A method and system for identifying MIMO signal modulation mode in a strong interference environment

Abstract

The application particularly relates to a MIMO signal modulation mode recognition method and system in a strong interference environment, which comprises the following steps: introducing a tensor analysis theory to construct a high-order cumulative tensor model of MIMO signals in a strong interference environment; combining an enhanced linear search algorithm to design a least square method, performing regular tensor decomposition on the high-order cumulative tensor model to reconstruct the MIMO signals; performing constellation mapping on the reconstructed MIMO signals to generate a constellation diagram, extracting features of the constellation diagram by using a fractional wavelet scattering network; constructing a feature collaborative representation classifier based on a Grassmann manifold to obtain the probability of a MIMO signal modulation category; constructing a target function with a maximum joint probability, converting an optimization problem into a process of finding optimal linear combination coefficients through weight design, and determining the modulation mode of the MIMO signals. The MIMO signal modulation mode recognition method has a good effect and can effectively realize MIMO signal modulation mode recognition in a Gaussian noise and strong interference environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of anti-interference communication technology in wireless communication, and particularly relates to a MIMO signal modulation mode identification method and system in a strong interference environment. BACKGROUND

[0002] With the accelerated iteration and evolution of information technology, the application scenarios of wireless communication technology are increasingly rich, but the system security risks and privacy risks are continuously upgraded. In view of the development of wireless communication link, the security protection strategy of the encryption type communication system cannot effectively resist malicious interference and intrusion threats, and the physical layer security technology fully utilizes the characteristics of the wireless channel and the physical layer technology to realize the privacy transmission of wireless communication, which can be an effective supplement to the encryption type security technology. As a kind of physical layer technology, modulation identification has the ability of malicious interference identification and potential threat early warning, which can provide an important basis for physical layer security strategy generation. For example, in the Internet of Things, the modulation identification technology can identify large-scale unknown attacks suffered by devices, assist the generation of physical layer security strategy, realize the dynamic adjustment of communication strategy, and avoid the threat attacks suffered by Internet of Things devices.

[0003] The current modulation identification method can effectively realize MIMO signal modulation identification in complex electromagnetic environments such as unknown noise, local communication interference and time-frequency asynchronous transmission. However, with the rapid popularization of frequency using devices and the continuous enrichment of frequency using business, malicious interference at the same frequency is emerging. Malicious communication interference causes MIMO signal features to be confused, which seriously restricts the recognition accuracy of the algorithm. Therefore, how to realize effective identification in the complex communication scene of insufficient prior information, inaccurate parameter estimation and strong same-frequency interference is a challenge for modulation identification research.

[0004] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present application, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The present application provides a MIMO signal modulation mode identification method and system in a strong interference environment, which can realize MIMO signal modulation mode identification in Gaussian noise and strong interference environment, and can overcome the defects in the prior art to some extent.

[0006] Other characteristics and advantages of the present application will become apparent from the following detailed description, or will be learned by practice of the present application.

[0007] According to a first aspect of the present application, a MIMO signal modulation mode identification method in a strong interference environment is provided, the method comprising:

[0008] The tensor analysis theory is introduced to construct a high-order cumulative tensor model of MIMO signals in a strong interference environment.

[0009] A least square method is designed in combination with an enhanced linear search algorithm, a regular tensor decomposition is performed on the high-order cumulative tensor model to reconstruct the MIMO signals.

[0010] A constellation diagram is generated by mapping the reconstructed MIMO signals to the constellation diagram, and a fractional-order wavelet scattering network is used to extract the fractional-order scattering features of the constellation diagram.

[0011] A feature collaborative representation classifier based on Grassmann manifold is constructed to obtain the probability of the modulation category of the MIMO signals.

[0012] A target function is constructed based on the maximum joint probability, and the optimization problem is converted into a process of finding the optimal linear combination coefficient by weight design, so as to determine the modulation mode of the MIMO signals.

[0013] In some example embodiments, the tensor analysis theory is introduced to construct a high-order cumulative tensor model of MIMO signals in a strong interference environment, wherein when the high order is four, the high-order cumulative tensor model comprises:

[0014] A four-order cumulative tensor of a group of received mixed signals is constructed Q Y Each element Q Y [i,j,k,l] is defined as:

[0015]

[0016] Let The four-order cumulative tensor relationship between the received signals and the channel matrix is as follows:

[0017]

[0018] Wherein, i, j, k, and l represent the serial numbers of the receiving antennas, y i (t), y j (t), y k (t), and y l (t) represent the signal vectors on the receiving antennas, h in , h jn , h kn , and h ln represent the attenuation of the propagation channels between the receiving antennas, M is the number of receiving antennas, (·) * represents a complex conjugate, x n (t) is the nth source signal.

