an original sending bit. The method is high in detection performance, less in time consumption, high in generalization and capable of adapting to scene changes.

Description

Technical Field

[0001] The invention relates to a super network assisted data model dual-driven MIMO-OFDM signal detection method, belonging to the technical field of wireless communication. Background Art

[0002] Channel estimation and signal detection are important tasks at the receiver end of MIMO-OFDM systems. Among them, the channel estimation method based on linear minimum mean square error has been widely used. In addition, the signal detection algorithm based on expected propagation has achieved a good balance between computational complexity and performance and has been widely studied. However, the imperfect channel state information obtained by the estimation algorithm will cause serious performance loss to the data inference method based on expected propagation. In addition, in scenarios with strong multi-user interference, the expected propagation detector still has a large performance gap compared to the optimal maximum likelihood detection baseline.

[0003] In recent years, deep learning technology has been widely used in fields such as image recognition, speech recognition, and natural language processing. Artificial intelligence-assisted signal detection design combines deep learning technology with expert knowledge, and uses a large amount of labeled data to learn the structured features of data embedded in multi-antenna detection, which is expected to make up for the shortcomings of model-based solutions. In related research directions, integrating graph neural networks into the iterative detection process can effectively learn embedded multi-user interference information and compensate for performance defects. However, the training of the detection network is often completed in specific scenarios, and there is a problem of insufficient robustness. When the environment changes significantly, the network cannot be adjusted adaptively, and the performance will be seriously deteriorated. In addition, under the condition of imperfect channel state information, it is necessary to further solve the problem of deviation of iterative detection from the posterior probability distribution of transmitted symbols to compensate for performance losses. Summary of the invention

[0004] Purpose of the invention: The present invention provides a hypernetwork-assisted data model dual-driven MIMO-OFDM signal detection method, which aims to utilize a data model dual-driven neural network to effectively improve the detection performance, and at the same time introduce a hypernetwork to improve the adaptability of the signal detection network.

[0005] Technical solution: The present invention adopts the following technical solution:

[0006] The hypernetwork-assisted data model dual-driven MIMO-OFDM signal detection method of the present invention comprises a signal detection network with a set of adjustable parameters and a trainable hypernetwork, and the method comprises the following steps:

[0007] (1) The channel state information is estimated using the linear minimum mean square error algorithm. The hypernetwork takes the channel state information as input and outputs an adjustable parameter Ω hyper Provided to a signal detection network;

[0008] (2) Estimated channel matrix H LMMSE The real-valued decomposition of the received signal vector y is performed, and the obtained real-domain channel matrix Real domain received signal vector And the derived equivalent noise variance information Input to the signal detection network, the signal detection network is composed of T graph neural network enhanced sub-modules in series, and the adjustable parameters are shared between different graph neural network modules; based on the real domain channel matrix estimation Real domain received signal vector Environmental Information The parameters provided by the hypernetwork and the adjustable parameters of the submodules give an estimate of the transmitted symbol vector

[0009] (3) Estimated symbol vector Demap to obtain an estimate of the original transmitted bits

[0010] Furthermore, the hypernetwork in step (1) is a hypernetwork with trainable parameters Θ hyper Neural network, some adjustable parameters Ω of the signal detection network hyper Generated by the super network; the channel state information input to the super network is the channel matrix H estimated by the linear minimum mean square error algorithm LMMSE , the super network channel matrix H LMMSE Perform singular value decomposition and perform real value decomposition on the obtained singular value vector s to obtain the singular value vector The hypernetwork is implemented using a deep neural network consisting of three fully connected layers, including {256,256,4N u} neurons, where N u is the feature vector u (0) The length of the hypernetwork output is the adjustable parameter set Ω hyper , corresponding to the input feature vector u of the first submodule t=1 of the signal detection network (0) A set of adjustable parameters required in the calculation process.

