A PCMA signal recognition method and device

Abstract

This application discloses a PCMA signal recognition method and device applicable to the field of satellite communications technology. The method comprises: obtaining a paired carrier multiple access (PCMA) signal to be identified; inputting the waveform and spectrum of the PCMA signal into a residual neural network (ResNet) to extract the spatial features of the waveform and spectrum; fusing the spatial features of the waveform and spectrum to obtain the spatiotemporal correlation between the waveform and spectrum; and identifying and classifying the PCMA signal using a fully connected layer based on the spatiotemporal correlation. This method utilizes time-frequency graphs and waveform graphs as input data to suppress interference from other information in the signal, perform preliminary extraction of target features, facilitate network model training, and improve the accuracy of signal detection and recognition when the number of symbols is small.

Description

Technical Field

[0001] The present application relates to the field of satellite communication technology, and in particular to a method and device for identifying PCMA signals. Background Art

[0002] Paired Carrier Multiple Access (PCMA) is an emerging satellite communication system that allows both communicating parties to transmit information simultaneously within the same frequency band. Communication is achieved by transmitting two digitally modulated signals at the same frequency. The components of these two PCMA signals overlap in the time domain and in the frequency domain. When receiving a single-channel PCMA signal, it is inevitable that other, non-target, single-channel digitally modulated signals from satellites will be received. Therefore, it is necessary to distinguish the PCMA signal from common single-channel satellite digitally modulated signals and then identify the specific modulation method of the PCMA signal.

[0003] Currently, existing detection and identification methods are divided into two categories: hypothesis-testing-based methods and feature-based methods. These methods share the common characteristic of extracting signal features by calculating high-order cumulants and simultaneously designing decision rules to achieve signal modulation recognition. However, because the estimation of high-order statistics requires a large number of symbols to achieve high accuracy, these methods perform poorly when the number of symbols is small. Therefore, designing a method to improve the accuracy of signal detection and identification when the number of symbols is small has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] In view of this, the embodiments of the present application provide a method and apparatus for identifying PCMA signals, aiming to meet the requirements for improving the accuracy of signal detection and identification.

[0005] In a first aspect, an embodiment of the present application provides a method for identifying a PCMA signal, the method comprising:

[0006] Acquire a paired carrier multiple access (PCMA) signal to be identified;

[0007] Inputting the waveform and spectrum of the PCMA signal into the residual neural network ResNet respectively to extract the spatial features of the waveform and spectrum;

[0008] fusing the spatial features of the waveform and spectrum and obtaining the spatiotemporal correlation between the waveform and spectrum;

[0009] The PCMA signal is identified and classified according to the spatiotemporal correlation through the fully connected layer.

[0010] Optionally, fusing the spatial features of the waveform and the spectrum and obtaining the spatiotemporal correlation between the waveform and the spectrum includes:

[0011] Extracting the temporal features of the PCMA signal through a long short-term memory (LSTM) layer;

[0012] The spatiotemporal correlation between the waveform and the spectrum is obtained according to the temporal feature and the spatial feature.

[0013] Optionally, before respectively inputting the waveform and spectrum of the PCMA signal into the residual neural network ResNet, the method further includes:

[0014] The waveform and spectrum of the PCMA signal are respectively input into the convolution layer to reduce the noise variance and abstract the waveform and spectrum of the PCMA signal.

[0015] Optionally, after fusing the spatial features of the waveform and the spectrum and obtaining the spatiotemporal correlation between the waveform and the spectrum, the following steps are further included:

[0016] The signal is mapped through a fully connected FC layer and fitted using a preset algorithm, wherein the preset algorithm is a dropout algorithm.

[0017] Optionally, identifying and classifying PCMA signals includes:

[0018] The PCMA signal is identified and classified by an activation function with 6 units, wherein the function is a Softmax function.

[0019] In a second aspect, an embodiment of the present application provides a device for identifying a PCMA signal, the device comprising:

[0020] A signal acquisition module, used to acquire a paired carrier multiple access (PCMA) signal to be identified;

[0021] A spatial feature acquisition module is used to input the waveform and spectrum of the PCMA signal into the residual neural network ResNet respectively to extract the spatial features of the waveform and spectrum;

[0022] A spatiotemporal correlation acquisition module is used to fuse the spatial features of the waveform and spectrum and obtain the spatiotemporal correlation of the waveform and spectrum;

[0023] The recognition and classification module is used to recognize and classify the PCMA signal according to the spatiotemporal correlation through a fully connected layer.

