Multi-domain feature fusion underwater acoustic communication modulation mode autonomous discrimination method

Abstract

The invention belongs to the technical field of underwater acoustic communication signal and digital signal processing. The invention provides a multi-domain feature fusion underwater acoustic communication modulation mode autonomous discrimination method. According to the embodiment of the invention, the data is preprocessed, and the orthogonal component and the in-phase component of the communication signal are extracted and input into a time domain network branch. And the frequency domain waveform obtained after Fourier transform is carried out on the signal is input into a frequency domain network branch. And inputting a time-frequency diagram obtained by performing continuous wavelet transform on the signal into a time-frequency diagram network branch. And carrying out correlation calculation on the time domain network branch, the frequency domain network branch and the continuous wavelet transform time-frequency graph network branch, and carrying out weighted fusion on the spliced network branches to obtain features. And after feature fusion is completed, the modulation modes of the underwater acoustic communication signals are classified through a full connection layer and a softmax layer. Feature information of a time domain, a frequency domain and a wavelet packet time-frequency graph is used as neural network input, feature results are fused flexibly, and robustness of underwater acoustic communication signal modulation mode autonomous recognition is improved.

Description

Technical Field

[0001] The disclosed embodiments relate to the technical field of underwater acoustic communication signals and digital signal processing, and in particular to a method for autonomously distinguishing underwater acoustic communication modulation modes by fusing multi-domain features. Background Art

[0002] With the rapid development of marine adaptive modulation and coding, non-cooperative communication, communication confrontation, and military intelligence collection and reconnaissance, the importance of autonomous discrimination of underwater acoustic communication modulation mode has become increasingly prominent. In the military field, autonomous discrimination of underwater acoustic communication modulation mode can efficiently collect and analyze enemy communication signals in complex underwater environments, provide key intelligence support, and thus provide an important basis for military decision-making; at the same time, by quickly and accurately identifying the modulation mode of enemy signals, effective interference and anti-interference strategies can be formulated to improve the effect of communication confrontation. In the civilian field, adaptive modulation and coding technology dynamically adjusts parameters such as modulation order and modulation mode by sensing changes in communication conditions, thereby improving the utilization of frequency band while ensuring the reliability of the communication system. At the receiving end, it is necessary to autonomously discriminate the modulation mode of underwater acoustic communication to improve the reliability and stability of communication nodes. Therefore, there is an urgent need for an autonomous discrimination method for underwater acoustic communication modulation mode that can quickly and accurately estimate the modulation mode of the received underwater acoustic signal.

[0003] Traditional methods for autonomous discrimination of underwater acoustic communication modulation modes include statistical classifiers, support vector machines, decision trees, and other methods. Traditional algorithms have achieved certain results in underwater acoustic communication modulation mode recognition, but they are highly dependent on feature selection and extraction. For unknown or complex signal features, the recognition effect is poor and can only be applied in restricted environments. In the face of complex and changing underwater acoustic environments and increasingly diverse modulation methods, the limitations of their performance and robustness are becoming increasingly obvious. Deep learning has been applied to autonomous discrimination of underwater acoustic communication modulation modes due to its powerful autonomous feature extraction capabilities, excellent generalization capabilities, and the ability to process high-dimensional features and nonlinear relationships of data. However, most existing recognition methods based on deep learning only use single domain information of the signal waveform (such as time domain information, frequency domain information, or time-frequency domain information) for feature representation. When processing complex underwater acoustic communication signals, they ignore the complementarity of feature representation between multiple domain information of the signal, do not consider that different communication methods have specific expressions in specific domains, and cannot fully capture all the characteristics of the signal, thus affecting recognition performance and robustness.

[0004] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0005] It should be noted that this section is intended to provide background or context for the technical solutions of the present disclosure stated in the claims. The description herein is not admitted to be prior art by virtue of being included in this section. Summary of the invention

[0006] The purpose of the embodiments of the present disclosure is to provide a method for autonomously distinguishing the modulation mode of underwater acoustic communication by fusing multi-domain features, thereby overcoming one or more problems caused by the limitations and defects of related technologies at least to a certain extent.

