A method and system for identifying modulation type of MIMO system under communication interference

Abstract

The present invention relates to a method and system for identifying the modulation type of a MIMO system under communication interference, belonging to the field of signal recognition technology. The method comprises: filtering, denoising, and whitening preprocessing the received MIMO signal; constructing a five-dimensional fourth-order cross-cumulant tensor using the preprocessed signal, and estimating a separation matrix through parallel factor decomposition to achieve suppressed interference reconstruction and separation; reconstructing a three-dimensional tensor from the separated and suppressed signal, and performing dimensionality reduction and feature extraction on the tensor using a generalized tensor discriminant analysis algorithm; and using Tucker decomposition to establish a core tensor and factor matrix for the reduced-dimensional tensor. The similarity of the core tensors of different modulation types is identified based on the Euclidean distance criterion to complete the MIMO signal modulation type identification. Simulation results show that the method of the present invention has good recognition accuracy under low signal-to-interference ratio conditions and does not require prior information such as channel coefficients and noise distribution.

Description

Technical Field

[0001] The present invention belongs to the technical field of communication signal modulation type identification in the field of electromagnetic spectrum monitoring, and in particular relates to a method and system for identifying the modulation type of multiple-input multiple-output (MIMO) signals under communication interference. Background Art

[0002] Electromagnetic spectrum monitoring is a crucial prerequisite for maintaining order in the electromagnetic spectrum, safeguarding the legitimate rights and interests of authorized users, and investigating and punishing illegal frequency usage by unauthorized users. With the advancement of information technology, electromagnetic spectrum monitoring services are becoming increasingly diverse. In recent years, the application scope of Multiple Input Multiple Output (MIMO) systems has continued to expand, making MIMO system awareness a key component of electromagnetic spectrum monitoring. However, with the frequent occurrence of illegal frequency usage cases such as black broadcasts, fake base stations, eavesdropping devices, jammers, and radio cheating, malicious electromagnetic interference types continue to emerge, posing numerous challenges to MIMO system awareness. Modulation type recognition is a crucial component of MIMO system awareness. Accurately identifying the modulation type of MIMO signals under malicious interference conditions is a crucial prerequisite for achieving in-depth MIMO system awareness.

[0003] Numerous research results have been achieved in MIMO signal modulation type recognition. Existing technologies can be roughly divided into two categories: decision-theoretic methods and pattern recognition-based methods. Decision-theoretic methods are primarily based on the principles of maximum likelihood ratio and probability theory. As a multiple hypothesis testing theory, their main approach is to statistically analyze signal characteristics through probability theory and other related mathematical theories. Decision thresholds and thresholds are usually set to achieve signal recognition (Nandi AK, Azzouz EE. Algorithms for automatic modulation recognition of communication signals [J]. IEEE Transactions on communications, 1998, 46(4): 431-436.). However, these methods still face the problems of complex theoretical derivation and high computational complexity in practical applications. The method based on pattern recognition is to extract the signal modulation mode features and identify them by building a classifier. At present, classifiers based on neural networks are usually used, which have the advantages of strong generalization ability, high efficiency and excellent performance. For example, signal modulation mode recognition based on deep learning (Xiao W, Luo Z, Hu Q. A review of research on signal modulation recognition based on deep learning [J]. Electronics, 2022, 11 (17): 2764.), modulation mode recognition based on hierarchical deep neural network (Karra K, Kuzdeba S, Petersen J. Modulation recognition using hierarchical deep neural networks [C] / / 2017 IEEE international symposium on dynamic spectrum access networks (DySPAN). IEEE, 2017: 1-3.), etc. Due to the complexity of the spectrum environment and the influence of interference noise on the modulation recognition task, exploring effective methods for modulation mode recognition in interference-noise environment has become an increasingly important proposition. For example, exploring the modulation recognition accuracy of convolutional network under hostile interference (Lin Y, Zhao H, Ma X, et al. Adversarial attacks in modulation recognition with convolutional neural networks[J].IEEE Transactions on Reliability, 2020,70(1):389-401.), modulation recognition under low signal-to-noise ratio based on deep convolutional network and transfer learning (Jiang K, Zhang J, Wu H, et al. A novel digital modulation recognition algorithm based on deep convolutional neural network [J]. Applied Sciences, 2020, 10 (3): 1166.), modulation recognition under channel interference based on multimodal feature fusion (Zhang X, Li T, Gong P, et al. Modulation recognition of communication signals based on multimodal feature fusion [J]. Sensors, 2022, 22 (17): 6539.), etc.