[0024] Optionally, fusing the spatial features of the waveform and the spectrum and obtaining the spatiotemporal correlation between the waveform and the spectrum includes:

[0025] Extracting the temporal features of the PCMA signal through a long short-term memory (LSTM) layer;

[0026] The spatiotemporal correlation between the waveform and the spectrum is obtained according to the temporal feature and the spatial feature.

[0027] Optionally, the spatial feature acquisition module further includes:

[0028] The signal processing module is used to input the waveform and spectrum of the PCMA signal into the convolution layer respectively, reduce the noise variance and abstract the waveform and spectrum of the PCMA signal.

[0029] Optionally, after fusing the spatial features of the waveform and the spectrum and obtaining the spatiotemporal correlation between the waveform and the spectrum, the following steps are further included:

[0030] The signal is mapped through a fully connected FC layer and fitted using a preset algorithm, wherein the preset algorithm is a dropout algorithm.

[0031] Optionally, identifying and classifying PCMA signals includes:

[0032] The PCMA signal is identified and classified by an activation function with 6 units, wherein the function is a Softmax function.

[0033] The present embodiment provides a method for identifying PCMA signals, comprising: obtaining a paired carrier multiple access (PCMA) signal to be identified; inputting the waveform and spectrum of the PCMA signal into a residual neural network (ResNet) to extract the spatial features of the waveform and spectrum; fusing the spatial features of the waveform and spectrum to obtain the spatiotemporal correlation between the waveform and spectrum; and identifying and classifying the PCMA signal based on the spatiotemporal correlation using a fully connected layer. This method utilizes time-frequency graphs and waveform graphs as input data to suppress interference from other information in the signal, perform preliminary extraction of target features, facilitate network model training, and improve the accuracy of signal detection and identification when the number of symbols is small.

[0034] In addition, the present application also provides a PCMA signal recognition device, the technical effect of which corresponds to the above method and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the following briefly introduces the drawings required for use in the embodiment or the prior art description. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1A flowchart of a method for identifying a PCMA signal provided in an embodiment of the present application;

[0037] Figure 2 Another method flow chart of the PCMA signal recognition method provided in an embodiment of the present application;

[0038] Figure 3 A flowchart of another method for identifying a PCMA signal provided in an embodiment of the present application;

[0039] Figure 4 A schematic diagram of a scenario of a method for identifying PCMA signals provided in an embodiment of the present application;

[0040] Figure 5 A schematic diagram comparing the PCMA signal detection rate with the signal-to-noise ratio using the PCMA signal recognition method provided in the embodiment of the present application and three other algorithms;

[0041] Figure 6 A schematic diagram showing how the recognition rate of different PCMA signals varies with the signal-to-noise ratio according to the PCMA signal recognition method provided in an embodiment of the present application;

[0042] Figure 7 Schematic diagram of how the recognition rate of different algorithms changes with the signal-to-noise ratio;

[0043] Figure 8 A schematic diagram of the structure of a PCMA signal recognition device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below in conjunction with the accompanying drawings and specific embodiments. Obviously, the embodiments described are only a part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present application.

[0045] Paired Carrier Multiple Access (PCMA) is an emerging satellite communication system proposed by Mark Dankberg in 1998. It allows both communicating parties to send information at the same time and in the same frequency band. Therefore, this technology can effectively improve the frequency band utilization of satellite communication channels, save frequency band resources, and has high utilization value and broad development space.

[0046] In satellite communications, paired carrier multiple access (PCMA) technology transmits two digitally modulated signals at the same frequency. The components of these two PCMA signals overlap in the time domain and completely overlap in the frequency domain. Detecting the target signal and identifying its modulation method are fundamental to subsequent signal processing and a critical issue in non-cooperative communication. Signals typically consist of a mixture of PCMA signals (BPSK, QPSK, or 8PSK) and single-channel signals (BPSK, QPSK, or 8PSK). When receiving a single-channel PCMA signal, it is inevitable that other non-target satellite digitally modulated single-channel signals will be received. Therefore, to ensure proper signal processing of the PCMA signal, it is necessary to distinguish the PCMA signal from common single-channel satellite digitally modulated signals and then identify its specific modulation method. Modulation identification of the PCMA signal provides essential guidance for subsequent signal parameter estimation and blind separation.

[0047] The present invention provides a method for identifying PCMA signals, the flow chart of which is as follows: Figure 1 As shown, the following steps are included:

[0048] S10: Acquire a paired carrier multiple access (PCMA) signal to be identified.