[0007] According to an embodiment of the present disclosure, a method for autonomously distinguishing a modulation mode of underwater acoustic communication by fusion of multi-domain features is provided, the method comprising: Constructing an underwater acoustic communication modulation mode data set, and preprocessing the underwater acoustic communication modulation mode data set; Dividing the preprocessed underwater acoustic communication modulation mode data set into a training set and a test set according to a preset ratio; Constructing a multi-domain fusion communication pattern recognition network; wherein the multi-domain fusion communication pattern recognition network includes a time domain branch network, a frequency domain branch network, a time-frequency graph branch network and a feature fusion network, and the time domain branch network, the frequency domain branch network and the time-frequency graph branch network are in parallel and are all connected to the feature fusion network; Using the training set and the test set to train the multi-domain fusion communication pattern recognition network respectively to obtain the trained multi-domain fusion communication pattern recognition network; The data to be judged is input into the trained multi-domain fusion communication pattern recognition network to obtain a judgment result.

[0008] Furthermore, constructing an underwater acoustic communication modulation mode data set, and preprocessing the underwater acoustic communication modulation mode data set includes: Based on the measured signals received in the underwater acoustic environment, the simulated signals generated by simulation and the underwater acoustic data set in the database, a data set of underwater acoustic communication modulation patterns is constructed; Demodulating the measured signal and the simulated signal to baseband, extracting the orthogonal components and in-phase components of the measured signal and the simulated signal, intercepting baseband signals of the same length, and classifying and labeling the measured signal and the simulated signal according to the signal-to-noise ratio and the type of the communication signal; For all data samples in the underwater acoustic communication modulation pattern dataset After normalization, we get , and each data sample Each element of is moved to the range of [0,1]; the normalized expression is:

[0009] In the formula, is the normalized underwater acoustic communication modulation pattern dataset, is the maximum amplitude value in the original data set, is the minimum amplitude value in the original data set; For each of the data samples Perform a 128-point Fourier transform to obtain a frequency domain signal of the same length as the input of the frequency domain branch network; wherein the expression of the Fourier transform is:

[0010] In the formula, is the nth element of a single underwater acoustic communication data after normalization, is the kth element of the frequency domain signal vector, N is the number of Fourier transform points, n is the coordinate of the time domain signal element, k is the coordinate of the frequency domain signal element, and j is an imaginary number; For each of the data samples Perform continuous wavelet transform and draw a wavelet transform time-frequency diagram as the input of the time-frequency diagram branch network; wherein the expression of wavelet transform is:

[0011] In the formula, is the nth element of a single underwater acoustic communication data after normalization, is the wavelet function, a is the ath row of the matrix after wavelet transformation, b is the bth column of the matrix after wavelet transformation, c o is a constant, d o is a constant.

[0012] Furthermore, the step of dividing the preprocessed underwater acoustic communication modulation mode data set into a training set and a test set according to a preset ratio includes: All the data samples in the preprocessed underwater acoustic communication modulation mode data set are randomly shuffled, and the training set and the test set are randomly extracted according to a preset ratio.

[0013] Furthermore, the time domain branch network includes: Time domain input layer, time domain convolution layer 1, time domain pooling layer 1, time domain convolution layer 2, time domain pooling layer 2, time domain convolution layer 3, time domain pooling layer 3, time domain convolution layer 4, time domain pooling layer 4 and time domain output layer; among them, The input of the time domain input layer of the time domain branch network is a matrix size of 2×128, the time domain convolution layer 1 is 2×128×256, the time domain pooling layer 1 is 2×64×256, the time domain convolution layer 2 is 2×32×128, the time domain pooling layer 2 is 2×32×128, the time domain convolution layer 3 is 2×32×64, the time domain pooling layer 3 is 2×16×64, the time domain convolution layer 4 is 2×16×64, the time domain pooling layer 4 is 2×8×32, and the time domain output layer outputs 128 time domain features.