[0004] The current modulation recognition technology in the context of non-cooperative communication usually studies the impact of signal-to-noise ratio changes on recognition accuracy under channel noise such as additive Gaussian noise. There are few studies on recognition methods in the presence of hostile interference (such as suppression interference, deceptive interference, etc.), which is of practical significance. In addition, the currently popular modulation recognition algorithms based on pattern recognition mostly rely on classifiers based on neural networks, and the features usually extracted as input to the classifier are basically one-dimensional feature vectors or two-dimensional feature matrices. The features of multiple signal attributes (such as time, space, and frequency domains) are usually extracted separately or fused, and the multi-attribute and multi-modal characteristics of the data are not actually considered comprehensively.

[0005] Through the above analysis, the problems and defects of the existing technology are as follows:

[0006] Most current modulation recognition technologies fail to account for strong suppressive interference environments and only consider the impact of noise interference. Furthermore, existing suppressive interference suppression requires complex processing, which consumes more spectrum resources and energy. Furthermore, existing modulation type recognition technologies often use low-dimensional features to characterize signal type differences, failing to consider the multidimensional and multi-attribute characteristics of signals, resulting in limited recognition performance.

[0007] The difficulty of solving the above problems and defects is: in order to save resource overhead, a method combining interference suppression and modulation mode is adopted. By utilizing the statistical characteristics of the signal, interference suppression based on tensor decomposition is performed, and tensor analysis is used to characterize the multi-dimensional and multi-attribute characteristics of the modulated signal, so as to achieve efficient, robust and effective MIMO signal modulation type identification.

[0008] The significance of solving the above problems and defects is that accurate identification of MIMO signal modulation type under suppression interference is an important prerequisite for implementing electromagnetic interference investigation and maintenance of electromagnetic spectrum order. Summary of the Invention

[0009] The technical problems to be solved by the present invention are:

[0010] In order to avoid the shortcomings of the existing technology, the present invention provides a method and system for identifying the modulation type of a MIMO system under communication interference. By utilizing tensor decomposition to effectively suppress suppression interference, and utilizing tensor analysis to construct multi-attribute features, the modulation type identification performance is significantly improved, providing technical support for real-time and efficient monitoring of the electromagnetic spectrum.

[0011] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0012] A method for identifying modulation types of a MIMO system under communication interference, characterized by comprising:

[0013] S1: Filter, denoise and whiten the received MIMO signal;

[0014] S2: Perform tensor representation on the preprocessed signal and construct a five-dimensional fourth-order cross-cumulant tensor. Based on the fourth-order cross-cumulant tensor, the separation matrix is ​​estimated through parallel factorization to achieve suppressed interference reconstruction and separation.

[0015] S3: Reconstruct a three-dimensional tensor for the separated and suppressed signal, and use the generalized tensor discriminant analysis algorithm to reduce the dimension and extract features of the tensor;

[0016] S4: Based on Tucker decomposition, the kernel tensor and factor matrix are established for the reduced-dimensional tensor. The similarity of the kernel tensors of different modulation types is identified according to the Euclidean distance criterion to complete the MIMO signal modulation type recognition.

[0017] A further technical solution of the present invention: Step S2 includes:

[0018] The whitened signal matrix is ​​segmented and obtained n=[z n1,z n2,...,zn k ],n i =N / K,i=1,2,3,...K;

[0019] For each signal matrix, the fourth-order cross-cumulant tensor X is calculated according to the fourth-order cross-cumulant formula. k ∈R M×M×M×M , k=1,2,3...K, get K tensors of order M×M×M×M, and concatenate them into a 5-dimensional tensor X∈R M×M×M×M×K ;

[0020] Perform high-order singular value decomposition on the tensor X, select rank R = M, and construct the factor matrix of the initial parallel factor decomposition CP;

[0021] The tensor X is decomposed using the alternating least squares method to obtain the first N factor matrices P1, P2, ..., P N ;

[0022] Select P1, P2, ..., P N , the i-th column of the N matrices constructs an M×N feature matrix T, i=1,2,3…,M, and performs singular value decomposition on the feature matrix T; the eigenvector corresponding to the maximum singular value of the left singular matrix is ​​taken as the actual estimated value of the i-th column of the mixing matrix H1, and the final separation matrix is ​​obtained Then the signal matrix after separation is

