Secondary modulation signal identification method and system based on joint model deep learning

Abstract

The invention discloses a secondary modulation signal identification method and system based on joint model deep learning, and relates to electronic countermeasure, wireless communication and signal processing technologies, and the method comprises the steps: obtaining a sample secondary modulation signal, and carrying out the preprocessing; for the preprocessed sample secondary modulation signal, extracting a spatial-temporal characteristic sample based on signal reconstruction and phase space statistical analysis permutation entropy; a lightweight deep learning model is adopted to jointly construct a data local spatial characteristic perception model and a time sequence backtracking control model, and the extracted spatial-temporal characteristic samples are utilized to train the joint model; and extracting spatial-temporal characteristics of the secondary modulation signal of the target, and outputting an identification result by using the trained joint model. The method can solve the problems that a secondary modulation signal method in the field of signal and information processing is low in recognition accuracy and large in calculation amount, and the number of needed prior samples is large.

Description

Technical Field

[0001] The present application relates to the fields of electronic countermeasures, wireless communications, and signal processing technology, and in particular to a method and system for identifying secondary modulation signals based on joint model deep learning. Background Art

[0002] A secondary modulation signal is a signal generated by performing two independent modulation operations on a signal in a communication system. It aims to achieve multi-dimensional signal optimization through a layered modulation strategy. Its essence is to use the modulated signal (the output of a primary modulation) as a baseband signal and reload it onto another carrier for secondary parameter adjustment (such as amplitude, frequency, or phase) to form a secondary modulation waveform. A secondary modulation signal actually modulates the signal twice, first performing inner modulation and then outer modulation, and is widely used in various communications and unmanned aerial vehicle (UAV) measurement and control. The secondary modulation identification signal reveals its inherent modulation parameters and information encoding mechanism through the hierarchical structure of the modulated signal, demonstrating key application value in multiple fields such as military and civilian communications. The secondary modulation identification signal can be used for anti-interception communications. In tactical radios and UAV telemetry links, secondary modulation hides signal characteristics. The receiver needs to analyze the spread spectrum code and modulation parameters to achieve reliable demodulation under low signal-to-noise ratios, making it difficult for adversaries to restore the signal through conventional spectrum analysis. Secondary modulation signals also play a vital role in satellite communications. For example, satellite television broadcasting (such as DVB-S2X) employs layered secondary modulation to adapt to different terminal reception capabilities. Ground stations identify the outer QPSK parameters and dynamically switch the inner demodulation mode to achieve adaptive transmission. Secondary modulation signals can also be used for network spectrum sharing. Due to their widespread application, the identification, analysis, and detection of secondary modulation signals are of paramount importance.

[0003] Traditional signal modulation recognition methods can be divided into two categories: likelihood-based and feature-based. Likelihood-based methods typically achieve optimal recognition accuracy in the Bayesian estimation sense, but at the expense of high computational complexity. Feature-based methods primarily learn representative features from training samples and use the trained model to classify input signals. Typical feature types used by feature-based methods include instantaneous time-domain features, transform-domain features, and statistical features. Compared to likelihood-based methods, feature-based methods typically only produce suboptimal solutions, but they offer lower computational complexity and stronger multi-modulation recognition capabilities.

[0004] ZL 202311297799.0 A method for automatic modulation identification of composite modulation signals in aerospace tracking and control links. This patent discloses a method for automatic modulation identification of composite modulation signals in aerospace tracking and control links. This method uses traditional signal processing methods (cyclic spectrum, wavelet transform, etc.) to identify secondary modulation signals, and does not adopt deep learning methods. The recognition accuracy and stability are insufficient, and prior knowledge of secondary modulation signals is required.

[0005] ZL 202310624327.5 A method for blind recognition of composite modulated signals based on compressed sensing of frequency domain data. This patent uses compressed sensing technology combined with a shallow neural network to blindly recognize secondary modulated signals, improving the recognition accuracy. However, the design of the neural network is relatively simple. It only uses the convolutional layer without the timing network structure, and cannot analyze the timing characteristics of the signal, resulting in insufficient recognition accuracy. Summary of the Invention

[0006] The embodiments of the present application provide a method and system for identifying secondary modulation signals based on joint model deep learning, which solve the problems of low recognition accuracy, large amount of calculation, and large number of prior samples required for secondary modulation signal methods in the field of signal and information processing.

