A method for recognizing a composite modulated signal based on structured decomposition
By constructing a phase difference matrix and performing low-rank sparse decomposition, combined with a convolutional neural network, the problem of recognizing composite modulation signals was solved, and robust recognition and noise suppression of composite modulation signals were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIAN INSTITUE OF SPACE RADIO TECH
- Filing Date
- 2024-06-28
- Publication Date
- 2026-07-07
AI Technical Summary
Existing phase-based modulation scheme identification methods are sensitive to noise and have difficulty effectively identifying composite modulation signals, especially phase-coded modulation signals. Furthermore, existing denoising methods are prone to causing signal distortion.
A structured decomposition-based method is adopted to construct a phase difference matrix through phase difference sequences, remove noise components using low-rank sparse decomposition, and input the low-rank sparse components into a multi-layer convolutional neural network for recognition. The multi-layer convolutional neural network is then constructed for training and testing.
It improves the recognition performance of composite modulation signals, reduces the impact of noise on recognition, is applicable to both conventional and composite modulation signals, and has robustness and universality.
Smart Images
Figure CN118861513B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of signal processing technology, particularly the field of radar electronic reconnaissance, and relates to a method for identifying complex modulation signals based on structured decomposition, which is used for robust identification of complex modulation signals in radar electronic reconnaissance. Background Technology
[0002] Composite modulation signals refer to signals jointly modulated by linear modulation, nonlinear frequency modulation, and frequency-coded signals combined with phase-coded modulation. For linear frequency modulation, nonlinear frequency modulation, and frequency-coded signals, the phase changes continuously. However, for phase-coded signals, the signal phase changes discretely, exhibiting abrupt phase changes. For single phase-coded modulation types, namely binary phase-coded signals (BPSK) and quadrature phase-coded signals (QPSK), they can be equivalent to the combination of conventional pulses and phase-coded signals.
[0003] Phase-based modulation scheme identification methods are significantly affected by noise. To overcome the influence of noise, an effective approach is to suppress noise in the signal phase. Principal Component Analysis (PCA) is a commonly used sequence denoising method; however, it assumes that the noise follows a zero-mean Gaussian distribution. For phase-coded modulation signal phase sequences, directly applying PCA denoising easily treats phase abrupt changes as noise signals, resulting in significant distortion of the denoised signal and severely impacting denoising and recognition performance. Summary of the Invention
[0004] The technical problem solved by this invention is to overcome the shortcomings of the prior art and propose a composite modulation signal identification method based on structured decomposition to solve the problem of robust identification of complex modulation signals in radar electronic reconnaissance.
[0005] The solution of this invention is as follows: The method of this invention is a supervised identification method, and the entire identification process is divided into two stages: training and testing. In the training stage, firstly, for the detected radar pulse complex signal... The method involves several steps: first, extracting the phase difference sequence; second, constructing a phase difference matrix using the phase difference sequence; third, performing a structured decomposition of the phase difference matrix using a low-rank sparse decomposition method; fourth, removing noise components from the decomposition result, with the remaining low-rank and sparse components serving as the basic modulation components of the composite modulation signal to achieve signal denoising; and finally, using the decomposed basic modulation components as input to a multi-layer convolutional neural network to train the network. During the testing phase, the acquisition and structured decomposition of the signal phase difference matrix are performed in the same way as during the training phase. After obtaining the structured decomposition result of the signal phase difference matrix, the low-rank and sparse components are input into the trained convolutional neural network for recognition, outputting category labels. Simulation verification and experimental results analysis show that this method is not only applicable to conventional modulation types but also significantly improves the recognition performance of composite modulation signal types, exhibiting good noise robustness.
[0006] Specifically, a method for identifying composite modulation signals based on structured decomposition is proposed, comprising the following steps:
[0007] Calculate the phase difference sequence of the signal ;
[0008] Constructing a phase difference matrix based on phase difference sequences: For the length of the window, For stepping, slide to extract the phase difference sequence And use these as columns of a matrix to construct a phase difference matrix. , This represents the number of columns in the phase difference matrix;
[0009] Structured decomposition and denoising of phase difference matrix: Construct an optimization objective function to perform structured decomposition of the phase difference matrix;
[0010] Optimize the objective function:
[0011] in, These are the low-rank components of the phase difference matrix. These are the sparse components of the phase difference matrix. The penalty factor is used; the structured decomposition of the phase difference matrix is obtained by using optimization tools: ,in, The noise component matrix is defined; noise components are removed from the decomposition, while low-rank and sparse components are retained.