[0019] In some example embodiments, the method further comprises reducing the data processing amount by dimension reduction of the constructed fourth-order tensor by truncated multi-linear singular value decomposition before performing the regular tensor decomposition.

[0020] In some example embodiments, the regular tensor decomposition of the high-order cumulant tensor model to reconstruct the MIMO signal comprises:

[0021] defining a cost function of the tensor decomposition:

[0022]

[0023] wherein, denotes the Khatri-Rao product, ||·||Fdenotes the F-norm, F denotes the F-norm, denotes the expansion of the tensor according to Mode-4, denotes the factor matrix calculated by the t-th iteration;

[0024] According to the cost function of the tensor decomposition, the reconstruction error is calculated to evaluate the fitting degree of the model until the iteration termination condition of the algorithm is reached, and the final factor matrices B, C, D, and E of the tensor decomposition are obtained.

[0025] The factor matrix B is taken as the estimated channel matrix, and the source signal is recovered by inverting the estimated channel matrix:

[0026]

[0027] wherein, (·) -1 denotes the matrix inversion, is a wireless communication channel matrix under the influence of phase bias, and w(t) is additive white Gaussian noise independent of the source signal.

[0028] In some example embodiments, the constellation mapping of the reconstructed MIMO signal to generate a constellation comprises:

[0029] The reconstructed MIMO signal is mapped to a two-dimensional constellation coordinate, and each signal component is represented by a pixel point of an image, and a constellation is generated:

[0030] The three-layer fractional wavelet scattering convolution network is constructed: the modulus of the fractional wavelet coefficient of the signal constellation is calculated based on the constellation diagram, and the spatial local average result of the first layer is output; the modulus of the fractional wavelet coefficient of the first layer is taken as input, the modulus of the fractional wavelet coefficient of the second layer is calculated, and the result of the spatial local average of the second layer is output; the modulus of the fractional wavelet coefficient of the second layer is taken as input, the modulus of the fractional wavelet coefficient of the third layer is calculated, and the result of the spatial local average of the third layer is output; the sum in the spatial dimension and the sum in the scale dimension are summed for each layer output, respectively, and the fractional scattering features of the constellation diagram are obtained by cascading the three-layer outputs.

[0031] In some example embodiments, the feature based on the Grassmann manifold is constructed to represent the classifier, and the probability of the MIMO signal modulation category is obtained, including:

[0032] The singular value decomposition is performed on the fractional scattering features to construct a training set S A and a test set S B as a point set on the Grassmann manifold G(q, d); the points on the Grassmann manifold are embedded into the symmetric matrix space Φ(S A ) = S A S A T and Φ(S B ) = S B S B T The sample points a in the symmetric space can be cooperatively represented by the samples in the set:

[0033]

[0034] wherein α is a coefficient vector, representing the weight of each sub-feature set in the combination;

[0035] For a to-be-tested sample b = S b S b T The probability that b belongs to the kth category is represented as:

[0036]

[0037] wherein v1 and v2 are positive numbers, and n k is the number of the kth modulation feature set.

[0038] In some example embodiments, the target function is constructed with the maximum joint probability, the optimization problem is converted into a process of finding the optimal linear combination coefficient through weight design, and the modulation mode of the MIMO signal is determined, including:

[0039] with the maximum joint probability The modulation mode of the MIMO signal is determined by constructing an objective function, converting the optimization problem into a process of finding optimal linear combination coefficients through weight design, and maximizing a joint probability:

[0040] The maximum joint probability under the collaborative framework can be expressed as:

[0041]

[0042] The modulation class of the test sample b can be predicted by adjusting λ1 and λ2 to achieve the best classification performance:

[0043]

[0044] A MIMO signal modulation mode recognition system in a strong interference environment comprises:

[0045] A high-order cumulant tensor construction module is configured to introduce a tensor analysis theory to construct a high-order cumulant tensor model of the MIMO signal in the strong interference environment.

[0046] An interference signal suppression module is configured to combine an enhanced linear search algorithm to design a least square method, perform regular tensor decomposition on the high-order cumulant tensor model, and reconstruct the MIMO signal.

[0047] A reconstructed signal fractional order graph feature extraction module is configured to perform constellation mapping on the reconstructed MIMO signal to generate a constellation graph, and extract fractional order scattering features of the constellation graph by using a fractional order wavelet scattering network.