[0023] A further technical solution of the present invention: Step S3 includes:

[0024] After each channel of communication signal is separated and interference suppressed, db4 wavelet is used to perform wavelet decomposition at different scales, and C is calculated for the decomposed signal at each scale through a sliding window. 20 ,C 40 ,C 60 Three high-order cumulants are used to construct a three-dimensional tensor sample of wavelet decomposition scale × high-order cumulant × time domain sequence;

[0025] The communication signals of different channels are processed as above to construct a tensor to obtain B samples, and construct χ={χ1,χ2,...χ B A 3D sample set with known category labels, belonging to C categories, each category has B C samples; and there are u n ,n=1,2,3,is the projection matrix of the nth dimension, calculate the total average tensor and the average tensor within each class c=1,2...C;according to the threshold value Q of the nth dimension n Determine the dimension P of the feature subspace after projection n ,n=1,2,3;

[0026] Calculate the n-th modulus divergence difference matrix of the original tensor sample set right Perform feature decomposition and select the first P n The characteristic matrix composed of the eigenvectors corresponding to the eigenvalues ​​is used as the initial projection matrix of the n-th module; the original tensor sample set is projected using the initial projection matrix to find the suboptimal solution set y of the n-th module i n , thereby constructing the inter-class discrete matrix of the suboptimal solution set of this module and the intra-class discrete matrix And get the total divergence difference matrix of the module Among them, ∈ is obtained by experiment; perform eigendecomposition on the total divergence difference matrix and take the first P n The eigenvectors corresponding to the largest eigenvalues ​​form the n-th module projection matrix, and the inter-class discreteness of the tensor set y after projection is calculated and intra-class dispersion Calculate the total divergence difference for each iteration separately:

[0027] Repeat the above operation for each module n, n = 1, 2... N, until the total divergence difference T vd If the difference between the iterations is less than a given value δ, the process is terminated.

[0028] A further technical solution of the present invention: Step S4 includes:

[0029] After projecting the set of tensors with known class labels using these projection matrices, calculate the average tensor within each class:

[0030]

[0031] Use the ALS algorithm to perform Tucker decomposition on the average tensor within each category to obtain the core tensor of each category. c ,c=1,2,...,C, and the corresponding factor matrix U ci , i=1,2,...N, for the unlabeled tensor sample t constructed after denoising separation and GTDA dimensionality reduction, its multiple core tensors are calculated by multiplying them by the transpose of the factor matrix of each category tensor. The Euclidean distances between the multiple core tensors and the core tensors of each category are compared and the sample type with the smallest distance is taken as the estimated type of the tensor.

[0032] A system for identifying modulation types of a MIMO system under communication interference, characterized by comprising:

[0033] The signal preprocessing module is used to process the observed signal using digital filters and whitening operations to provide conditions for tensor decomposition;

[0034] Interference reconstruction and separation module, used to construct tensors for the processed signals and perform parallel factor decomposition to reconstruct and separate the suppressed interference;

[0035] Tensor dimension reduction and feature extraction module, which is used to construct feature tensors for the reconstructed signal and perform dimension reduction and feature extraction using generalized tensor decision analysis;

[0036] The modulation type identification module is used to perform Tucker decomposition on the tensor after dimensionality reduction, construct the core tensor and factor matrix, and use the Euclidean distance between the core tensors to identify the modulation type.

[0037] A computer system, characterized in that it includes: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method.

[0038] A computer-readable storage medium is characterized by storing computer-executable instructions, which are used to implement the above method when executed.

[0039] A computer program product, characterized by comprising computer executable instructions, wherein the instructions are used to implement the above method when executed.

[0040] The beneficial effects of the present invention are:

[0041] The present invention provides a method for identifying the modulation type of a MIMO system under communication interference. The method can utilize the high-order characteristics of the signal to more efficiently perform a blind source separation algorithm, realize the processing of suppressive interference, and simultaneously construct a tensor form of signal samples in multiple dimensions to better characterize the multi-attribute characteristics of the signal, thereby facilitating improved recognition performance. At the same time, the method uses a GTDA+Tucker decomposition algorithm to analyze the similarity between kernel tensors, with low computational cost and short computation time, and still has a certain robustness when there are fewer training samples and a low signal-to-interference ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like parts throughout the drawings.