[0012] Composite Modulation Signal Recognition Based on Structured Decomposition: Using the results of structured decomposition, the low-rank and sparse components of the phase difference matrix are used as inputs to a convolutional neural network to train the network. The trained network is then used to identify test samples and output category labels.
[0013] Furthermore, the calculated signal phase difference sequence Specifically:
[0014] For complex signals , Sampling time, For signal amplitude, For instantaneous phase, with Construct phase difference sequences for the interval ;
[0015] The phase difference sequence is obtained using complex number division, as shown in the following formula:
[0016]
[0017] Phase extraction operation for complex signals.
[0018] Furthermore, the phase difference matrix is constructed as follows:
[0019]
[0020] in, , , For the sequence window length, The sequence step interval.
[0021] Furthermore, the RPCA method is used to perform low-rank sparse decomposition on the phase difference matrix, which yields low-rank components, sparse components, and noise components of the phase difference matrix, corresponding to the three parts of phase continuous modulation, phase coding abrupt change, and phase noise, respectively.
[0022] Furthermore, the composite modulation signal identification based on structured decomposition is specifically as follows:
[0023] During the training phase, a multi-layer convolutional neural network is constructed, with the input layer being... It employs the ReLU activation function and selects max pooling, with the output layer being a fully connected layer;
[0024] Network training: Using the results of structured decomposition, the low-rank and sparse components of the phase difference matrix are used as inputs to train the network.
[0025] Test sample identification: The phase difference matrix of the test sample is obtained and structured decomposed. The low-rank and sparse components of the decomposition result are input into the trained neural network to identify the signal modulation pattern and output the category label.
[0026] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the composite modulation signal recognition method based on structured decomposition.
[0027] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the composite modulation signal recognition method based on structured decomposition.
[0028] The advantages of this invention compared to the prior art are:
[0029] (1) The present invention uses complex division to obtain the phase difference sequence, which avoids the phase ambiguity problem caused by the carrier frequency signal. At the same time, the phase difference can sharpen the phase change and improve the recognition of the modulated signal.
[0030] (2) The present invention constructs a phase difference matrix as the recognition feature domain, which has the characteristics of low feature extraction complexity;
[0031] (3) This invention obtains the basic modulation components of the signal through structured decomposition, which has the characteristic of clear physical meaning;
[0032] (4) The present invention removes the noise component in the decomposition result and removes phase noise, which can effectively suppress the impact of phase noise on recognition performance;
[0033] (5) The principle of this invention is simple, easy to implement in engineering, and robust. It can adapt to not only conventional signal types, but also better adapt to composite modulation signal types, and has universality. Attached Figure Description
[0034] Figure 1 This is a schematic diagram of the structured decomposition of the phase difference matrix;
[0035] Figure 2 This is a schematic diagram of the CNN convolutional neural network structure;
[0036] Figure 3 This is a flowchart of a composite modulation signal recognition method based on structured decomposition;
[0037] Figure 4 The results of modulation scheme identification under different signal-to-noise ratios;
[0038] Figure 5 Analysis of the performance improvement effect of noise reduction. Detailed Implementation
[0039] To ensure detection power, radar signals typically use maximum transmission power and modulate only the phase, without amplitude modulation. Therefore, the phase information of a radar signal contains all the modulation information of the radar modulation signal, which is the basis for this invention's ability to identify radar signal modulation. In view of this, this invention utilizes phase characteristics to identify radar modulation methods.
[0040] like Figure 3 As shown, this invention proposes a method for recognizing composite modulation signals based on structured decomposition, comprising the following steps:
[0041] (1) Calculate the phase difference sequence of the signal: For complex signals , Sampling time, For signal amplitude, For instantaneous phase, with Construct a phase difference vector for the interval ,in Phase extraction operation for complex signals;
[0042] (2) Constructing a phase difference matrix based on phase difference sequences: For the length of the window, For stepping, slide to extract the difference vector. And use these as columns of a matrix to construct a phase difference matrix. ;
[0043] (3) Structured decomposition and denoising of phase difference matrix: Construct an optimization objective function and perform structured decomposition of phase difference matrix;
[0044] Optimize the objective function:
[0045] in, These are the low-rank components of the phase difference matrix. These are the sparse components of the phase difference matrix. The penalty factor is used; the structured decomposition of the phase difference matrix is obtained by using optimization tools: ,in, The noise component matrix is defined; noise components are removed from the decomposition, while low-rank and sparse components are retained.