[0049] The paired carrier multiple access PCMA signal is generally composed of a mixed signal consisting of BPSK, QPSK, 8PSK PCMA signals and BPSK, QPSK, 8PSK single-channel signals, and is received to obtain the paired carrier multiple access PCMA signal.

[0050] S20, inputting the waveform and spectrum of the PCMA signal into the residual neural network ResNet respectively to extract spatial features of the waveform and spectrum.

[0051] The waveform and spectrum of the PCMA signal are respectively input into the residual neural network ResNet. Compared with the method of directly inputting signal waveform data in the prior art, this application adopts time-frequency diagram and waveform diagram as input data form, which suppresses the interference of other information in the signal. It can also be regarded as a preliminary extraction of target features, which helps the training of the network model. Therefore, it shows better recognition performance in some methods. The method based on the signal transformation domain (time-frequency diagram and waveform diagram) can better reflect the signal feature distribution.

[0052] S30: Fusing the spatial features of the waveform and the spectrum to obtain the spatiotemporal correlation between the waveform and the spectrum.

[0053] The image features of the signal time-frequency diagram and waveform diagram are extracted by the neural network architecture in step S20, and the time-frequency characteristics existing in the signal are used for detection and recognition. The features extracted from the two channels are fused to obtain the joint features, that is, the spatiotemporal correlation of the waveform and spectrum.

[0054] S40: Identify and classify the PCMA signal according to the spatiotemporal correlation through a fully connected layer.

[0055] The features are mapped through the fully connected layer, and the PCMA signals are identified and classified according to the preset function.

[0056] It can be seen that this method uses time-frequency diagrams and waveform diagrams as input data to suppress interference from other information in the signal, performs preliminary extraction of target features, helps train the network model, and improves the accuracy of signal detection and recognition when the number of symbols is small.

[0057] In some specific embodiments, Figure 2 As shown in the flowchart of another method for identifying the PCMA signal, step S30 specifically includes S301 and S302:

[0058] S301, extracting the temporal features of the PCMA signal through a long short-term memory (LSTM) layer.

[0059] LSTM can extract temporal features, while the 2D convolutional layer can extract spatial features of PCMA signals. The temporal features of the signal are extracted through the LSTM layer so that they can be combined with the spatial features obtained by the convolutional layer to obtain spatiotemporal correlation.

[0060] S302: Obtaining a spatiotemporal correlation between a waveform and a spectrum according to the temporal feature and the spatial feature.

[0061] MCLDNN integrates CNN, LSTM, and fully connected layer deep neural networks into a unified structure to leverage their complementarity and synergy for spatiotemporal feature extraction and classification. The features extracted from the two channels are fused to obtain joint features, and the effective features obtained are then input into the classification layer for signal classification, detection, and recognition.

[0062] In some specific embodiments, Figure 2 As shown in the flowchart of another method for identifying the PCMA signal, step S10 further includes step S11:

[0063] S11 , inputting the waveform and spectrum of the PCMA signal into a convolution layer respectively, reducing the noise variance and abstracting the waveform and spectrum of the PCMA signal.

[0064] The waveform and spectrum of the PCMA signal are input into the two-dimensional convolution layer respectively to reduce the noise variance and provide better features for LSTM. The multi-channel input and processing structure captures the features of input representation at different scales and effectively utilizes the complementary information of I channel, Q channel and I / Q multi-channel data.

[0065] In some specific embodiments, Figure 3 As shown in the flowchart of another method for identifying the PCMA signal, step S30 further includes step S31:

[0066] S31, mapping the signal through a fully connected FC layer and fitting it using a preset algorithm, wherein the preset algorithm is a dropout algorithm.

[0067] Dropout significantly reduces overfitting by ignoring half of the feature detectors in each training batch (setting half of the hidden layer node values ​​to 0). This approach reduces interactions between feature detectors, where some detectors rely on others to function. Using dropout prevents model overfitting.

[0068] In some specific embodiments, identifying and classifying the PCMA signal in step S40 specifically includes:

[0069] The PCMA signal is identified and classified by an activation function with 6 units, wherein the function is a Softmax function.

[0070] The fully connected layer with 6 units and the activation function of the Softmax function classifies the signal, and there are 6 classification results corresponding to different detection and recognition results.