[0043] Figure 1 This is a flow chart of a method for identifying modulation types of MIMO systems under communication interference provided by an embodiment of the present invention.

[0044] Figure 2 This is a system structure block diagram of a method for identifying modulation types of a MIMO system under communication interference provided by an embodiment of the present invention.

[0045] Figure 3 This is a schematic diagram of the accuracy of identifying the modulation mode of channel noise changes under communication interference provided by an embodiment of the present invention.

[0046] Figure 4 This is a schematic diagram of the accuracy of identifying the modulation mode of interference power changes under communication interference provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only intended to illustrate the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0048] like Figure 1 As shown, the method for identifying the modulation type of a MIMO system under communication interference provided by an embodiment of the present invention includes the following steps:

[0049] S101, performing filtering, denoising and whitening preprocessing on the received MIMO signal;

[0050] S102, constructing a five-dimensional fourth-order cross-cumulant tensor using the preprocessed signal, and estimating a separation matrix through parallel factor decomposition to achieve suppressed interference reconstruction separation;

[0051] S103, reconstructing a three-dimensional tensor for the separated and suppressed signal, and performing dimensionality reduction and feature extraction on the tensor using a generalized tensor discriminant analysis algorithm;

[0052] S104: Tucker decomposition is used to establish a core tensor and a factor matrix for the reduced-dimensional tensor, and the similarity of the core tensors of different modulation types is identified based on the Euclidean distance criterion to complete the MIMO signal modulation type recognition.

[0053] like Figure 2 As shown, the modulation mode identification system of a MIMO system under communication interference provided by an embodiment of the present invention includes:

[0054] The signal preprocessing module is used to process the observed signal using digital filters and whitening operations to provide conditions for tensor decomposition;

[0055] Interference reconstruction and separation module, used to construct tensors for the processed signals and perform parallel factor decomposition to reconstruct and separate the suppressed interference;

[0056] Tensor dimension reduction and feature extraction module, which is used to construct feature tensors for the reconstructed signal and perform dimension reduction and feature extraction using generalized tensor decision analysis;

[0057] The modulation type identification module is used to perform Tucker decomposition on the tensor after dimensionality reduction, construct the core tensor and factor matrix, and use the Euclidean distance between the core tensors to identify the modulation type.

[0058] The present invention will be further described below with reference to the embodiments.

[0059] Example 1

[0060] An embodiment of the present invention provides a method for identifying a modulation type of a MIMO system under communication interference, comprising the following steps:

[0061] The first step is data preprocessing, which involves filtering, denoising, and whitening the received MIMO signal.

[0062] The multi-path noisy mixed signal observed by the MIMO system is filtered using a 40th-order FIR filter. The main purpose of this step is to reduce the Gaussianity of the signal, and then the signal is whitened. First, the M-path instantaneous linear mixed signal passing through the noisy channel is filtered (here the number of observed signals is equal to the number of source signals), and the denoised mixed signal x(n)[x1 n,...,x M n] T ,n=1,2,3...N, N is the number of sampling points, then calculate the covariance matrix of the mixed signal

[0063] C x =E{x(n)x T (n)}

[0064] Perform eigendecomposition on the covariance matrix:

[0065] C x =EDE T

[0066] Compute the whitening matrix:

[0067] V=ED 1 / 2 E T

[0068] Get the whitened signal:

[0069]

[0070] The second step is to construct a five-dimensional fourth-order cross-cumulant tensor using the preprocessed signal and estimate the separation matrix through parallel factorization to achieve the reconstructed separation of the suppressed interference. This includes:

[0071] The signal matrix after whitening is divided into K segments, and we get n=[z n1,z n2,z n1,...,z n2],n i =N / K, i=1,2,3,...K, and then use the following formula to calculate the fourth-order cross-accumulation tensor X for each signal matrix k ∈R M×M×M×M , k=1,2,3...K, the calculation formula is:

[0072] X k =Cum[z n1,z n1,z n1,z n1]

[0073] cum a,b,c,d i,j,k,l =E a i b j c k d l -E a i b j E c k d l -E a i c k E b j d l -E a i d l E b j c k