[0048] A Grassmann manifold mapping module is configured to construct a feature collaborative representation classifier based on a Grassmann manifold to obtain a probability of the modulation class of the MIMO signal.

[0049] A modulation mode recognition module is configured to construct an objective function by maximizing a joint probability, convert an optimization problem into a process of finding optimal linear combination coefficients through weight design, and determine the modulation mode of the MIMO signal.

[0050] According to a second aspect of the present application, a storage medium having a computer program stored thereon is provided, and the computer program is executed by a processor to implement the MIMO signal modulation mode recognition method in a strong interference environment according to the first aspect.

[0051] According to a third aspect of the present application, a computer program product having a computer program stored thereon is provided, and the computer program is executed by a processor to implement the MIMO signal modulation mode recognition method in a strong interference environment according to the first aspect.

[0052] According to a fourth aspect of the present application, an electronic device is provided, comprising:

[0053] a processor; and

[0054] Memory for storing the executable instructions of the processor;

[0055] The processor is configured to implement the MIMO signal modulation mode identification method under strong interference environment described in the first aspect by executing the executable instructions.

[0056] The MIMO signal modulation pattern identification method provided by embodiments of the present invention introduces tensor analysis theory to construct a high-order cumulative tensor model of MIMO signals under strong interference. It then combines an enhanced linear search algorithm with a least-squares-based tensor decomposition method to achieve strong interference separation and MIMO signal reconstruction. Based on this, constellation diagram mapping is performed on the reconstructed MIMO signal, and a feature representation model is designed using a fractional-order wavelet scattering network to fully explore the deep discriminative features of the MIMO signal map domain. A feature collaborative representation classifier based on the Grassmann manifold is constructed to enhance the differences between MIMO signal modulation categories. An objective function is constructed with the maximum joint probability, transforming the optimization problem into finding the optimal linear combination coefficients, thereby achieving MIMO signal modulation pattern identification under strong interference. The MIMO signal modulation pattern identification method of the present invention performs well and can effectively identify MIMO signal modulation patterns under Gaussian noise and strong interference environments. When the signal-to-noise ratio is higher than 0 dB, the correct identification probability of the modulation pattern reaches over 90%, and the method also exhibits good performance for different types of intentional interference.

[0057] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0058] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0059] Figure 1 This is a flowchart of the MIMO signal modulation mode identification method provided in the embodiments of the present invention.

[0060] Figure 2 This is a block diagram of a method for identifying MIMO signal modulation modes under strong interference conditions provided in an embodiment of the present invention;

[0061] In the diagram: 1-High-order cumulative tensor construction module; 2-Interference signal suppression module; 3-Reconstructed signal fractional-order graph feature extraction module; 4-Glassman manifold mapping module; 5-Modulation mode identification module.

[0062] Figure 3 This is a schematic diagram illustrating the performance of MIMO signal modulation method identification under strong interference conditions provided in an embodiment of the present invention.

[0063] Figure 4 This is a schematic diagram illustrating the recognition performance of different modulated MIMO signals under strong interference conditions, provided by an embodiment of the present invention.

[0064] Figure 5 This is a schematic diagram illustrating the impact of different interference suppression methods on the performance of the strong interference environment provided in the embodiments of the present invention. Detailed Implementation

[0065] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0066] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0067] In related technologies, traditional modulation recognition techniques are mainly divided into likelihood function-based and feature parameter-based methods. These methods either treat the recognition problem as a multivariate hypothesis testing problem or have high computational complexity and require known noise distribution. To improve recognition performance, many scholars have conducted extensive research in the field of modulation recognition by combining artificial intelligence technology.

[0068] Existing technologies propose a cooperative modulation recognition method for MIMO systems based on convolutional neural networks (CNNs), employing cooperative decision rules to determine the modulation mode of the MIMO system. Two low-complexity deep neural network modulation recognition schemes are also proposed: a sparse coding network based on deep neural networks and a radial basis function network. These two schemes respectively utilize a restricted-domain quasi-Newton method and a least-squares method to optimize network weights, improving the performance of deep network modulation recognition. Furthermore, an intelligent modulation recognition scheme for MIMO-OFDM systems is proposed based on a 3D convolutional network architecture. This involves introducing a data transformation method to construct a high-dimensional data matrix and using a 3D convolutional layer to design a deep learning network to extract deep signal features, thereby improving the performance of MIMO signal modulation recognition. A joint decision-making scheme for modulation recognition and direction-of-arrival estimation based on multi-task CNNs is also proposed. This scheme designs a deep learning network by connecting multiple residual modules and considers different Y-shaped connections to achieve optimal recognition. Finally, a semi-supervised modulation recognition scheme based on transfer learning is proposed. This scheme utilizes convolutional autoencoders and convolutional neural networks to construct a deep learning network and employs a transfer learning strategy to train the network weights. It also proposes to extract signal features using causal convolution and structural reparameterization techniques, and dynamically adjust feature weights to maximize the retention of useful features, thereby achieving modulation recognition.