[0014] Furthermore, the frequency domain branch network includes: Frequency domain input layer, frequency domain convolution layer 1, frequency domain pooling layer 1, frequency domain convolution layer 1, frequency domain pooling layer 2, frequency domain convolution layer 3, frequency domain pooling layer 3, frequency domain convolution layer 4, frequency domain pooling layer 4 and frequency domain output layer; among them, The input of the frequency domain input layer of the frequency domain branch network is a matrix size of 1×128, the frequency domain convolution layer 1 is of size 1×128×256, the frequency domain pooling layer 1 is of size 1×64×256, the frequency domain convolution layer 1 is of size 1×32×128, the frequency domain pooling layer 2 is of size 1×32×128, the frequency domain convolution layer 3 is of size 1×32×64, the frequency domain pooling layer 3 is of size 1×16×64, the frequency domain convolution layer 4 is of size 1×16×64, the frequency domain pooling layer 4 is of size 1×8×32, and the output layer outputs 64 frequency domain features.

[0015] Furthermore, the time-frequency graph branch network includes: Time-frequency graph input, time-frequency graph convolution layer 1, dense connection block 1, time-frequency graph pooling layer 1, dense connection block 2, time-frequency graph convolution layer 2, time-frequency graph pooling layer 2, time-frequency graph convolution layer 3, time-frequency graph pooling layer 3 and time-frequency graph output layer; among them, The frequency graph input layer size of the frequency graph branch network is 128×128×3, the frequency graph convolution layer 1 outputs 64×64×144, the dense connection block 1 outputs a size of 64×64×144, the frequency graph pooling layer 1 outputs 32×32×72, the dense connection block 2 outputs a size of 32×32×144, the frequency graph convolution layer 2 outputs 16×16×72, the frequency graph pooling layer 2 outputs 16×16×36, the frequency graph convolution layer 3 outputs 16×16×36, the frequency graph pooling layer 3 outputs 8×8×18, and the frequency graph output layer outputs 64 frequency graph features.

[0016] Furthermore, the training step of the multi-domain fusion communication pattern recognition network includes: Inputting the time domain signal in the data sample into the time domain branch network to obtain the time domain feature; Inputting the frequency domain signal in the data sample into the frequency domain branch network to obtain the frequency domain feature; Inputting the time domain graph in the data sample into the time domain graph branch network to obtain the time domain graph feature; Calculating weight values ​​of the time domain feature, the frequency domain feature and the time domain graph feature respectively; Based on the weight values ​​of the respective features, weighted serial concatenation is performed on the feature descriptors of the time domain feature, the frequency domain feature and the time-frequency graph feature to obtain a fusion feature; The fused features are passed through a fully connected layer and a softmax layer to complete the classification of all modulation modes.

[0017] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects: In the embodiment of the present disclosure, through the above-mentioned multi-domain feature fusion underwater acoustic communication modulation mode autonomous discrimination method, on the one hand, the data is preprocessed to extract the orthogonal components and in-phase components of the communication signal and input into the time domain network branch. The frequency domain waveform after Fourier transformation of the signal is input into the frequency domain network branch. The time-frequency diagram obtained by continuous wavelet transform of the signal is input into the two-dimensional time-frequency diagram network branch. The correlation is calculated for the time domain network branch, the frequency domain network branch, and the continuous wavelet transform time-frequency diagram network branch, and the features obtained by weighted fusion splicing network branches. After the feature fusion is completed, the modulation mode of the underwater acoustic communication signal is classified through the fully connected layer and the softmax layer. On the other hand, by combining the feature information of the time domain, frequency domain and wavelet packet time-frequency diagram, a plurality of deep learning networks are used to extract comprehensive features of the signal, and finally the multi-domain data is fused, which makes full use of the complementarity between different domains and different transformation features, fills the limitations of extracting high-dimensional features from single domain data, and more comprehensively captures the complex characteristics of the signal, thereby achieving higher performance modulation recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0019] Figure 1 A step diagram showing a method for autonomously distinguishing a modulation mode of underwater acoustic communication by fusion of multi-domain features in an exemplary embodiment of the present disclosure is shown; Figure 2 A flow chart showing simulation signal generation in an exemplary embodiment of the present disclosure; Figure 3A specific flow chart showing the autonomous discrimination method of underwater acoustic communication modulation mode by fusion of multi-domain features in an exemplary embodiment of the present disclosure is shown; Figure 4 A schematic diagram showing the structure of the convolutional layer and the fully connected layer of the multi-domain fusion communication pattern recognition network in an exemplary embodiment of the present disclosure is shown; Figure 5 A result diagram showing the recognition accuracy of communication modulation mode by fusion of multi-domain features in an exemplary embodiment of the present disclosure; Figure 6 A result diagram showing the average recognition accuracy of the multi-domain feature fusion communication modulation mode in an exemplary embodiment of the present disclosure; Figure 7 The overall confusion matrix of the convolutional network in the exemplary embodiment of the present disclosure is shown; Figure 8 The confusion matrices under different signal-to-noise ratios in an exemplary embodiment of the present disclosure are shown. DETAILED DESCRIPTION