[0011] Furthermore, step (2) specifically includes:

[0012] (2.1) For the channel matrix H LMMSE And the received signal vector y is decomposed into real values ​​to obtain the real domain channel matrix and the real domain received signal vector

[0013] (2.2) According to the estimated channel matrix error, the second-order statistical characteristics of the noise are corrected R z, the real domain correlation matrix R is obtained by real value decomposition e , and extract equivalent environmental information The expression is:

[0014]

[0015] Where Re(·) represents the real part of a complex number, N r is the number of receiving antennas, tr(·) represents the trace of the matrix;

[0016] (2.3) The expected propagation algorithm enhanced by the graph neural network is used as the signal detection network. The deep neural network contains T sub-modules connected in series, and the adjustable parameters are shared between different graph neural network modules. The real domain channel matrix Real domain received signal vector and equivalent environmental information As the input of the signal detection network; the input feature vector u of the first submodule t=1 of the signal detection network (0) The calculation process involves the adjustable parameter Ω hyper , provided by the hypernetwork; the input feature vector u of the subsequent t=2,...,T submodules (0) According to the output feature vector u of the graph neural network in the t-1th module (L) Assignment, where L is the number of iterations of the feature vector in the graph neural network; the t-th submodule of the signal detection network is based on the real domain channel matrix Real domain received signal vector Equivalent environmental information And the input feature vector u (0) To complete the estimation of this layer, the t-th submodule is divided into three parts: observation, graph neural network enhancement, and estimation. The observation part is based on the expected propagation detection algorithm. First, the Gaussian posterior approximation is constructed using the exponential cluster distribution and its mean is calculated. With covariance Then the likelihood function is calculated based on the Gaussian posterior function, and the mean of the likelihood function is With covariance The graph neural network is used as prior information to input the graph neural network; the enhanced part of the graph neural network uses the likelihood distribution characteristics and multi-user interference information to characterize the latent function relationship between variable nodes and factor nodes, and strengthens the posterior probability inference. Input estimation module. In addition, the feature vector is continuously updated in the message passing process between the variable node and the factor node. After L iterations, the output feature vector u is obtained. (L) ; The estimation part makes soft decisions on the transmitted symbols based on the posterior probability and calculates the mean and variance Combine the prior information of the likelihood function to update the parameter pair (γ (t) ,Λ (t) ), and the output feature vector u (L) are input together to the t+1th submodule; the Tth submodule outputs the final estimated symbol vector

[0017] Furthermore, the second-order statistical characteristic R of the noise z The correction of is obtained by deriving the second-order moment of the channel matrix estimation error ΔH; Indicates the first n=1,...,N t The pilot subcarrier position occupied by the root transmit antenna is Indicates the data subcarrier position, where P is the pilot sequence length, D is the number of data subcarriers, and N t is the number of transmitting antennas, using the received pilot component Known pilot sequence The frequency correlation characteristics of the channel are calculated for m=1,...,N r LMMSE estimation of the channel coefficient between the nth receiving antenna and the nth transmitting antenna W LMMSE,n is the interpolation matrix used in the linear minimum mean square error estimation calculation, and the corresponding true channel coefficient vector is expressed as h m,n ; At the kth data subcarrier, there is a received vector According to the estimated channel matrix calculated and omit Construct the following equivalent relationship:

[0018] y=H LMMSE x+z,z=ΔHx+w,

[0019] The variance of the white noise vector w is According to the row index d k and column index From the interpolation matrix W LMMSE,n Extract the column vector w from LMMSE,n , and calculate the matrix The second-order moment R of the equivalent noise vector z z =E{zz H} is represented as:

[0020] R Δh,n =R A +R D -R B +R C ,R A =R hh ,

[0021]

[0022] Among them, the average signal energy E s =1 / N t , I is the identity matrix, w k′ and Corresponding to vector w LMMSE,n and the elements in the matrix M, The main diagonal elements are The diagonal matrix of Related information hh , The definition is as follows:

[0023]

[0024] in, The above related information is extracted using the MIMO-OFDM channel correlation matrix.

[0025] Furthermore, in the graph neural network enhancement part of the t-th submodule, a paired Markov random field model is used to characterize the latent function relationship between variable nodes and factor nodes, and to extract factor features in and Respectively refer to No. j column, For the first submodule t=1, the variable feature vector The calculation of is based on the above latent function relationship and the adjustable parameters provided by the hypernetwork. The adjustable parameter Ω hyper Contains the weight matrix W hyper and the bias vector b hyper , the variable eigenvector is calculated as follows:

[0026]

[0027] Eigenvector The length is N u ; In the l=1,...,Lth iteration, the feature With variable eigenvector Input factor nodes, aggregate through multilayer perceptron network 1, output message information Prior information with likelihood function After integration by the gated recurrent unit, the feature vector at the variable node is updated Iterate L times to complete the message transmission between the variable node and the factor node, and finally output the feature vector Computational graph neural network enhanced posterior probability inference