[0069] Existing methods have made significant strides in feature extraction and classifier design; however, most of these methods assume that the wireless channel is an ideal MIMO channel. For time-frequency asynchronous transmission channels, existing techniques propose preprocessing the received signal to compensate for time-frequency offsets, and then using normalized higher-order cumulants in the frequency domain to construct features, thereby completing modulation identification for MIMO-OFDM systems. Existing techniques also propose using a portion of the signal to estimate carrier frequency offset compensation coefficients, reconstructing the received signal symbols, and combining asymmetric convolution to extract intra- and inter-symbol correlations for modulation classification. Furthermore, existing techniques propose a dual-stream network, with each stream consisting of a CNN, LSTM, and an attention module, which can classify higher-order PSK and QAM modulations under time-frequency asynchronous and unknown channel conditions.

[0070] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0071] (1) In existing MIMO signal modulation identification methods, due to the high power interference covering the communication transmission frequency band, strong interference severely damages the statistical characteristics of MIMO signals, resulting in a serious degradation of the performance of existing MIMO signal modulation identification methods under strong interference environment.

[0072] (2) Existing technologies do not involve the mining and utilization of fractional graph domain features.

[0073] The challenges in addressing these issues and shortcomings lie in the following: Under common strong interference environments, signals capable of covering specific communication frequency bands often exhibit statistical characteristics that can be confused with those of MIMO signals. Therefore, constructing a high-order cumulative tensor model that can suppress strong interference signals and capture richer higher-order statistical information in the modulated signal is crucial for reconstructing the MIMO signal. Furthermore, mining the grayscale constellation features of the reconstructed signal using a fractional-order wavelet scattering network serves as the discriminative feature set. During classification, the feature set is modeled onto a Grassmann manifold. Introducing probabilistic collaborative representation transforms the identification problem into an optimal problem, thus overcoming the technical difficulties in identifying MIMO signal modulation schemes.

[0074] To address the shortcomings and deficiencies of existing technologies, this example embodiment provides a method for identifying MIMO signal modulation patterns under strong interference environments. This method can provide technical support for intelligent wireless systems, not only expanding the application scenarios of intelligent wireless systems but also effectively improving their anti-interference capabilities.

[0075] refer to Figure 1 As shown, the method for identifying MIMO signal modulation modes under strong interference conditions may specifically include the following steps:

[0076] S101, Introducing tensor analysis theory to construct a high-order cumulative tensor model of MIMO signals under strong interference conditions.

[0077] S102, combining an enhanced linear search algorithm with a least squares method, performs regular tensor decomposition to separate strong interference and reconstruct the MIMO signal.

[0078] S103, Perform constellation diagram mapping on the reconstructed MIMO signal to generate constellation diagram Ω. S A feature mining and characterization model is designed using a fractional wavelet scattering network to fully extract the deep discriminative features S of the MIMO signal constellation diagram;

[0079] S104, Construct a feature co-representation classifier based on Grassman manifold to enhance the differences between MIMO signal modulation categories;

[0080] S105, with the maximum joint probability By constructing an objective function and designing weights, the optimization problem is transformed into a process of finding the optimal linear combination coefficients, thereby determining the modulation scheme of the MIMO signal.

[0081] The following will describe in more detail each step of the MIMO signal modulation mode identification method under strong interference environment in this example embodiment, with reference to the accompanying drawings and embodiments.

[0082] In step S101, tensor analysis theory is introduced to construct a high-order cumulative tensor model of MIMO signals under strong interference conditions.

[0083] Specifically, a non-cooperative MIMO system under strong interference conditions consists of K transmit antennas and M receive antennas, where s(t) = [s1(t), s2(t), ..., s...]. K (t)] T This indicates that K transmitting antennas are at t th The message symbols transmitted at each moment, [J1(t), J2(t), ..., J N-K (t)] T Let M ≥ N represent NK interference signals. Then, the N source signals include K communication signals and NK interference signals, as follows:

[0084]

[0085] in,(·) T This indicates transpose.