[0007] The present invention provides a method for identifying secondary modulation signals based on joint model deep learning, including:

[0008] Obtaining sample secondary modulation signals and performing preprocessing;

[0009] For the secondary modulation signal of the sample after preprocessing, the spatiotemporal feature samples are extracted based on signal reconstruction and phase space statistical analysis and permutation entropy;

[0010] A lightweight deep learning model is used to jointly build a data local spatial feature perception model and a temporal backtracking control model, and the joint model is trained using the extracted spatiotemporal feature samples.

[0011] The spatiotemporal features of the target secondary modulation signal are extracted, and the trained joint model is used to output the recognition results. The extracted spatiotemporal features are respectively input into the trained data local spatial characteristic perception model and the temporal backtracking control model. After the two models complete their respective classification calculations, the secondary modulation signal classification weights of the two models are optimized based on the optimal correlation information criterion, and the modulation mode recognition results of the secondary modulation signal are fused according to the classification weights to output the recognition results.

[0012] An embodiment of the present application provides a secondary modulation signal recognition system based on joint model deep learning, including a processor and a memory, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the secondary modulation signal recognition method based on joint model deep learning as described above are implemented.

[0013] The embodiment of the present application adopts the idea of lightweight processing, optimizes the amount of calculation, has a high classification and recognition accuracy in complex and changeable electromagnetic environments, and can optimize signal classification calculations. At the same time, this method requires less prior knowledge, has strong robustness, and has lower training overhead than other methods with the same recognition accuracy level, and has strong engineering practicality.

[0014] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present application. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0016] Figure 1 This is a schematic diagram of the basic process of the secondary modulation signal recognition method based on joint model deep learning in an embodiment of the present application;

[0017] Figure 2 This is a comparison of the recognition accuracy performance of the method in the embodiment of the present application and other methods. DETAILED DESCRIPTION

[0018] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0019] The technical problem to be solved by this application is that the recognition rate of traditional identification methods for secondary modulation signals in complex electromagnetic environments is low, and there are many steps in feature extraction, the classification calculation complexity is high, and more prior information about signal parameters such as signal code rate, modulation index, etc. is required. In addition, some traditional signal recognition methods have too large a computational load and long delays, and the high computational complexity hinders their deployment on devices with limited hardware resources. The embodiment of this application provides a secondary modulation signal recognition method based on joint model deep learning, such as Figure 1 As shown, the following steps are included:

[0020] In step S101, a sample secondary modulated signal is acquired and preprocessed. In some embodiments, a fast Fourier transform is used to scan the frequency band to locate the main carrier and sidelobe characteristics of the secondary modulated signal; furthermore, the signal power spectral density is calculated, and a threshold trigger is used to lock the valid signal window. The secondary modulated signal is then effectively captured using an analog-to-digital converter (ADC), and the front-end amplifier gain is dynamically adjusted to avoid saturation of the ADC or deterioration of quantization noise. Through the above process, data input for the secondary modulated signal is acquired, providing an initial data source for subsequent steps.

[0021] After the signal is collected and input, the signal is decomposed into multiple levels based on the dynamic envelope multi-level decomposition method to obtain the corresponding envelope, and then reorganized to achieve signal data preprocessing.

[0022] In step S102, the pre-processed sample secondary modulation signal is subjected to signal reconstruction and phase space statistical analysis permutation entropy to extract spatiotemporal feature samples. In a specific example, after completing the pre-processing of the secondary modulation signal, the obtained signal normalized sequence is decomposed, and the decomposed sequence is then mapped from a low-dimensional to a high-dimensional phase space to achieve phase space reconstruction of the signal. An adaptive scaling factor and a noise suppression weight coefficient are introduced to optimize the phase space statistical analysis of the reconstructed signal to achieve nonlinear feature extraction of the signal.