[0028] Furthermore, the complete neural network consisting of the signal detection network and the hypernetwork is trained end-to-end with supervised learning using a mini-batch gradient descent algorithm for multiple rounds. A mini-batch of the training set is represented as is a set of S randomly generated samples, where the real domain symbol vector sent As labels, the real field receives symbolic vectors and the real-domain accurate channel matrix As the input features of the complete neural network; the adaptive momentum estimation optimizer is selected to optimize the trainable parameters Θ in the neural network, and the learning rate is set to 0.001; the cross entropy loss function is used for training, and the specific formula is:

[0029]

[0030] in, For built-in functions, In the case of , it is 1, and in the other cases it is 0. It is the posterior probability inference output by the Tth submodule of the signal detection network enhanced by the graph neural network.

[0031] Beneficial effects: Compared with the prior art, the present invention has the following beneficial effects:

[0032] The present invention constructs a signal detection neural network based on the expected propagation algorithm, enhances the inference of the posterior probability of the transmission symbol with the help of the graph neural network, and avoids the deviation of the iterative detection algorithm caused by the imperfect channel state information through the correction of environmental information. Compared with the existing scheme, the present invention has achieved considerable performance gain. In addition, the present invention introduces a hypernetwork to dynamically capture the changes in channel characteristics, so as to specifically adjust a small number of key parameters in the signal detection network, which has the advantages of strong generalization and adaptability to scene changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a system block diagram of an embodiment of the present invention;

[0034] Figure 2 A schematic diagram of the architecture of a complete neural network used in the present invention;

[0035] Figure 3 A schematic diagram of the architecture of a signal detection network used in the present invention. DETAILED DESCRIPTION

[0036] The present invention will be described in detail below with reference to the accompanying drawings and embodiments of the MIMO system.

[0037] 1. System Model Used in This Example

[0038] Consider a MIMO-OFDM system using spatial multiplexing technology. The transmitter has N t antennas, the receiving end has N r antennas, the number of subcarriers is K. The training pilot length of each transmitting antenna is P, and it occupies subcarriers for transmission in a comb-like manner. The rest D = K-PN t subcarriers are used to transmit data. Indicates the pilot subcarrier position of the nth transmitting antenna, using Indicates the data subcarrier position. In this embodiment, N is set t =N r =8, K=512, P=16, D=384. In this embodiment, the modulation method adopts QPSK constellation modulation. In one symbol time, the data bits of the nth transmitting antenna are 384×2=768 bits. After QPSK modulation, the transmission data symbol vector is obtained. And the pilot symbol vector composes OFDM symbols, where A is the QPSK constellation symbol set. At the mth receiving antenna, Refers to the received data component, with Refers to the received pilot component corresponding to the nth transmitting antenna, then the signal model of the MIMO-OFDM system can be expressed as:

[0039]

[0040] in, The main diagonal elements are The diagonal matrix of and They correspond to the frequency domain channels through which the pilot and data transmission pass, respectively. and is an additive Gaussian white noise vector with a noise variance of Using the received pilot component Send pilot sequence And the frequency-dependent characteristics of the channel, the linear minimum mean square error estimate of the channel coefficient between the mth receiving antenna and the nth transmitting antenna can be calculated:

[0041]

[0042] Where W LMMSE,n is the LMMSE interpolation matrix. In addition, in d k There is a MIMO subsystem at the data subcarrier position, and the signal model can be expressed as:

[0043]

[0044] in, After the omission Simplify the vector representation in the MIMO subsystem.

[0045] In this embodiment, a tapped delay line model is used to generate a MIMO-OFDM channel, and its spatial correlation characteristics are simulated using a matrix R. The specific formula is:

[0046]

[0047] Among them, N r ×N r Dimensional matrix R r and N t ×N t Dimensional matrix R t are the correlation matrices of the receiving end and the sending end respectively, which are generated using the exponential correlation model. The (i, j)th element r of the matrix ij satisfy:

[0048]

[0049] Wherein, ρ represents the correlation coefficient of the channel matrix.