[0086] After transmission through a MIMO wireless communication channel, the received signal may exhibit random phase deviation. The received signal at the receiver can be expressed as:

[0087]

[0088] Among them, y(t)=[y1(t),y2(t),...,y M (t)] T Each of its elements y i w(t) represents the signal vector on each receiving antenna; w(t) = [w1(t), w2(t), ..., w M (t)] T H represents additive white Gaussian noise independent of the source signal; H is an M×N dimensional complex channel matrix, and H={h ij}, 1≤i≤M, 1≤j≤N, where each element h ij This represents the attenuation of the propagation channel between the j-th source and the i-th receiving antenna; Φ is the phase error matrix, and each element... The π represents the phase difference in channel gain between different antennas; ⊙ represents the Hadamard product. It is the wireless communication channel matrix under the influence of phase deviation.

[0089] For ease of observation, the signals on each receiving antenna of the MIMO system receiver are represented in vector form:

[0090]

[0091] Among them, each received signal yi (t)(1≤i≤M) are linear combinations of N source signals, and contain both communication interference and channel noise.

[0092] Strong interference is characterized by three common types of communication interference:

[0093] Linear frequency sweep interference:

[0094]

[0095] Where A is the amplitude of the linear sweep interference signal, f0 is the starting frequency, and k is the sweep slope, also called the sweep coefficient. The initial phase is T, and the duration of the interference signal is T.

[0096] Partial frequency band noise interference:

[0097]

[0098] Among them, U n (t) is a vector with mean zero and variance . Gaussian noise, f c This refers to the center frequency of the partial frequency band noise interference signal. This is the initial phase.

[0099] Noise FM interference:

[0100]

[0101] Among them, U j (t) is the amplitude of the noise frequency-modulated interference signal, f j K is the center frequency of the noise FM interference signal. FM Here, τ is the frequency modulation coefficient, and u(τ) is narrowband Gaussian white noise with a mean of 0 and a variance of 0. Belongs to the Wiener process and obeys distributed.

[0102] Construct a set of fourth-order cumulative tensors for the mixed signals at the receiver. Q Y Each element Q Y [i,j,k,l] is defined as:

[0103]

[0104] Since the source signals are statistically independent, the fourth-order cumulative tensor of the mixed signal can be written as:

[0105]

[0106] make The fourth-order cumulative tensor relationship between the received signal and the channel matrix is ​​obtained as follows:

[0107]

[0108] Among them, (v) * Let x represent the complex conjugate, i,j,k,l represent the receiving antenna numbers, and x represent the receiving antenna numbers. n (t) represents the nth source signal.

[0109] In step S102, the least squares method is designed in conjunction with the enhanced linear search algorithm to perform regular tensor decomposition to separate strong interference and reconstruct the MIMO signal.

[0110] Specifically, the standard decomposition of the constructed fourth-order cumulative tensor model can be expressed as:

[0111]

[0112] Where, B = [b1 b2…b N C = [c1 c2…c N ]、D=[d1 d2…d N ] and E = [e1 e2…e N ] are the factor matrices obtained after tensor decomposition. The channel matrix can be obtained by solving the matrix corresponding to each dimension of the tensor.

[0113] Before performing tensor regularization, the constructed fourth-order tensor is first reduced in dimensionality by truncating multilinear singular value decomposition. The matrices of the fourth-order tensor Q4 expanded in Mode-k form are denoted as follows: Calculate the singular value decompositions of matrices C1, C2, C3, and C4 respectively:

[0114]

[0115] Where, ∑ i (i = 1, 2, 3, 4) represents a diagonal matrix composed of eigenvalues.

[0116] Extract the first p columns of each of the four left singular matrices U1, U2, U3, and U4, and denote them as follows: and Constructing the kernel tensor using a truncated left singular matrix:

[0117]

[0118] Among them, × k This represents the k-modal product.

[0119] nuclear tensor As input to the tensor decomposition method, the factor matrix is ​​initialized and updated. A linear search strategy is used to determine the optimal step size, and linear interpolation is performed on the currently estimated unknown factor matrix. The interpolation matrix can be represented as:

[0120]

[0121] in, Let represent the estimated factor matrix obtained in the (t-1)th iteration, and μ represent the step size in the search direction, i.e., the relaxation factor.

[0122] Define the cost function for tensor decomposition:

[0123]

[0124] in, Denotes the Khatri-Rao product, ||·|| F Describing the F-norm, This indicates that the tensor is expanded using Mode-4. Therefore, the optimal relaxation factor is...