[0074] The calculated K-order M×M×M×M tensors are concatenated into a 5-dimensional tensor X∈R M×M×M×M×K Then, the tensor decomposition is initialized. First, the tensor X is decomposed by Higher-Order Singular Value Decomposition (HOSVD), the rank R = M is selected, and the initial factor matrix of Canonical Polyadic Decomposition / Parallel Factor Analysis (CP) is constructed. Then, the tensor X is decomposed by Alternating Least Squares (ALS) to obtain the first N factor matrices P1, P2, ..., P N , the fusion method is used to estimate the final mixing matrix H1 for these N matrices. First, take P1, P2, ..., P N , the i-th column of N matrices constructs an M×N feature matrix T;

[0075] Perform singular value decomposition (SVD) on the feature matrix T:

[0076] T=SVD T

[0077] Take the eigenvector of the left singular matrix corresponding to the largest singular value as the actual estimated value of the i-th column of the mixing matrix H1, i = 1, 2, ..., N, and the final separation matrix Then the signal matrix after separation is

[0078] The third step is to reconstruct a three-dimensional tensor for the separated and suppressed signal, and use the generalized tensor discriminant analysis algorithm to reduce the dimension and extract features of the tensor:

[0079] First, the communication signal after each channel is separated and suppressed for interference is decomposed using db4 wavelet at different scales, and C is calculated for the signal at each scale after decomposition through a sliding window. 20 ,C 40 ,C 60 Three high-order cumulants are used to construct a three-dimensional tensor sample of wavelet decomposition scale × high-order cumulant × time domain sequence.

[0080] For B samples, χ={χ1,χ2,...χ B} is an N-dimensional sample set with known category labels. These B samples belong to C categories, and each category has B C samples. n ,n=1,2...N, is the projection matrix of the nth dimension, after projection, the subspace dimension P n ,n=1,2,...,N. Calculate the total average tensor Compute the average tensor within each class c=1,2...C.

[0081]

[0082] Where, Represents the bth tensor sample in the cth class sample. n Determine the dimension of the feature subspace after projection:

[0083]

[0084] in, is the eigenvalue obtained by eigendecomposing the divergence difference matrix of the sample tensor along each mode, I n The module dimensions of the input tensor for the GTDA algorithm, and P n is the scale of the nth mode after projection, here by selecting the front P n eigenvalues ​​to make the Q value meet the requirements, thereby determining P n , and then use Full Projection Truncation (FPT) to obtain the initial projection matrix.

[0085] First, calculate the n-th modulus divergence difference matrix of the original tensor sample set:

[0086]

[0087] Where, χ c n is the input sample χ c The modulo n expansion matrix, and are the modulo n expansion matrix of the total average tensor and the modulo n expansion matrix of the total average tensor within the class corresponding to the b-th sample, and ε is obtained by experiment.

[0088] Afterwards Feature decomposition, select the first P n The characteristic matrix composed of the eigenvectors corresponding to the eigenvalues ​​is used as the initial projection matrix of the n-th module. The original tensor sample set is projected using the initialization matrix to find the suboptimal solution set of the n-th module:

[0089] y i n =χ i ×1u1×2u2...× n-1 u n-1 × n+1 u n+1 ...× N u N

[0090] i=1,2...B

[0091] The matrix is ​​expanded through the suboptimal solution set, and then the inter-class discrete matrix and the intra-class discrete matrix are calculated by the following formulas:

[0092]

[0093] here is the total average matrix of the suboptimal solution after expansion along the module n, It is the total average matrix after the expansion of various types of inner edges modulo n in the suboptimal solution set.

[0094] Thus, the total divergence difference matrix of the module is obtained ∈ is obtained from the experiment, the total divergence difference matrix is ​​decomposed into features, and the first P n The eigenvector corresponding to the largest eigenvalue is used as the n-th module projection matrix. Then, for the tensor set y obtained after projection, the inter-class discreteness is calculated as:

[0095]

[0096] The intra-class dispersion is calculated as:

[0097]

[0098] The “||·||” in the above formula represents the Frobenius norm, and the total divergence difference in each iteration is calculated as:

[0099] For each module n, n = 1, 2...N, the above operation is repeated until the difference in the total divergence before and after the iteration is less than a small amount δ, and the algorithm converges and terminates.