[0050] 2. Specific steps of this embodiment

[0051] like Figure 1 As shown, an embodiment of the present invention provides a system block diagram of a data model dual-driven MIMO-OFDM signal detection method assisted by a hypernetwork, including real value decomposition and equivalent environmental information correction, a signal detection neural network with a set of adjustable parameters, a hypernetwork for generating some adjustable parameters in the signal detection network, and a demapping module. In this embodiment, the receiving end obtains non-perfect channel state information through a linear minimum mean square error algorithm; the complete neural network composed of the signal detection network and the hypernetwork is shown in FIG. Figure 2 As shown in the figure, specifically, the signal detection network is composed of T graph neural network enhanced submodules connected in series, and adjustable parameters are shared between different graph neural network modules; the super network is a deep neural network composed of three fully connected layers with trainable parameters Θ hyper , based on the channel state information, an adjustable parameter set Ω is generated hyper , corresponding to a set of adjustable parameters required in computing the input features of the first submodule of the signal detection network.

[0052] Offline end-to-end supervised learning training of the complete neural network composed of the signal detection network and the hypernetwork is performed to optimize the trainable parameters Θ in the hypernetwork hyperThe trained network is used to complete the forward estimation when deploying online. The forward estimation is divided into four steps: generating adjustable parameters through the super network, real value decomposition and correction of equivalent environmental information, forward signal detection, and symbol demapping:

[0053] (1) Hypernetwork generation with adjustable parameters

[0054] For each channel realization, the super network generates some adjustable parameters of the signal detection network based on the channel state information. The channel state information input to the super network is the channel matrix estimated by the linear minimum mean square error algorithm The super network first calculates the channel matrix H LMMSE Perform singular value decomposition to obtain the singular value vector Then perform real value decomposition to obtain the singular value vector like Figure 2 As shown, the hypernetwork will Input deep neural network, output adjustable parameter set Ω hyper Contains the weight matrix W hyper and the bias vector b hyper , the calculation formula is:

[0055]

[0056] Where g(·) contains three fully connected layers, and the neurons are set to {256, 256, 4N u},N u is the feature vector u (0) Length; Output Ω hyper The first submodule t=1 of the signal detection network is used to realize the conversion from external information to input feature u (0) 's mapping.

[0057] (2) Real value decomposition and equivalent environmental information correction

[0058] In order to facilitate algebraic operations and the use of deep learning methods, we first use the estimated channel matrix H LMMSE The system represented by equation (3) is equivalently constructed as:

[0059] y=H LMMSE x+z,z=ΔHx+w (4)

[0060] Among them, the channel matrix estimation error ΔH = HH LMMSE ; According to the row index d k and column index From the interpolation matrix W LMMSE,n Extract the column vector w from LMMSE,n , and calculate the matrix The second-order moment R of the equivalent noise vector z z =E{zzH} means the following:

[0061] R Δh,n =R A +R D -R B +R C ,R A =R hh ,

[0062]

[0063] Among them, the average signal energy E s =1 / N t , I is the identity matrix, w k′ and Corresponding to vector w LMMSE,n and the elements in the matrix M, tr(·) represents the trace of the matrix, and the relevant information R hh , The definition is as follows:

[0064]

[0065] in, The above related information is extracted using the MIMO-OFDM channel correlation matrix.

[0066] Then, the formula (4) is decomposed into the following real value:

[0067]

[0068] Re(·) and Im(·) represent the real and imaginary parts of a complex number, respectively. T Represents the transpose operation. After real value decomposition, the equivalent real number form of the MIMO subsystem is as follows:

[0069]

[0070] The equivalent noise The variance of The expression is:

[0071]

[0072] (3) Forward signal detection

[0073] For signal detection, the expectation propagation algorithm enhanced by graph neural network is used as the signal detection network. The deep neural network contains T sub-modules connected in series, and the adjustable parameters are shared between different graph neural network modules, such as Figure 3 As shown; the signal detection network is based on the real domain channel matrix Real domain received signal vector and equivalent environmental information Solve the system represented by equation (4) to complete the transmission symbol vector Estimates.