[0071] Scenario 1:

[0072] For ease of understanding, the PCMA signal recognition method in this application is applied to Figure 4 In the network structure scenario shown in Figure 1, Figure 4 The network structure shown is a multi-channel residual fusion long-term deep neural network (hereinafter referred to as MCRLDNN). The experimental environment and parameter settings are as follows: the source signal set is set to {BPSK, QPSK, 8PSK, BPSK mixed, QPSK mixed, 8PSK mixed}.

[0073] MCLDNN exploits the complementarity and synergy of them for spatiotemporal feature extraction and classification. The network has two functional parts: multi-channel input and spatial feature mapping, temporal feature extraction, and fully connected classification.

[0074] The multi-level spatial feature extraction component consists of three 2D convolutional layers (Conv1, Conv2, and Conv3) and two ResNet18 modules (Residual Neural Networks ResNet1 and ResNet2). First, the waveform and time-frequency graphs of the received I / Q multi-channel signal data are resized to 256×2×1 via a resizing layer to meet the required dimensions. The input data is then fed into ResNet1 and ResNet2, respectively. These are then fused in Concatenate and fed into Conv3 to extract spatial correlation. This multi-channel input and processing structure effectively utilizes the spatial feature data of the signal.

[0075] The long short-term memory layer consists of two LSTM layers (LSTM1 and LSTM 2). The LSTM layer can effectively process sequential data to extract the temporal correlation of signals.

[0076] The fully connected layer consists of two FC layers (FC1 and FC2). The FC layer maps features to a more separable space and uses dropout to prevent overfitting. The signal is then classified using a fully connected layer with 6 units and a softmax activation function. Six classification results correspond to different detection and recognition outcomes, achieving more accurate classification.

[0077] When the output is a BPSK, QPSK, or 8PSK single-channel signal, it indicates a non-PCMA signal. When the output is a BPSK, QPSK, or 8PSK PCMA modulated signal, it indicates the corresponding three PCMA mixed modulation signals. This allows detection of single-channel signals and PCMA signals, as well as identification of the modulation method of the corresponding PCMA signal.

[0078] Scenario 2:

[0079] The number of samples in each symbol period is L = 4000 (corresponding to 1000 symbols), the roll-off coefficient of the raised cosine shaping filter is α = 0.25, the noise is Gaussian white noise, the oversampling factor is 4, and the normalized frequency offset of the two signals (relative to the symbol rate) is randomly and uniformly selected within a certain range, that is, Δf_1∈(0,10^(-3)). The recognition rate, which measures the recognition performance, is the ratio of the number of correct recognitions to the total number of trials. The following comparison reflects the beneficial effects achieved by the solution in this application:

[0080] (1) PCMA signal detection based on MCRLDNN

[0081] A simulation comparison is conducted on the signal detection algorithms based on MCRLDD algorithm, CNN algorithm, constellation zero point clustering feature and amplitude flatness, and the latter two recognition algorithms are denoted as CZCF and AF respectively. Figure 5The detection rate of PCMA signals for the four algorithms varies with the signal-to-noise ratio.

[0082] Figure 5 When the amplitude ratio of the two signal components is fixed and the time delay is 0, the changes in the recognition rate of the PCMA signal by the four recognition algorithms under different signal-to-noise ratios can be seen. It can be seen that under high signal-to-noise ratio, the four algorithms have good recognition rates, but under low signal-to-noise ratio, the algorithms MCRLDD and CNN still have high recognition rates. Since the time delay of the two signal components is 0 at this time, the CZCF algorithm can accurately extract the optimal sampling value of the PCMA signal, and the recognition performance is higher than that of AF. The performance of this algorithm is better than that of the CNN-based algorithm and the traditional algorithm under medium signal-to-noise ratio conditions.

[0083] (2) Modulation recognition rate based on MCRLDNN varies with signal-to-noise ratio

[0084] The relationship between the recognition accuracy and SNR based on the MCRLDD algorithm is as follows: Figure 6 The PCMA signal recognition rate is shown as the signal-to-noise ratio changes.

[0085] Depend on Figure 6 As can be seen, the recognition accuracy of each modulation type gradually increases with increasing signal-to-noise ratio. When the SNR is 10dB, the recognition accuracy is almost always above 90%. When the SNR increases to 12dB, the recognition accuracy of BPSK approaches 100%, and the recognition accuracy of the other modulation types also exceeds 95%. This demonstrates that the algorithm has high modulation recognition accuracy at medium and high signal-to-noise ratios. Comparing the recognition accuracy curves of the four modulation types versus SNR reveals that, under the same signal-to-noise ratio, the recognition accuracy of BPSK, QPSK, and 8PSK decreases in that order, indicating that higher-order modulation types have relatively lower recognition accuracy.