[0046] (4) Composite modulation signal recognition based on structured decomposition: Using the structured decomposition results, the low-rank component and sparse component of the phase difference matrix are used as inputs to the convolutional neural network to train the convolutional neural network, and the trained network is used to identify test samples and output category labels.
[0047] In step (1), the phase difference sequence is obtained by using complex division and phase taking operations to obtain the complex signal difference signal, which can effectively remove the influence of the signal carrier frequency and avoid signal phase ambiguity. The formula is as follows:
[0048]
[0049] Phase extraction operation for complex signals.
[0050] In step (2), a sliding window is used to construct the phase difference matrix:
[0051]
[0052] in, , , For the sequence window length, For sequence step interval, is the number of columns in the phase difference matrix.
[0053] In step (3), the phase difference matrix is decomposed into low-rank sparse components using the RPCA (Robust Principal Component Analysis) method. The decomposition yields low-rank components, sparse components, and noise components of the phase difference matrix. The low-rank components represent the gradual phase variation components of the phase difference matrix, corresponding to the phase gradual modulation information. The sparse components represent the phase abrupt change components, corresponding to the phase coding modulation information. The noise components are the random fluctuations of the phase difference matrix.
[0054] In step (4), during the training phase, a multi-layer convolutional neural network (CNN) is constructed, with the input layer being... The ReLU function is used as the activation function, max pooling is used, and the output layer is a fully connected layer; Network training: The low-rank component and sparse component of the phase difference matrix are used as the input of the network to train the network using the structured decomposition results; Test sample identification: The phase difference matrix is obtained and structured decomposed according to the above steps (1)-(3). The low-rank component and sparse component of the decomposition result are input into the trained neural network to identify the signal modulation pattern and output the category label.
[0055] The present invention will be further described below with reference to the embodiments.
[0056] Example 1
[0057] In this embodiment, a method for identifying composite modulation signals based on structured decomposition includes the following steps:
[0058] 1) Extracting the phase difference sequence: For complex signals , Sampling time, For signal amplitude, For instantaneous phase, with Construct a phase difference vector for the interval ,in This involves taking the phase of a complex signal. The phase difference method can sharpen abrupt changes in the phase of a radar signal.
[0059] 2) A sliding window method is used to construct the phase difference matrix for the phase difference sequence. For the length of the window, For stepping, slide to extract the difference vector. And use these as columns of a matrix to construct a phase difference matrix. .
[0060] 3) As attached Figure 1 As shown, the phase difference matrix is decomposed and denoised in a structured manner: an optimization objective function is constructed, and the phase difference matrix is decomposed in a structured manner.
[0061]
[0062] in, These are the low-rank components of the phase difference matrix. These are the sparse components of the phase difference matrix. The penalty factor is used. The structured decomposition of the phase difference matrix is obtained by solving using optimization tools: .in, This is the noise component matrix. Noise components are removed from the decomposition, while low-rank and sparse components are retained.
[0063] 4) As attached Figure 2 As shown, a multi-layer convolutional neural network (CNN) is constructed: a three-layer convolutional neural network (CNN) is built, with the input layer being... The number of convolutional kernels in the first and second layers are set to 16 and 9 respectively, and the third layer is a fully connected layer;
[0064] 5) As attached Figure 3 The process of constructing a composite modulation signal recognition system based on structured decomposition is as follows: In the training phase, the low-rank and sparse components of the phase difference matrix are used as inputs to train the network using the structured decomposition results; In the testing phase, the phase difference matrix is obtained and structured decomposed for the test samples according to steps 1)-4) above, and the low-rank and sparse components of the decomposition results are input into the trained neural network for recognition, and the recognition label is output.