[0020] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0021] In addition, the accompanying drawings are only schematic illustrations of the embodiments of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0022] This example embodiment provides a method for autonomously distinguishing modulation modes of underwater acoustic communication by fusing multi-domain features. Figure 1 As shown in , the multi-domain feature fusion underwater acoustic communication modulation mode autonomous identification method may include: step S101 to step S105.

[0023] Step S101: constructing an underwater acoustic communication modulation mode data set, and preprocessing the underwater acoustic communication modulation mode data set; Step S102: dividing the preprocessed underwater acoustic communication modulation mode data set into a training set and a test set according to a preset ratio; Step S103: constructing a multi-domain fusion communication pattern recognition network; wherein the multi-domain fusion communication pattern recognition network includes a time domain branch network, a frequency domain branch network, a time-frequency graph branch network and a feature fusion network, and the time domain branch network, the frequency domain branch network and the time-frequency graph branch network are in parallel and are all connected to the feature fusion network; Step S104: training the multi-domain fusion communication pattern recognition network using the training set and the test set respectively to obtain the trained multi-domain fusion communication pattern recognition network; Step S105: input the data to be judged into the trained multi-domain fusion communication pattern recognition network to obtain a judgment result.

[0024] Through the above-mentioned multi-domain feature fusion underwater acoustic communication modulation mode autonomous discrimination method, on the one hand, the data is preprocessed to extract the orthogonal components and in-phase components of the communication signal and input into the time domain network branch. The frequency domain waveform after Fourier transform of the signal is input into the frequency domain network branch. The time-frequency graph obtained by continuous wavelet transform of the signal is input into the two-dimensional time-frequency graph network branch. The correlation is calculated for the time domain network branch, the frequency domain network branch, and the continuous wavelet transform time-frequency graph network branch, and the features obtained by weighted fusion splicing network branches. After the feature fusion is completed, the modulation mode of the underwater acoustic communication signal is classified through the fully connected layer and the softmax layer. On the other hand, by combining the feature information of the time domain, frequency domain and wavelet packet time-frequency graph, multiple deep learning networks are used to extract comprehensive features of the signal, and finally the multi-domain data is fused, which makes full use of the complementarity between different domains and different transformation features, fills the limitations of extracting high-dimensional features from single domain data, and more comprehensively captures the complex characteristics of the signal, thereby achieving higher performance modulation recognition.

[0025] Next, we will refer to Figures 1 to 8 Each step of the above-mentioned multi-domain feature fusion underwater acoustic communication modulation mode autonomous discrimination method in this example implementation is described in more detail.

[0026] In step S101, a data set of underwater acoustic communication modulation modes is constructed. The measured signals obtained by sampling underwater communication signals, the simulation signals obtained by simulating underwater acoustic channels, and the existing public data sets of communication signals are used as the data set of the autonomous discrimination method of underwater acoustic communication modulation modes; the 11 modulation modes of the data set constructed in this paper are 8PSK, AM-DSB, AM-SSB, BPSK, CPFSK, GFSK, 4-PAM, 16-QAM, 64-QAM, QPSK, and WBFM.