[0023] In step S103, a lightweight deep learning model is used to jointly construct a data local spatial characteristic perception model and a temporal backtracking control model, and the joint model is trained using the extracted spatiotemporal feature samples. In some embodiments, the jointly constructed data local spatial characteristic perception model and temporal backtracking control model include: a data local spatial characteristic perception module, a temporal backtracking control module, and a joint classification calculation module. The joint model is used to process the extracted temporal and spatial features of the secondary modulation signal using the data local spatial characteristic perception model and the temporal backtracking control model, respectively, and to fuse the classification results of the two models according to the joint classification calculation module.

[0024] In step S104, the spatiotemporal features of the target secondary modulation signal are extracted, and the trained joint model is used to output the recognition result, wherein the extracted spatiotemporal features are respectively input into the trained data local spatial characteristic perception model and the timing backtracking control model. After the two models complete their respective classification calculations, the secondary modulation signal classification weights of the two models are optimized based on the optimal correlation information criterion, and the modulation mode recognition results of the secondary modulation signal are fused according to the classification weights to output the recognition result.

[0025] The embodiment of the present application adopts the idea of lightweight processing, optimizes the amount of calculation, has a high classification and recognition accuracy in complex and changeable electromagnetic environments, and can optimize signal classification calculations. At the same time, this method requires less prior knowledge, has strong robustness, and has lower training overhead than other methods with the same recognition accuracy level, and has strong engineering practicality.

[0026] In some embodiments, obtaining the sample secondary modulation signal and performing preprocessing further includes:

[0027] The acquired secondary modulation signal is divided into sampling points, saved and output in vector format, and the vector is normalized.

[0028] After that, all extreme points of the analytical signal are calculated and the extreme envelope v is fitted by the cubic spline function + (t).

[0029] Then get all the extreme values of the function and fit their effect envelope v - (t), and calculate the mean m(t) of the two envelopes.

[0030] Subtract the calculated mean from the normalized secondary modulated signal to obtain a new signal. Repeat the above decomposition process until a signal that meets the conditions is obtained. The output result can be expressed as:

[0031] r1(t)=x(t)-k1 n (t) = x(t) - imf1(t)

[0032] Where r1(t) represents the decomposed secondary modulation signal, x(t) represents the input sampling data, and imf1(t) represents the first-order intrinsic mode function of the signal. Based on the decomposition of the signal data, the envelope of the signal can be preliminarily analyzed and processed to achieve signal denoising.

[0033] Repeat the decomposition process for the new signal generated by each iteration, perform decomposition on the new signal obtained each time, obtain the second-order component, third-order component, ..., n-order component, and realize the acquisition of the noise component of the secondary modulation signal, where the residual signal r nWhen (t) becomes a monotonic function and cannot be further decomposed into an oscillating component or becomes a constant, the decomposition is terminated. The multi-stage filtering process of the signal is performed through the above process, and the signal is represented as follows:

[0034]

[0035] Among them, IMF n (t) is the n-th order envelope component.

[0036] The sum of the third-order component and the residual term is selected as the signal preprocessing result.

[0037] X input (t) = imf last (t)+imf last-1 (t)+imf last-2 (t)+r n (t)

[0038] Among them, X input (t) is the result of output signal preprocessing, imf1(t) is the envelope component of each order r n (t) is the nth-order residual term. By filtering the multi-level components of the secondary modulated signal, the influence of noise components on the signal is eliminated, achieving efficient preprocessing of the input data. The processed results are input into the next step, where feature extraction is performed based on signal reconstruction and phase space statistical analysis.

[0039] After completing the preprocessing of the secondary modulated signal, the obtained signal normalized sequence is decomposed, and then the decomposed sequence is mapped from a low-dimensional to a high-dimensional phase space to achieve phase space reconstruction of the signal. An adaptive scaling factor and a noise suppression weight coefficient are introduced to optimize the phase space statistical analysis of the reconstructed signal to achieve nonlinear feature extraction of the signal. In some embodiments, extracting spatiotemporal feature samples based on signal reconstruction and phase space statistical analysis permutation entropy for the sample secondary modulated signal after preprocessing includes:

[0040] The phase parameters of the pre-processed sample secondary modulation signal are transformed to construct the phase space, and the input y row parameter is inversely transformed to meet the following requirements:

[0041]

[0042] Among them, x is the preprocessed signal data, is the phase parameter of the signal, is the output after phase transformation, R and I are the real and imaginary part operation functions of the data.