[0065] 6) Simulation Verification: Recognition simulations were performed on radar signals under different signal-to-noise ratios (SNRs), including two categories: conventional radar signals and composite modulation signals. Conventional radar signals included six types: conventional pulse, binary phase-shift keying (BPSK), quadrature phase-shift keying (QPSK), frequency-coded (FSK), linear frequency modulation (LFM), and nonlinear frequency modulation (NLFM). Composite modulation signals included three types: frequency-coded composite binary phase-shift keying (FSKCompBPSK), linear frequency modulation composite binary phase-shift keying (LFMCompBPSK), and nonlinear frequency modulation composite binary phase-shift keying (NLFMCompBPSK), for a total of nine signals for simulation experiments. The SNR ranged from 0dB to 20dB, in 2dB increments. 300 signals of each type were generated at each SNR, for a total of 29,700 signals. The signal samples were divided into training and test sets. At each SNR, 50 samples of each signal type were selected as the training set, and the remaining samples were used as the test set. The experiment was repeated 100 times. The experimental results of the recognition method in this embodiment are attached. Figure 4 As shown in the figure, the proposed method has good recognition performance. To verify the impact of noise removal on recognition performance, simulations were conducted to compare the recognition performance of the denoised and non-denoised methods. The recognition results are attached. Figure 5 As shown in the figure, the proposed method significantly outperforms the method without noise removal after removing the noise component.
[0066] In summary, this invention employs a low-rank sparse decomposition method to model the signal phase, using low-rank and sparse components to model continuous phase changes and abrupt phase changes respectively. The remaining residual is modeled using Gaussian white noise, which can better model the signal phase sequence. Furthermore, the phase components after low-rank sparse decomposition can well fit the composite modulation process, and their physical meaning is clear.
[0067] This invention utilizes the RPCA decomposition method to obtain sparse components, low-rank components, and noise components of the structured decomposition. Each component of the RPCA decomposition has a clear physical meaning, enabling effective pre-classification of modulated signals. By combining this with a neural network, the results of the structured decomposition are used as input to the CNN network, discarding the noise components, which effectively reduces the impact of noise on recognition. Furthermore, using the structured decomposition as input, different components can classify different signal types, effectively imposing physical constraints on the CNN network, thereby improving the recognition ability of complex modulated signal categories and the network's generalization ability.
[0068] This application provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform... Figure 3 The method described.
[0069] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.
[0070] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0071] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0072] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0073] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.
[0074] The contents not described in detail in this specification are common knowledge to those skilled in the art.
Claims
1. A method for recognizing composite modulated signals based on structured decomposition, characterized in that, Includes the following steps: Calculate the phase difference sequence of the signal ; Constructing a phase difference matrix based on phase difference sequences: For the sequence window length, with The phase difference sequence is truncated by sliding at the sequence step interval. And use these as columns of a matrix to construct a phase difference matrix. , This represents the number of columns in the phase difference matrix; Structured decomposition and denoising of phase difference matrix: Construct an optimization objective function to perform structured decomposition of the phase difference matrix; Optimize the objective function: in, These are the low-rank components of the phase difference matrix. These are the sparse components of the phase difference matrix. The penalty factor is used; the structured decomposition of the phase difference matrix is obtained by using optimization tools: ,in, The noise component matrix is defined; noise components are removed from the decomposition, while low-rank and sparse components are retained. Composite Modulation Signal Recognition Based on Structured Decomposition: Using the structured decomposition results, the low-rank and sparse components of the phase difference matrix are used as inputs to a convolutional neural network to train the network. The trained network is then used to identify test samples and output category labels. The calculated signal phase difference sequence Specifically: For complex signals , Sampling time, For signal amplitude, For instantaneous phase, with Construct phase difference sequences for the interval ; The phase difference sequence is obtained using complex number division, as shown in the following formula: Phase extraction operation for complex signals; The phase difference matrix is constructed as follows: in, , , For the sequence window length, The sequence step interval; The identification of composite modulation signals based on structured decomposition is as follows: During the training phase, a multi-layer convolutional neural network is constructed, with the input layer being... It employs the ReLU activation function and selects max pooling, with the output layer being a fully connected layer; Network training: Using the results of structured decomposition, the low-rank and sparse components of the phase difference matrix are used as inputs to train the network. Test sample identification: The phase difference matrix of the test sample is obtained and structured decomposed. The low-rank and sparse components of the decomposition result are input into the trained neural network to identify the signal modulation pattern and output the category label.
2. The method for recognizing composite modulated signals based on structured decomposition according to claim 1, characterized in that, The phase difference matrix is decomposed into low-rank sparse components, sparse components, and noise components using the RPCA method. These components correspond to phase continuous modulation, phase coding abrupt change, and phase noise, respectively.
3. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 2.
4. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 2.