[0086] (3) Recognition rates of different algorithms vary with signal-to-noise ratio

[0087] The relationship curves between the recognition accuracy and SNR of different algorithms are as follows: Figure 7 shown.

[0088] Simulation comparisons were carried out on the recognition algorithms based on MCRLDD algorithm, CNN algorithm, discrete spectral line features based on high-order cumulant features and quartic spectrum, modulation recognition algorithm based on joint feature parameters and modulation recognition algorithm based on 4 high-order cumulant features and 1 likelihood feature. The latter three recognition algorithms are denoted as DLSF, UCP and CLC respectively.

[0089] from Figure 7As can be seen, under high signal-to-noise ratios, the average recognition accuracy of this algorithm and other algorithms exceeds 95%. However, under medium signal-to-noise ratio conditions, the average recognition accuracy of this method exceeds 90%, outperforming other methods. When the signal-to-noise ratio is 7dB, the average recognition accuracy of this method reaches 84%, while the average recognition accuracy of other methods is only 56% to 82%.

[0090] The MCRLDNN algorithm can effectively detect PCMA signals and identify their modulation modes. Moreover, compared with traditional detection and modulation recognition algorithms and CNN algorithms, the recognition accuracy is higher under medium signal-to-noise ratio conditions.

[0091] Based on the PCMA signal recognition method provided in the above embodiment, the present invention provides a device for performing the above PCMA signal recognition. The structural diagram of the PCMA signal recognition device is shown in FIG. Figure 8 As shown, the PCMA signal recognition device includes:

[0092] In some specific embodiments, the text image tampering detection device further includes:

[0093] The signal acquisition module 10 is used to acquire a paired carrier multiple access (PCMA) signal to be identified.

[0094] The paired carrier multiple access PCMA signal is generally composed of a mixed signal consisting of BPSK, QPSK, 8PSK PCMA signals and BPSK, QPSK, 8PSK single-channel signals, and is received to obtain the paired carrier multiple access PCMA signal.

[0095] The spatial feature acquisition module 20 is used to input the waveform and spectrum of the PCMA signal into the residual neural network ReNet respectively to extract the spatial features of the waveform and spectrum.

[0096] The waveform and spectrum of the PCMA signal are respectively input into the residual neural network ResNet. Compared with the method of directly inputting signal waveform data in the prior art, this application adopts time-frequency diagram and waveform diagram as input data form, which suppresses the interference of other information in the signal. It can also be regarded as a preliminary extraction of target features, which helps the training of the network model. Therefore, it shows better recognition performance in some methods. The method based on the signal transformation domain (time-frequency diagram and waveform diagram) can better reflect the signal feature distribution.

[0097] The spatiotemporal correlation acquisition module 30 is configured to fuse the spatial features of the waveform and the spectrum and acquire the spatiotemporal correlation of the waveform and the spectrum.

[0098] The spatial feature acquisition module 20 uses a neural network architecture to extract the image features of the signal time-frequency diagram and waveform diagram, and uses the time-frequency characteristics existing in the signal for detection and recognition. The features extracted from the dual channels are fused to obtain a joint feature, that is, the spatiotemporal correlation of the waveform and spectrum.

[0099] The recognition and classification module 40 is used to recognize and classify the PCMA signal according to the spatiotemporal correlation through a fully connected layer.

[0100] The features are mapped through the fully connected layer, and the PCMA signals are identified and classified according to the preset function.

[0101] In some specific embodiments, the temporal features of the PCMA signal are extracted through a long short-term memory (LSTM) layer.

[0102] LSTM can extract temporal features, while the 2D convolutional layer can extract spatial features of PCMA signals. The temporal features of the signal are extracted through the LSTM layer so that they can be combined with the spatial features obtained by the convolutional layer to obtain spatiotemporal correlation.

[0103] The spatiotemporal correlation between the waveform and the spectrum is obtained according to the temporal feature and the spatial feature.

[0104] MCLDNN integrates CNN, LSTM, and fully connected layer deep neural networks into a unified structure to leverage their complementarity and synergy for spatiotemporal feature extraction and classification. The features extracted from the two channels are fused to obtain joint features, and the effective features obtained are then input into the classification layer for signal classification, detection, and recognition.