[0043] According to the structure of phase parameter transformation, the time delay embedding method is used to reconstruct the phase space from the time series data, so that data feature analysis can be performed in high-dimensional space, as shown in the following formula:

[0044]

[0045] in, is the output reconstruction result, is the signal data after phase transformation; N is the length of the preprocessed signal, s is the scale of data reconstruction, and the reconstructed signal vector is obtained according to the scale size

[0046] After obtaining the reconstructed signal, the signal sequence in the phase space is quantized. The quantization interval is set according to the reconstructed signal vector, and the reconstructed signals within the same interval are assigned the same discrete value to obtain the sequence symbol. The number of occurrences of each different sequence symbol is counted. The phase space statistical feature permutation entropy of the reconstructed signal is calculated as shown in the following formula to extract the nonlinear characteristics of the signal:

[0047]

[0048] Among them, p l =N l / N represents the proportion of the number of the lth symbol in the total number of symbols, L is the total number of symbols, and m=N / s is the length of the reconstructed signal sequence.

[0049] The phase space statistical feature vector extracted from the reconstructed signal is used as the extracted spatiotemporal feature sample and output to the subsequent data local spatial characteristic perception model and timing backtracking control model for secondary modulation signal feature analysis and classification recognition.

[0050] After extracting the secondary modulation signal features, a lightweight deep learning model is used to jointly construct a data local spatial feature perception model and a time sequence backtracking control model, and the model is trained for subsequent classification and recognition of the obtained signal features. In some embodiments, training the joint model using the extracted spatiotemporal feature samples includes:

[0051] First, the signal spatial features are compressed by the local perception calculation unit of the data local spatial feature perception model;

[0052] The local perception calculation unit intercepts the local vector of the signal feature and multiplies it with the perception calculation kernel, and performs sliding operations to realize the successive analysis of each local feature of the signal, obtaining the spatial feature vector after scale compression, and calculating the category probability vector through the exponential normalization function, thereby improving the classification and recognition accuracy of the secondary modulation signal and reducing the computational complexity.

[0053] The spatial feature vectors obtained by the data local spatial feature perception model are rescaled and reconstructed to adapt the parameter input requirements of the time series backtracking control model;

[0054] After that, the vector obtained by the data local spatial feature perception model is rescaled and reconstructed to adapt to the parameter input requirements of the time series backtracking control model, and then the rescaled vector is input into the time series backtracking control model. The model uses the state selective update unit to save the feature data involved in the operation and uses the update parameter z in each training calculation output process. t Control history status data h t-1 and status data to be updated The mixing ratio satisfies:

[0055]

[0056] in, is the new state of the output model, h t-1 is the historical state data, W h is the scaling function, x t is the input data of the model, that is, the feature vector obtained by the local spatial characteristic perception model of the data, b h The bias term, tanh, and hyperbolic tangent function are used to activate the model's calculation results. Exponential vector normalization is then used to convert the results into a probability vector, i.e., the modulation mode identification vector for the quadratic modulation signal. By selectively updating historical and current data, the vanishing gradient problem caused by the model processing long time series data can be mitigated, improving the accuracy of quadratic modulation signal feature calculation.

[0057] In some embodiments, training the joint model using the extracted spatiotemporal feature samples also includes: after calculating the data local spatial characteristic perception model and the temporal backtracking control model, fusing the outputs of the two models according to the joint classification calculation module, calculating the loss of the negative expected logarithmic probability optimization loss function of the model output and the sample data, that is, calculating the loss of the negative expected logarithmic probability optimization loss function of the model output and the sample data, and using the loss as the gradient of the iterative parameter update to train the model, as shown in the following formula.

[0058]

[0059] Among them, W h , b h is the linear calculation parameter and offset parameter of the model, y t (i) is the probability that the sample value at time t is i, that is, 0 or 1, is the probability that the signal recognition classification result is i at time t, that is, the output result of the joint model of data local spatial feature perception model and temporal backtracking control, [W h ,b h ] is the joint vector matrix composed of the corresponding parameters of the model, η is the update step size, which can be set to a fixed value.