[0125] Complete the update of the factor matrix:

[0126]

[0127] in, This indicates the Moore-Penrose pseudo-inverse.

[0128] Among them, by b i The factor matrix B contains channel matrix information. Therefore, factor matrix B is used as the estimated channel matrix. The source signal is recovered by inverting the estimated channel matrix, thus suppressing communication interference.

[0129]

[0130] in,(·) -1 This represents finding the inverse of a matrix.

[0131] In step S103, constellation mapping is performed on the reconstructed MIMO signal to generate constellation diagram Ω. S A feature mining and characterization model is designed using a fractional wavelet scattering network to fully extract the deep discriminative features S of the MIMO signal constellation diagram.

[0132] Specifically, the MIMO signal will be reconstructed. Mapped to two-dimensional constellation coordinates:

[0133]

[0134] in, This represents the scale factor that defines the range of the signal, and imageSize is the size of the output image.

[0135] Represent the coordinates of each signal component using image pixels to generate a constellation diagram:

[0136]

[0137] Given the constellation Ω S A three-layer fractional wavelet scattering convolutional network is constructed: In the construction of the first layer, the modulus of the fractional wavelet coefficients of the signal constellation diagram is calculated. In the formula, U β [λ1] denotes the first-level fractional-order scattering propagation operator, Θ β Let β denote the fractional convolution operator, where β represents the fractional order. Represent the first-level wavelet basis functions and output the spatial local averaging results of the first level. In the formula, Represents the scattering characteristics of the first layer, φ J This represents the scaling function used for local smoothing. The second-level fractional wavelet coefficient modulus is calculated using the first-level fractional wavelet coefficient modulus as input. And output the result S of the second-level spatial local averaging. (2) =S β [λ1]Ω S =U β [λ1]Ω S Θ β φ J Similarly, the modulus of the fractional wavelet coefficients of the third layer of the scattering network can be obtained as follows: And output the result S of the third-level spatial local averaging. (3) =S β [λ1,λ2]Ω S =U β [λ1,λ2]Ω S Θ β φ J Finally, the features output from each layer are summed in the spatial dimension. In the formula, This represents the value of the m-th order scattering feature at the i, j spatial location, summed with the scale dimension. Aggregation yields a single feature vector, and cascading the outputs of three layers yields the overall output feature of the entire fractional-order scattering network.

[0138]

[0139] In step S104, a feature co-representation classifier based on Grassman manifold is constructed to enhance the differences between MIMO signal modulation categories.

[0140] Specifically, feature mapping is performed based on the Grassmann manifold. The training set S is constructed using singular value decomposition. A With test set S B As the set of points on the Grassman manifold G(q,d); embed the points on the Grassman manifold into the symmetric matrix space Φ(S A ) = S A S A T and Φ(S) B ) = S B S B T A sample point 'a' in the symmetric space spanned by the entire training set can be cooperatively represented by all samples in the set:

[0141]

[0142] Where α is a coefficient vector, representing the weight of each sub-feature set in the combination.

[0143] For the sample to be tested, b = S b S b T The probability that it belongs to the k-th class can be expressed as:

[0144]

[0145] Where v1 and v2 are positive numbers, and n k It is the number of modulation feature sets of the kth class.

[0146] In step S105, with the maximum joint probability By constructing an objective function and designing weights, the optimization problem is transformed into a process of finding the optimal linear combination coefficients, thereby determining the modulation scheme of the MIMO signal.

[0147] Specifically, for different subspaces k, maximizing the probability of P(l(b)=k) is different, which can lead to classification instability and affect the discriminative performance when inter-class differences are not significant. Collaborative representation integrates all class information into a single framework by maximizing the joint probability. Since l(b)=k are independent, maximizing the joint probability under the collaborative framework can be expressed as:

[0148]

[0149] Where λ1=v1, λ2=v2 / K.

[0150] Optimal classification performance can be achieved by adjusting λ1 and λ2. The predicted modulation category for test sample b is:

[0151]

[0152] The technical effects of the present invention will be described in detail below with reference to simulation.