[0074] like Figure 3 As shown in Figure 1, the t-th submodule of the signal detection network significantly improves the detection performance through the posterior probability inference enhanced by the graph neural network. The input feature vector u of the first submodule t=1 of the signal detection network (0) The calculation process involves the adjustable parameter Ω hyper , provided by the super network; the t=1,...,T submodules are based on the real domain channel matrix Real domain received signal vector Equivalent environmental information And the input feature vector u (0) Complete the estimation of this layer; each submodule has the same structure and shares adjustable parameters, which can be divided into three parts: observation, graph neural network enhancement and estimation. The following takes the tth submodule as an example to explain the workflow of these three parts:

[0075] (3.1) Observation part: The observation part includes a minimum mean square error operation unit and an external information operation unit. First, the mean of the Gaussian posterior distribution is constructed using the exponential cluster distribution. With covariance The essence is the minimum mean square error estimation:

[0076]

[0077] Among them, (γ (t) ,Λ (t) ) parameter pair according to Initialize, Then the mean of the likelihood function is calculated based on the Gaussian posterior function With covariance Diagonal Matrix The main diagonal Elements and the mean No. Elements The specific formula used for calculation is:

[0078]

[0079] in Corresponding Gaussian posterior distribution covariance matrix The main diagonal elements, Corresponding mean No. elements; external mean With covariance As prior information input to the graph neural network part.

[0080] (3.2) Graph neural network enhancement part: Based on the paired Markov random field model, the latent function relationship between variable nodes and factor nodes in the graph neural network is characterized, and factor features are extracted in and Respectively refer to No. j column, For the first submodule t=1, the variable feature vector The calculation of is based on the above latent function relationship and the adjustable parameter Ω provided by the hypernetwork hyper = {W hyper ,b hyper}Calculated:

[0081]

[0082] Eigenvector The length is N u ; The message transmission process and feature vector between variable nodes and factor nodes in the l=1,...,Lth iteration The iterative update calculation of is as follows:

[0083] a. Features With vector Input factor nodes, aggregate through multilayer perceptron network 1, output message

[0084] The multi-layer perceptron f MLP1 (·) Implemented by a three-layer deep neural network;

[0085] b. Message Prior information with likelihood function After integration by the gated recurrent unit, the feature vector at the variable node is updated

[0086]

[0087] in and They are gated recurrent units f GRU (·) previous and current hidden states, f D ense1(·) is a single-layer deep neural network;

[0088] After L iterations, according to the output feature vector Calculate the posterior probability inference

[0089] Multilayer Perceptron MLP2 (·) is implemented by a three-layer deep neural network, Ω is the modulation symbol set composed of the real part of the transmitted constellation point; in addition, according to the output feature vector The variable feature vector of the t+1th submodule Assign a value.

[0090] (3.3) Estimation part: The estimation part includes soft decision calculation and parameter pair update; first, according to the posterior probability Soft decision is made on the transmission symbol. In this embodiment, The components of come from the modulation symbol set composed of the real part of the QPSK constellation point So the mean No. Elements And the diagonal matrix The main diagonal Elements The specific formula used for calculation is:

[0091]

[0092] The above soft estimation results are used together with the prior information of the likelihood function to update (γ (t) ,Λ (t) ) parameter pair; each of the T submodules performs operations according to the above steps (3.1)-(3.3), and finally outputs the estimated symbol vector

[0093] The complete neural network composed of the signal detection network and the hypernetwork is trained end-to-end using a mini-batch gradient descent algorithm. In this embodiment, a total of 2000 rounds of training are performed, each round containing 64 mini-batches. A mini-batch of the training set is represented as is a set of S randomly generated samples, where the real domain symbol vector sent As labels, the real field receives symbolic vectors and the real-domain accurate channel matrix As the input feature of the complete neural network, i is the sample number. In this embodiment, S is taken as 500, and the correlation coefficient ρ of the training set channel matrix is ​​randomly selected in [0.3, 0.6]. The adaptive momentum estimation optimizer is selected to optimize the trainable parameter θ in the neural network, and the learning rate is set to 0.001; the training adopts the cross entropy loss function, and the specific formula is:

[0094]

[0095] in, For built-in functions, In the case of , it is 1, and in the other cases it is 0. It is the posterior probability inference output by the Tth submodule of the signal detection network enhanced by the graph neural network; during training, 500 samples are sent into the network for forward propagation each time, and the loss function is calculated and then back-propagated to optimize the parameter Θ.