[0100] The fourth step is to use Tucker decomposition to build the core tensor and factor matrix of the reduced dimensionality tensor. The similarity of the core tensors of different modulation types is identified based on the Euclidean distance criterion to complete the MIMO signal modulation type recognition. This includes:

[0101] First, after the dimensionality reduction of the constructed tensor in the third step, since the output of the GTDA algorithm is the iterative projection matrix, the tensor set with known category labels is projected using these projection matrices and the average tensor within each category is calculated:

[0102]

[0103] Use the ALS algorithm to perform Turcker decomposition on the average tensor within each category (referred to as category tensor) to obtain the category core tensor core c , and the corresponding factor matrix U ci , i=1,2,...N, for the unlabeled tensor sample t constructed after denoising separation and GTDA dimensionality reduction, calculate its multiple core tensors:

[0104] core tc =t×1u1 T ×2u2 T ... N u N T ,c=1,2,...C

[0105] The calculated Euclidean distances of multiple core tensors and each category of core tensors:

[0106] dis c =||core tc -core c ||,c=1,2...C

[0107] Compare their sizes and take the sample type with the smallest distance as the estimated type of the tensor.

[0108] The technical effects of the present invention are described in detail below with reference to simulations.

[0109] In order to evaluate the performance of the present invention, a simulation is conducted. A 5×4 MIMO system is used here. The suppression interference is linear frequency sweep interference and the channel is a flat fading channel. The simulation generates four communication signals: 2PSK, 4PSK, 8QAM, and 16QAM. 2000 groups of each type of signal are generated. Each group of signals consists of four communication signals and one interference signal. The number of sampling points for each signal is 10. 5, respectively, at the receiving end, low signal-to-interference ratio and interference signal are mixed, and blind source separation is performed by decomposing the fourth-order mutual accumulation tensor. Then, the separated communication signal is decomposed by wavelet using db4 wavelet. The high-order cumulants are calculated along the sliding window of the wavelet decomposition scale, and a tensor sample with the size of 6×6×186 of wavelet decomposition scale×high-order cumulants×time domain sequence is constructed. Then, the GTDA algorithm is used for dimensionality reduction to obtain a 3×3×83 dimensionality reduction sample. The core tensor of the average tensor within the class is calculated by Tucker decomposition, and the similarity between the core tensor and the test sample is calculated (here the size of the core tensor is specified to be 3×3×3) to make a category judgment. Under different channels, the signal-to-interference ratio remains unchanged, and the classification accuracy obtained by increasing the channel noise power is as follows Figure 2 As shown in the figure, when the low signal-to-noise ratio channel condition is bad, the performance of the blind source separation algorithm deteriorates sharply, and the accuracy of the back-end classification algorithm is an uncertain random value, so the algorithm is considered to be invalid at this time; when the interference power is increased and the channel condition remains unchanged, the modulation mode recognition accuracy is obtained as follows Figure 3 As shown. Figure 3 It can be seen that the classification accuracy of the method of the present invention is still robust to a certain extent when the signal interference is relatively low.

[0110] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented in whole or in part in the form of a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL), or wireless (e.g., infrared, wireless, microwave, etc.)) method. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0111] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for identifying modulation types of a MIMO system under communication interference, characterized in that: include: S1: Filter, denoise and whiten the received MIMO signal; S2: Perform tensor representation on the preprocessed signal and construct a five-dimensional fourth-order cross-cumulant tensor. Based on the fourth-order cross-cumulant tensor, the separation matrix is ​​estimated through parallel factorization to achieve suppressed interference reconstruction and separation. S3: Reconstruct a three-dimensional tensor for the separated and suppressed signal, and use the generalized tensor discriminant analysis algorithm to reduce the dimension and extract features of the tensor; S4: Based on Tucker decomposition, the kernel tensor and factor matrix are established for the reduced-dimensional tensor. The similarity of the kernel tensors of different modulation types is identified according to the Euclidean distance criterion to complete the MIMO signal modulation type recognition.