[0153] To evaluate the performance of this invention, simulation verification was performed. The non-cooperative MIMO system had 4 transmit antennas (K=4) and 6 receive antennas (M=6), with modulation schemes primarily involving QPSK, 8PSK, 16PSK, 8QAM, and 16QAM. The signal length N was 10000. Three types of interference signals were considered: linear frequency sweep interference, partial-band noise interference, and noise-frequency modulation interference. The power ratio of the interference signal to the communication signal was set to 1:1. The maximum scale of the fractional-order wavelet scattering network was initialized to 2. J =16, the number of angles is initialized to 8, one fractional-order parameter α1 is fixed at 1, and the other parameter takes a value every 0.05 between 0 and 1. The regularization parameters of the probabilistic collaborative classifier are λ1 = 3 and λ2' = 0.2. This invention uses the average relative error of channel matrix estimation. Correct recognition rate of modulated signal As evaluation metrics, interference suppression simulation experiments employed 800 iterations of statistical simulation, while modulation recognition simulations used 200 cross-validations to verify performance. The proposed modulation scheme recognition method is compared and analyzed with existing methods (Fr-SVM, CNN, Quasi-Newton), and the simulation results are as follows. Figure 3 As shown. By Figure 3 It can be seen that the method of this invention has strong adaptability to strong interference environments, and compared with existing algorithms, the method of this invention has significant performance advantages. The modulation scheme identification method proposed in this invention is compared and analyzed for different modulation schemes, and the simulation results are as follows: Figure 4 As shown. By Figure 4 It can be seen that the method of the present invention has a strong adaptability to strong interference environments. Figure 5 The effects of different types of suppression interference on the interference signal suppression method proposed in this invention are presented. Figure 5 It can be seen that the method proposed in this invention is robust and adaptable to common suppression interference signals.

[0154] It should be noted that, as another aspect, this application also provides a MIMO signal modulation mode identification system under strong interference environment, characterized by comprising:

[0155] The higher-order cumulative tensor construction module is used to introduce tensor analysis theory to construct a higher-order cumulative tensor model of MIMO signals under strong interference conditions.

[0156] The interference signal suppression module is used to combine the enhanced linear search algorithm to design the least squares method to perform regular tensor decomposition on the high-order cumulative tensor model to reconstruct the MIMO signal.

[0157] The reconstructed signal fractional-order graph feature extraction module is used to generate a constellation graph by mapping the reconstructed MIMO signal into a constellation graph, and to extract the fractional-order scattering features of the constellation graph using a fractional-order wavelet scattering network.

[0158] The Grassmann manifold mapping module is used to construct a feature co-representation classifier based on the Grassmann manifold to obtain the probability of the modulation category of MIMO signals;

[0159] The modulation scheme identification module is used to construct the objective function with the maximum joint probability. Through weight design, the optimization problem is transformed into the process of finding the optimal linear combination coefficients, thereby determining the modulation scheme of the MIMO signal.

[0160] In one embodiment, this application also provides a storage medium, which may be included in an electronic device or may exist independently and not assembled into the electronic device. The storage medium carries one or more programs that, when executed by an electronic device, cause the electronic device to perform the methods described in the following embodiments. For example, the electronic device may perform... Figure 1 The steps of the method shown.

[0161] In one embodiment, this application provides a computer program product including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0162] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0163] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0164] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is defined only by the appended claims.

Claims

1. A method for identifying MIMO signal modulation schemes under strong interference conditions, characterized in that, The method includes: A high-order cumulative tensor model of MIMO signals under strong interference environment is constructed by introducing tensor analysis theory. A least squares method is designed by combining an enhanced linear search algorithm to perform regular tensor decomposition on a high-order cumulative tensor model to reconstruct the MIMO signal. Constellation diagrams are generated by mapping the reconstructed MIMO signals to constellation diagrams, and fractional wavelet scattering networks are used to extract the fractional scattering features of the constellation diagrams. Construct a feature co-representation classifier based on the Grassman manifold to obtain the probability of MIMO signal modulation category; The objective function is constructed with the maximum joint probability. By designing weights, the optimization problem is transformed into the process of finding the optimal linear combination coefficients, thereby determining the modulation scheme of the MIMO signal.

2. The method according to claim 1, characterized in that, The method introduces tensor analysis theory to construct a high-order cumulative tensor model of MIMO signals under strong interference conditions. When the higher order is fourth order, it includes: Construct a set of fourth-order cumulative tensors for the mixed signals at the receiver. , Each element Defined as: ； make The fourth-order cumulative tensor relationship between the received signal and the channel matrix is ​​obtained as follows: ； in, This is represented by the receiving antenna number. This represents the signal vector on the receiving antenna. This indicates the attenuation of the propagation channel between receiving antennas. M The number of receiving antennas, To indicate complex conjugate, For the first Individual source signals.