[0105] In some specific embodiments, the waveform and spectrum of the PCMA signal are respectively input into a convolutional layer to reduce the noise variance and abstract the waveform and spectrum of the PCMA signal.

[0106] The waveform and spectrum of the PCMA signal are input into the two-dimensional convolution layer respectively to reduce the noise variance and provide better features for LSTM. The multi-channel input and processing structure captures the features of input representation at different scales and effectively utilizes the complementary information of I channel, Q channel and I / Q multi-channel data.

[0107] In some specific embodiments, the signal is mapped through a fully connected FC layer and fitted using a preset algorithm, wherein the preset algorithm is a dropout algorithm.

[0108] Dropout significantly reduces overfitting by ignoring half of the feature detectors in each training batch (setting half of the hidden layer node values ​​to 0). This approach reduces interactions between feature detectors, where some detectors rely on others to function. Using dropout prevents model overfitting.

[0109] In some specific embodiments, PCMA signals are identified and classified using an activation function with 6 units, wherein the function is a Softmax function.

[0110] The fully connected layer with 6 units and the activation function of the Softmax function classifies the signal, and there are 6 classification results corresponding to different detection and recognition results.

[0111] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0112] The above is a detailed introduction to the solution provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for identifying a PCMA signal, characterized in that: The method comprises: Acquire a paired carrier multiple access (PCMA) signal to be identified; Inputting the waveform and spectrum of the PCMA signal into the residual neural network ResNet respectively to extract the spatial features of the waveform and spectrum; fusing the spatial features of the waveform and spectrum and obtaining the spatiotemporal correlation between the waveform and spectrum; The PCMA signal is identified and classified according to the spatiotemporal correlation through the fully connected layer.

2. The method according to claim 1, characterized in that The step of fusing the spatial features of the waveform and the spectrum and obtaining the spatiotemporal correlation between the waveform and the spectrum includes: Extracting the temporal features of the PCMA signal through a long short-term memory (LSTM) layer; The spatiotemporal correlation between the waveform and the spectrum is obtained according to the temporal feature and the spatial feature.

3. The method according to claim 2, characterized in that Before the waveform and spectrum of the PCMA signal are respectively input into the residual neural network ResNet, the method further includes: The waveform and spectrum of the PCMA signal are respectively input into the convolution layer to reduce the noise variance and abstract the waveform and spectrum of the PCMA signal.

4. The method according to claim 1, wherein After fusing the spatial features of the waveform and spectrum and obtaining the spatiotemporal correlation between the waveform and spectrum, it also includes: The signal is mapped through a fully connected FC layer and fitted using a preset algorithm, wherein the preset algorithm is a dropout algorithm.

5. The method according to claim 1, wherein The identifying and classifying the PCMA signal includes: The PCMA signal is identified and classified by an activation function with 6 units, wherein the function is a Softmax function.

6. A PCMA signal recognition device, characterized in that: The device comprises: A signal acquisition module, used to acquire a paired carrier multiple access (PCMA) signal to be identified; A spatial feature acquisition module is used to input the waveform and spectrum of the PCMA signal into the residual neural network ResNet respectively to extract the spatial features of the waveform and spectrum; A spatiotemporal correlation acquisition module is used to fuse the spatial features of the waveform and spectrum and obtain the spatiotemporal correlation of the waveform and spectrum; The recognition and classification module is used to recognize and classify the PCMA signal according to the spatiotemporal correlation through a fully connected layer.

7. The device according to claim 6, characterized in that The step of fusing the spatial features of the waveform and the spectrum and obtaining the spatiotemporal correlation between the waveform and the spectrum includes: Extracting the temporal features of the PCMA signal through a long short-term memory (LSTM) layer; The spatiotemporal correlation between the waveform and the spectrum is obtained according to the temporal feature and the spatial feature.

8. The device according to claim 7, characterized in that The spatial feature acquisition module further includes: The signal processing module is used to input the waveform and spectrum of the PCMA signal into the convolution layer respectively, reduce the noise variance and abstract the waveform and spectrum of the PCMA signal.

9. The device according to claim 6, characterized in that After fusing the spatial features of the waveform and spectrum and obtaining the spatiotemporal correlation between the waveform and spectrum, it also includes: The signal is mapped through a fully connected FC layer and fitted using a preset algorithm, wherein the preset algorithm is a dropout algorithm.

10. The device according to claim 6, characterized in that The identifying and classifying the PCMA signal includes: The PCMA signal is identified and classified by an activation function with 6 units, wherein the function is a Softmax function.

Images