[0060] By calculating the loss function, the model parameters are iteratively updated step by step until the model converges. Afterwards, signal classification and recognition are achieved based on a joint model of local spatial feature perception and temporal backtracking control.

[0061] After completing the construction and training analysis of the joint model in the above steps, based on the optimal correlation information criterion, the data local spatial characteristic perception model and the temporal backtracking control model are combined to achieve refined mining of data characteristics and improve the accuracy of weight distribution.

[0062] First, the obtained secondary modulation signal feature data is input into the trained data local spatial characteristic perception model and the corresponding modulation mode recognition output result. At the same time, the feature vector is input into the timing backtracking control model, and the output type recognition probability vector is calculated and normalized. After completing the classification calculation for each of the two models, the secondary modulation signal classification weights of the two models are optimized based on the optimal correlation information criterion to reduce the classification error and meet the following requirements:

[0063]

[0064] Among them, w i is the model classification weight, I is the association information function, p i The classification probability is calculated by the data local spatial feature perception model and the temporal retrospective control model.

[0065] In some embodiments, the modulation mode recognition results of the secondary modulated signal are fused according to the classification weight to output a recognition result that satisfies:

[0066] C=w1c1+w2c2

[0067] Where C is the joint classification vector, w1 and w2 are the weights of the data local spatial characteristic perception model and the temporal backtracking control model, respectively, and c1 and c2 are the classification result vectors of the data local spatial characteristic perception model and the temporal backtracking control model, respectively. After calculating the joint classification vector, the vector is normalized and the modulation mode recognition probability is calculated and sorted. The category corresponding to the highest probability is selected as the final classification result of the secondary modulation signal modulation mode.

[0068] The method of this application fully considers the characteristics of secondary modulation signals and uses a parameter transformation module to reduce the adverse effects of the channel on signal data. In terms of recognition accuracy, the method of this application significantly outperforms existing deep learning recognition methods based on CNN network models. The method of this application uses fewer parameters than other existing deep learning-based modulation recognition methods.

[0069] The method proposed in this application can directly use raw modulation recognition data as input, without consuming additional resources for data preprocessing. This method comprehensively considers recognition accuracy and model complexity, achieving a recognition rate close to the highest recognition accuracy of current deep learning modulation recognition methods while using a minimal number of model parameters.

[0070] The present application also proposes an implementation case of a secondary modulation signal recognition method based on joint model deep learning:

[0071] In this example, the secondary modulated communication signal data input is first collected, and preprocessing is performed based on the dynamic envelope multi-level decomposition of the input signal to obtain a denoised secondary modulated signal. The signal is then reconstructed in the phase space based on the phase statistical transformation of the signal data, and the signal phase space statistical characteristics are calculated to achieve feature extraction. The obtained feature vector is then input into the joint model of data local spatial characteristic perception and timing backtracking control. The joint model is trained and analyzed based on the sample data to process the secondary modulated signal characteristics and perform classification and recognition, ultimately achieving the output of the recognition result of the secondary modulated signal.

[0072] The specific implementation process of the efficient recognition method of secondary modulation signals based on joint model deep learning:

[0073] Initially, all models were evaluated on simulated, secondary modulation signals, including 2FSK-PM, BPSK-FM, BPSK-PM, MSK-FM, OQPSK-FM, OQPSK-PM, QPSK-FM, QPSK-PM, PCM-FM, PCM-PM, as well as MSK-DSB and PCM-DSB modulation. The dataset was randomly split into training, validation, and test sets in a 6:2:2 ratio. The model was implemented using a GeForce GTX 1080Ti GPU and Keras with TensorFlow as the backend. No expert feature extraction or other preprocessing was performed on the raw signal. Instead, raw time series features were learned directly from the high-dimensional data using a real network, with the real and imaginary parts of the input signal being fed into the model as two dimensions.

[0074] Representative State-of-the-art (SOTA) models are used to provide baseline comparisons, including the IC-AMCNET model and MCNET model based on CNN networks, the LSTM2 model and GRU2 model based on RNN networks, and the MCLDNN model with the highest recognition accuracy.