3. The method according to claim 2, characterized in that, The method also includes reducing the dimensionality of the constructed fourth-order tensor by truncating multilinear singular value decomposition before performing regular tensor decomposition, thereby reducing the amount of data processing.

4. The method according to claim 3, characterized in that, The method of performing regularized tensor decomposition on a high-order cumulative tensor model to reconstruct the MIMO signal includes: Define the cost function for tensor decomposition: in, Represents the Khatri-Rao product. Describing the F-norm, This indicates that the tensor is expanded in Mode-4. , , , Indicates the first The factor matrix obtained from the next iteration; Based on the cost function of tensor decomposition, the reconstruction error is calculated to evaluate the model's fit until the algorithm's iterative termination condition is met, yielding the final factor matrix of the tensor decomposition. , , , ; factor matrix As the estimated channel matrix, the source signal is recovered by inverting the estimated channel matrix: ； in, This represents finding the inverse of a matrix. The wireless communication channel matrix under the influence of phase deviation. It is additive white Gaussian noise independent of the source signal.

5. The method according to claim 4, characterized in that, The process of generating a constellation diagram by mapping the reconstructed MIMO signal to a constellation diagram and extracting the fractional-order scattering features of the constellation diagram using a fractional-order wavelet scattering network includes: Reconstructing the MIMO signal Mapping to two-dimensional constellation coordinates, and then representing the coordinates of each signal component using image pixels, generates a constellation diagram: A three-layer fractional wavelet scattering convolutional network is constructed: the magnitude of the fractional wavelet coefficients of the signal constellation diagram is calculated based on the constellation diagram, and the spatial local averaging result of the first layer is output; the magnitude of the fractional wavelet coefficients of the second layer is calculated as input using the magnitude of the first layer fractional wavelet coefficients, and the spatial local averaging result of the second layer is output; the magnitude of the fractional wavelet coefficients of the third layer is calculated as input using the magnitude of the second layer fractional wavelet coefficients, and the spatial local averaging result of the third layer is output; the outputs of each layer are summed in both the spatial dimension and the scale dimension, and the three layers of outputs are cascaded to obtain the fractional scattering features of the constellation diagram.

6. The method according to claim 5, characterized in that, The construction of a feature cooperative representation classifier based on the Grassmann manifold to obtain the probability of MIMO signal modulation category includes: Singular value decomposition is performed on fractional-order scattering features to construct a training set. With test set As Grassmann manifold A set of points on a Grassman manifold; embedding points on a Grassman manifold into a symmetric matrix space. and Sample points in symmetric space It can be represented collaboratively by samples in the set: ； in, It is a coefficient vector, representing the weight of each sub-feature set in the combination; For the sample to be tested It belongs to the first The probability of a class is expressed as: ； in, , It is a positive number. It is the first The number of modulation feature sets.

7. The method according to claim 6, characterized in that, The process of constructing the objective function with the maximum joint probability, and transforming the optimization problem into finding the optimal linear combination coefficients through weight design, to determine the modulation scheme of the MIMO signal, includes: With maximum joint probability By constructing the objective function and designing weights, the optimization problem is transformed into finding the optimal linear combination coefficients, thus determining the modulation scheme of the MIMO signal. The maximization of joint probability under the collaborative framework can be expressed as: ； ； It can be adjusted and To achieve optimal classification performance, test samples The modulation category is predicted as follows: 。 8. A MIMO signal modulation mode identification system under strong interference environment, characterized in that, include: The higher-order cumulative tensor construction module is used to introduce tensor analysis theory to construct a higher-order cumulative tensor model of MIMO signals under strong interference conditions. The interference signal suppression module is used to combine the enhanced linear search algorithm to design the least squares method to perform regular tensor decomposition on the high-order cumulative tensor model to reconstruct the MIMO signal. The reconstructed signal fractional-order graph feature extraction module is used to generate a constellation graph by mapping the reconstructed MIMO signal into a constellation graph, and to extract the fractional-order scattering features of the constellation graph using a fractional-order wavelet scattering network. The Grassman manifold mapping module is used to construct a feature co-representation classifier based on the Grassman manifold to obtain the probability of the modulation category of MIMO signals; The modulation scheme identification module is used to construct the objective function with the maximum joint probability. Through weight design, the optimization problem is transformed into the process of finding the optimal linear combination coefficients, thereby determining the modulation scheme of the MIMO signal.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the MIMO signal modulation mode identification method under strong interference environment as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the MIMO signal modulation mode identification method under strong interference environment as described in any one of claims 1 to 7.

Images