[0096] After completing the above training offline, the network is deployed online. In this embodiment, the correlation coefficient ρ of the spatial correlation channel used to evaluate the network performance is taken as 0.8. Since the super network has learned the law of initialization mapping of signal detection network features in different scenarios during training, it can still generate appropriate parameters for new scenarios in the test and provide them to the signal detection network. The signal detection network can maintain considerable detection performance based on these parameters. As a result, the adaptability of the detection method is greatly enhanced.

[0097] (4) Symbol Demapping

[0098] After signal detection is completed, the estimated symbol vector Restore to the complex domain, then demap, in this embodiment, perform hard decision, obtain an estimate of the original transmitted bit

[0099] What is described above in conjunction with the accompanying drawings is only a preferred embodiment of the present invention and cannot be used to limit the scope of rights included in the present invention. It should be understood that any equivalent changes made without departing from the spirit of the present invention are within the scope of protection covered by the claims of the present invention.

Claims

1. A super network assisted data model dual-driven MIMO-OFDM signal detection method, characterized in that: The MIMO-OFDM system includes a signal detection network having a set of adjustable parameters and a trainable super network, and the method includes the following steps: (1) The channel state information is estimated using the linear minimum mean square error algorithm. The hypernetwork takes the channel state information as input and outputs an adjustable parameter Ω hyper Provided to a signal detection network; (2) Estimated channel matrix H LMMSE The real-valued decomposition of the received signal vector y is performed, and the obtained real-domain channel matrix Real domain received signal vector And the derived equivalent noise variance information Input to the signal detection network, the signal detection network is composed of T graph neural network enhanced sub-modules in series, and the adjustable parameters are shared between different graph neural network modules; based on the real domain channel matrix estimation Real domain received signal vector Environmental Information The parameters provided by the hypernetwork and the adjustable parameters of the submodules give an estimate of the transmitted symbol vector (3) Estimated symbol vector Demap to obtain an estimate of the original transmitted bits 2. The method for detecting MIMO-OFDM signals using a super-network-assisted data model dual-drive method according to claim 1, characterized in that: The hypernetwork in step (1) is a network with trainable parameters Θ hyper Neural network, some adjustable parameters Ω of the signal detection network hyper Generated by the super network; the channel state information input to the super network is the channel matrix H estimated by the linear minimum mean square error algorithm LMMSE , the super network channel matrix H LMMSE Perform singular value decomposition and perform real value decomposition on the obtained singular value vector s to obtain the singular value vector The hypernetwork is implemented using a deep neural network consisting of three fully connected layers, including {256,256,4N u } neurons, where N u is the feature vector u (0) The length of the hypernetwork output is the adjustable parameter set Ω hyper , corresponding to the input feature vector u of the first submodule t=1 of the signal detection network (0) A set of adjustable parameters required in the calculation process.