[0075] In Table 1, multiple indicators were selected for performance comparison and complexity analysis, including parameter count, training time, test time, the highest recognition accuracy under all signal-to-noise ratios (SNRs), and the average recognition accuracy under all SNRs. The method of the present application has the fewest parameters and the shortest training time. The test time is only 0.003 seconds / sample worse than the fastest IC-AMCNET model. The accuracy performance is slightly lower than the MCLDNN model, but comparable to the LSTM2 model. Overall, compared with other methods, the model of the present method has the fewest parameters. Even compared with the MCNET model with the fewest parameters, the number of parameters is less than 70%. Compared with GRU2, LSTM2, and MCLDNN, which have superior performance but high complexity, the time cost of the method of the present application has a significant advantage. Compared with the IC-ACMNET and MCNET models based on the CNN architecture, the time cost of this method is comparable, but the accuracy is higher.

[0076] Table 1 Performance comparison between the present invention and the existing method

[0077]

[0078] Based on the above initial conditions, simulation and experiments are carried out. Figure 2 The paper provides a graph showing how the recognition accuracy of each model changes with the signal-to-noise ratio. The MCLDNN model has the highest recognition accuracy, but the method proposed in this patent, with less than one-fifth of the parameters of the MCLDNN model, achieves a recognition accuracy very close to that of the MCLDNN model. Furthermore, the recognition accuracy of the method proposed in this patent is superior to that of the GRU2 model. The IC-AMCNET and MCNET models based on CNN networks, in data simulation, have lower recognition accuracy than the method proposed in this patent, achieving only a maximum recognition accuracy of around 80%.

[0079] From the perspective of comprehensive performance, the method of this application takes into account both complexity and accuracy, and is an efficient method for identifying secondary modulation signals based on deep learning of joint models.

[0080] An embodiment of the present application also provides a secondary modulation signal recognition system based on joint model deep learning, including a processor and a memory, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the secondary modulation signal recognition method based on joint model deep learning as described above are implemented.

[0081] It should be noted that, in the various embodiments of the present application, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0082] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0083] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0084] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are protected by this application.

Claims

1. A method for identifying secondary modulation signals based on joint model deep learning, characterized in that: include: Obtaining sample secondary modulation signals and performing preprocessing; For the secondary modulation signal of the sample after preprocessing, the spatiotemporal feature samples are extracted based on signal reconstruction and phase space statistical analysis and permutation entropy; A lightweight deep learning model is used to jointly build a data local spatial feature perception model and a temporal backtracking control model, and the joint model is trained using the extracted spatiotemporal feature samples. The spatiotemporal features of the target secondary modulation signal are extracted, and the trained joint model is used to output the recognition results. The extracted spatiotemporal features are respectively input into the trained data local spatial characteristic perception model and the timing backtracking control model. After the two models complete their respective classification calculations, the secondary modulation signal classification weights of the two models are optimized based on the optimal correlation information criterion, and the modulation mode recognition results of the secondary modulation signal are fused according to the classification weights to output the recognition results.

2. The method for identifying secondary modulation signals based on joint model deep learning according to claim 1, wherein: Obtaining a sample secondary modulation signal and performing preprocessing includes: Scanning the frequency band by fast Fourier transform to locate the main carrier and sidelobe characteristics of the secondary modulation signal; and Calculate the signal power spectrum density and lock the valid signal window according to the threshold trigger; The secondary modulation signal is obtained based on the analog-to-digital converter ADC.

3. The method for identifying secondary modulation signals based on joint model deep learning according to claim 2, wherein: Acquiring a sample secondary modulation signal and performing preprocessing also includes: The acquired secondary modulation signal is divided into sampling points and normalized; Calculate the extreme points of the normalized secondary modulation signal and fit the extreme envelope to obtain all extreme values; Fit the effect envelopes for the extreme values and calculate the means of the two envelopes; Subtract the calculated mean from the normalized secondary modulation signal to obtain a new signal; Decomposition is performed on each new signal obtained to obtain second-order components, third-order components, ..., n-order components, thereby obtaining the noise component of the secondary modulation signal. The decomposition is terminated when the residual signal becomes a monotonic function and cannot be further decomposed into an oscillating component or becomes a constant. The sum of the third-order component and the residual term is selected as the signal preprocessing result.