2. The method for identifying modulation types of a MIMO system under communication interference according to claim 1, wherein: Step S2 includes: The signal matrix after whitening is segmented and zn=[zn1,zn2,...,zn k ],n i =N / K,i=1,2,3,...K; For each signal matrix, the fourth-order cross-cumulant tensor X is calculated according to the fourth-order cross-cumulant formula. k ∈R M×M×M×M , k=1,2,3...K, get K tensors of order M×M×M×M, and concatenate them into a 5-dimensional tensor XR M×M×M×M×K ; Perform high-order singular value decomposition on the tensor X, select rank R = M, and construct the factor matrix of the initial parallel factor decomposition CP; The tensor X is decomposed using the alternating least squares method to obtain the first N factor matrices P1, P2, ..., P N ; Select P1, P2, ..., P N , the i-th column of the N matrices constructs an M×N feature matrix T, i=1,2,3…,M, and performs singular value decomposition on the feature matrix T; the eigenvector corresponding to the maximum singular value of the left singular matrix is ​​taken as the actual estimated value of the i-th column of the mixing matrix H1, and the final separation matrix W=H1 -1 ; Then the separated signal matrix y = Wz.

3. The method for identifying modulation types of a MIMO system under communication interference according to claim 1, wherein: Step S3 includes: After each channel of communication signal is separated and interference suppressed, db4 wavelet is used to perform wavelet decomposition at different scales, and C is calculated for the decomposed signal at each scale through a sliding window. 20 ,C 40 ,C 60 Three high-order cumulants are used to construct a three-dimensional tensor sample of wavelet decomposition scale × high-order cumulant × time domain sequence; The communication signals of different channels are processed as above to construct a tensor to obtain B samples, and construct χ={χ1,χ2,...χ B A 3D sample set with known category labels, belonging to C categories, each with B C samples; and there are u n ,n=1,2,3,is the projection matrix of the nth dimension, calculate the total average tensor and the average tensor within each class According to the threshold value Q of the nth dimension n Determine the dimension P of the feature subspace after projection n ,n=1,2,3; Calculate the n-th modulus divergence difference matrix of the original tensor sample set right Perform feature decomposition and select the first P n The characteristic matrix composed of the eigenvectors corresponding to the eigenvalues ​​is used as the initial projection matrix of the n-th module; the original tensor sample set is projected using the initial projection matrix to find the suboptimal solution set y of the n-th module in , thereby constructing the inter-class discrete matrix of the suboptimal solution set of this module and the intra-class discrete matrix And get the total divergence difference matrix of the module in, ∈ Obtained from the experiment; perform eigendecomposition on the total divergence difference matrix and take the first P n The eigenvectors corresponding to the largest eigenvalues ​​form the n-th module projection matrix, and the inter-class discreteness of the tensor set y after projection is calculated and intra-class dispersion Calculate the total divergence difference for each iteration separately: For each modulo n, n ＝ Repeat the above operation in 1,2...N cycles until the total divergence difference T vd If the difference between the iterations is less than a given value δ, the process is terminated.

4. The method for identifying modulation types of a MIMO system under communication interference according to claim 1, wherein: Step S4 includes: After projecting the set of tensors with known class labels using these projection matrices, calculate the average tensor within each class: Use the ALS algorithm to perform Tucker decomposition on the average tensor within each category to obtain the core tensor of each category. c ,c=1,2,...,C, and the corresponding factor matrix U ci , i=1,2,...N, for the unlabeled tensor sample t constructed after denoising separation and GTDA dimensionality reduction, its multiple core tensors are calculated by multiplying them by the transpose of the factor matrix of each category tensor. The Euclidean distances between the multiple core tensors and the core tensors of each category are compared and the sample type with the smallest distance is taken as the estimated type of the tensor.

5. A MIMO system modulation type identification system under communication interference, characterized in that: include: The signal preprocessing module is used to process the observed signal using digital filters and whitening operations to provide conditions for tensor decomposition; Interference reconstruction and separation module, which is used to construct tensors of processed signals and perform parallel factor decomposition to reconstruct and separate suppressed interference; Tensor dimension reduction and feature extraction module, which is used to construct feature tensors for the reconstructed signal and perform dimension reduction and feature extraction using generalized tensor decision analysis; The modulation type identification module is used to perform Tucker decomposition on the tensor after dimensionality reduction, construct the core tensor and factor matrix, and use the Euclidean distance between the core tensors to identify the modulation type.

6. A computer system, characterized in that include: One or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method of claim 1.

7. A computer-readable storage medium, characterized in that Computer-executable instructions are stored, and when the instructions are executed, they are used to implement the method of claim 1.

8. A computer program product, characterized in that The invention comprises computer executable instructions, which are used to implement the method of claim 1 when the instructions are executed.

Images