3. The method for detecting MIMO-OFDM signals using a super-network-assisted data model dual-drive method according to claim 1, characterized in that: Step (2) specifically includes: (2.1) For the channel matrix H LMMSE And the received signal vector y is decomposed into real values ​​to obtain the real domain channel matrix and the real domain received signal vector (2.2) According to the estimated channel matrix error, the second-order statistical characteristics of the noise are corrected R z , the real domain correlation matrix R is obtained by real value decomposition e , and extract equivalent environmental information The expression is: Where Re(·) represents the real part of a complex number, N r is the number of receiving antennas, tr(·) represents the trace of the matrix; (2.3) The expected propagation algorithm enhanced by the graph neural network is used as the signal detection network. The deep neural network contains T sub-modules connected in series, and the adjustable parameters are shared between different graph neural network modules. The real domain channel matrix Real domain received signal vector and equivalent environmental information As the input of the signal detection network; the input feature vector u of the first submodule t=1 of the signal detection network (0) The calculation process involves the adjustable parameter Ω hyper , provided by the super network; the input feature vector u of the subsequent t=2,...,T submodules (0) According to the output feature vector u of the graph neural network in the t-1th module (L) Assignment, where L is the number of iterative updates of the feature vector in the graph neural network; the t-th submodule of the signal detection network is based on the real domain channel matrix Real domain received signal vector Equivalent environmental information And the input feature vector u (0) To complete the estimation of this layer, the t-th submodule is divided into three parts: observation, graph neural network enhancement, and estimation. The observation part is based on the expected propagation detection algorithm. First, the Gaussian posterior approximation is constructed using the exponential cluster distribution and its mean is calculated. With covariance Then the likelihood function is calculated based on the Gaussian posterior function, and the mean of the likelihood function is With covariance The graph neural network is used as prior information to input the graph neural network; the enhanced part of the graph neural network uses the likelihood distribution characteristics and multi-user interference information to characterize the latent function relationship between variable nodes and factor nodes, and strengthens the posterior probability inference. Input estimation module. In addition, the feature vector is continuously updated in the message passing process between the variable node and the factor node. After L iterations, the output feature vector u is obtained. (L) ; The estimation part makes soft decisions on the transmitted symbols based on the posterior probability and calculates the mean and variance Combine the prior information of the likelihood function to update the parameter pair (γ (t) ,Λ (t) ), and the output feature vector u (L) are input together to the t+1th submodule; the Tth submodule outputs the final estimated symbol vector 4. The method for detecting MIMO-OFDM signals using a super-network-assisted data model dual-drive method according to claim 3, characterized in that: Second-order statistical characteristics of noise R z The correction of is obtained by deriving the second-order moment of the channel matrix estimation error ΔH; Indicates the first n=1,...,N t The pilot subcarrier position occupied by the root transmit antenna is Indicates the data subcarrier position, where P is the pilot sequence length, D is the number of data subcarriers, and N t is the number of transmitting antennas, using the received pilot component Known pilot sequence The frequency correlation characteristics of the channel are calculated for m=1,...,N r Linear minimum mean square error estimation of the channel coefficients between the nth receiving antenna and the nth transmitting antenna W LMMSE,n is the interpolation matrix used in the linear minimum mean square error estimation calculation, and the corresponding true channel coefficient vector is expressed as h m,n ; At the kth data subcarrier, there is a received vector According to the estimated channel matrix calculated and omit Construct the following equivalent relationship: y=H LMMSE x+z,z=ΔHx+w, The variance of the white noise vector w is According to the row index d k and column index From the interpolation matrix W LMMSE,n Extract the column vector w from LMMSE,n , and calculate the matrix The second-order moment R of the equivalent noise vector z z =E{zz H } is represented as: Among them, the average signal energy E s =1 / N t , I is the identity matrix, w k′ and Corresponding to vector w LMMSE,n and the elements in the matrix M, The main diagonal elements are The diagonal matrix of Related information hh , The definition is as follows: in, The above related information is extracted using the MIMO-OFDM channel correlation matrix.

5. The method for detecting MIMO-OFDM signals using a super-network-assisted data model dual-drive method according to claim 3, characterized in that: In the graph neural network enhancement part of the t-th submodule, the paired Markov random field model is used to characterize the latent function relationship between variable nodes and factor nodes, and to extract factor features in and Respectively refer to No. j column, For the first submodule t=1, the variable feature vector The calculation of is based on the latent function relationship and the adjustable parameters provided by the hypernetwork. The adjustable parameter Ω hyper Contains the weight matrix W hyper and the bias vector b hyper , the variable eigenvector is calculated as follows: Eigenvector The length is N u ; In the l=1,...,Lth iteration, the feature With variable eigenvector Input factor nodes, aggregate through multilayer perceptron network 1, output message information Prior information with likelihood function After integration by the gated recurrent unit, the feature vector at the variable node is updated Iterate L times to complete the message transmission between the variable node and the factor node, and finally output the feature vector Computational graph neural network enhanced posterior probability inference 6. The method for detecting MIMO-OFDM signals using a data model dual-driven by a hyper-network according to any one of claims 1 to 3, characterized in that: The complete neural network consisting of the signal detection network and the hypernetwork is trained end-to-end by using a mini-batch gradient descent algorithm, and trained for multiple rounds; a mini-batch of the training set is represented as is a set of S randomly generated samples, where the real domain symbol vector sent As labels, the real field receives symbolic vectors and the real-domain accurate channel matrix As the input features of the complete neural network; the adaptive momentum estimation optimizer is selected to optimize the trainable parameters Θ in the neural network, and the learning rate is set to 0.001; the cross entropy loss function is used for training, and the specific formula is: in, For built-in functions, In the case of , it is 1, and in the other cases it is 0. It is the posterior probability inference output by the Tth submodule of the signal detection network enhanced by the graph neural network.

Images