4. The method for identifying secondary modulation signals based on joint model deep learning according to claim 1, wherein: For the secondary modulation signal of the sample after preprocessing, the temporal and spatial feature samples are extracted based on signal reconstruction and phase space statistical analysis and substitution entropy, including: The phase parameters of the pre-processed sample secondary modulation signal are transformed to construct the phase space to meet the following requirements: Among them, x is the preprocessed signal data, is the phase parameter of the signal, is the output after phase transformation, R and I are the real and imaginary part operation functions of the data; According to the structure of phase parameter transformation, the phase space is reconstructed from time series data using the time delay embedding method: in, is the output reconstruction result, is the signal data after phase transformation; N is the length of the preprocessed signal, s is the scale of data reconstruction, j represents the output reconstruction result data number, i represents the signal data number of phase transformation, and the reconstructed signal vector is obtained according to the scale size. The quantization interval is set according to the reconstructed signal vector, and the reconstructed signal in the same interval is assigned the same discrete value to obtain the sequence symbol, and the number of occurrences of each different sequence symbol is counted; The phase space statistical feature vector extracted from the reconstructed signal is used as the extracted spatiotemporal feature sample.

5. The method for identifying secondary modulation signals based on joint model deep learning according to claim 4, wherein: The jointly constructed data local spatial characteristic perception model and timing backtracking control model include: a data local spatial characteristic perception module, a timing backtracking control module and a joint classification calculation module. The joint model is used to process the time characteristics and spatial characteristics of the extracted secondary modulation signal separately using the data local spatial characteristic perception model and the timing backtracking control model, and fuse the classification results of the two models according to the joint classification calculation module.

6. The method for identifying secondary modulation signals based on joint model deep learning according to claim 5, wherein: Training the joint model using the extracted spatiotemporal feature samples includes: The signal spatial features are compressed by the local perception calculation unit of the data local spatial feature perception model; The local perception calculation unit intercepts the local vector of the signal feature and multiplies it with the perception calculation kernel, and performs sliding operations to realize the successive analysis of each local feature of the signal to obtain the spatial feature vector after scale compression; The spatial feature vectors obtained by the data local spatial feature perception model are rescaled and reconstructed to adapt the parameter input requirements of the time series backtracking control model; The transformed vector is input into the time series backtracking control model, which uses the state selective update unit to save the feature data involved in the operation and uses the update parameter z in each training calculation output process. t Control history status data h t-1 and status data to be updated The mixing ratio satisfies: in, is the new state of the output model, h t-1 is the historical state data, x t Input data for the model, b h is the bias term, W h is the scaling function, tanh is the hyperbolic tangent function, and t represents the time that the state is in.

7. The method for identifying secondary modulation signals based on joint model deep learning according to claim 6, wherein: Training the joint model using the extracted spatiotemporal feature samples also includes: After calculating the data local spatial feature perception model and the temporal backtracking control model, the outputs of the two models are fused according to the joint classification calculation module, and the loss function loss of the negative expected logarithmic probability optimization of the model output and the sample data is calculated to train the model.

8. The method for identifying secondary modulation signals based on joint model deep learning according to claim 7, wherein: The classification weights of the secondary modulation signals of the two models are optimized based on the optimal correlation information criterion to meet the following requirements: Among them, w i is the model classification weight, I is the association information function, p i and p j They represent the classification probabilities calculated by the data local spatial feature perception model and the temporal backtracking control model, respectively, where i and j represent any two different data sequence numbers in the model, and i≠j.

9. The method for identifying secondary modulation signals based on joint model deep learning according to claim 8, wherein: The modulation mode recognition results of the secondary modulation signal are fused according to the classification weight to output the recognition result that satisfies: C=w1c1+w2c2 Among them, C is the joint classification vector, w1 and w2 are the weight values of the data local spatial feature perception model and the temporal backtracking control model, respectively, and c1 and c2 are the classification result vectors of the data local spatial feature perception model and the temporal backtracking control model, respectively.

10. A secondary modulation signal recognition system based on joint model deep learning, characterized in that: It includes a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the secondary modulation signal recognition method based on joint model deep learning as described in any one of claims 1 to 9 are implemented.

Images