[0027] Perform necessary preprocessing on the data set to make it more reliable. Demodulate the signal to baseband and extract the orthogonal and in-phase components of the communication signal. Add Gaussian noise to the simulation data and classify the data set according to the signal-to-noise ratio. After normalization, we get , move each element of each data sample to the range of [0,1] to ensure that each element has the same scale and avoid the influence of different scales on the judgment result. The normalized expression is as follows.

[0028]

[0029] In step S102, the data set is randomly shuffled, and a training set and a test set are randomly extracted. The size of each sample matrix is ​​(2, 128), that is, the data of the same direction component and the orthogonal component. The two data are combined and Fourier transformed to obtain the frequency domain signal , the specific expression of Fourier transform is as follows:

[0030] The signal time-frequency diagram is obtained by continuous wavelet transform, and the expression is:

[0031] In step S103 to step S105, a branch network model is constructed. The network of the present application includes a time domain network branch, a frequency domain branch, and a continuous wavelet transform frequency spectrum branch. The input of the time domain network branch is a matrix size of 2×128, the size of convolution layer 1 is 2×128×256, the size of pooling layer 1 is 2×64×256, the size of convolution layer 2 is 2×32×128, the size of pooling layer 2 is 2×32×128, the size of convolution layer 3 is 2×32×64, the size of pooling layer 3 is 2×16×64, the size of convolution layer 4 is 2×16×64, the size of pooling layer 4 is 2×8×32, and finally 128 features are output.

[0032] The input of the frequency domain network branch is a matrix of size 1×128, the size of convolution layer 1 is 1×128×256, the size of pooling layer 1 is 1×64×256, the size of convolution layer 1 is 1×32×128, the size of pooling layer 2 is 1×32×128, the size of convolution layer 3 is 1×32×64, the size of pooling layer 3 is 1×16×64, the size of convolution layer 4 is 1×16×64, the size of pooling layer 4 is 1×8×32, and finally 64 features are output.

[0033] The time-frequency graph is first compressed by dimensionality reduction to reduce network complexity and improve running speed. The input layer size is 128×128×3, the convolution layer 1 outputs 64×64×144, the dense connection block 1 outputs 64×64×144, the pooling layer 1 outputs 32×32×72, the dense connection block 2 outputs 32×32×144, the convolution layer 2 outputs 16×16×72, the pooling layer 2 outputs 16×16×36, the convolution layer 3 outputs 16×16×36, the pooling layer 3 outputs 8×8×18, and finally outputs 64 features.

[0034] Construct a feature fusion network. The three branch networks output a total of 256 features. In order to make the three types of feature descriptors cooperate with each other to achieve better results, they need to be fused. Since the features output by different branch networks have different relevance to the final communication judgment, weighted fusion of different network features is adopted to reflect the characterization ability of each feature for the communication modulation mode. In terms of weight value calculation, the Pearson correlation coefficient method is selected, and the expression is as follows:

[0035]

[0036]

[0037]

[0038] in, is the Pearson correlation coefficient; cov(X, Y) is the covariance of two variables X and Y, which is used to reflect the relationship between variables X and Y; is the standard deviation of the X variable. is the standard deviation of the Y variable.

[0039] After the weight value calculation is completed, the feature descriptors of time domain features, frequency domain features and continuous wavelet transform time-frequency graph features are weighted serially spliced, and the fused feature descriptors are expressed as:

[0040] According to the analysis of the correlation coefficient in the experiment, the feature weight value of the time domain network branch is 0.35, the feature weight of the frequency domain network branch is 0.25, and the feature weight of the continuous wavelet transform time-frequency graph is 0.4. Finally, through the fully connected layer and the softmax classification layer, the output size is 11 modulation mode classifications.

[0041] In one embodiment, the underwater acoustic communication modulation mode discrimination method of the present application integrates time domain signals, frequency domain signals and continuous wavelet transform time-frequency diagrams. High-dimensional features in different domains are effectively extracted, avoiding the limitation of feature extraction using single-dimensional data.

[0042] In one embodiment, the Pearson correlation method is used to calculate the correlation of the time domain network branch, frequency domain network branch, and continuous wavelet transform time-frequency graph network branch proposed in this article, and the calculated weights are: the time domain network branch feature weight value is 0.35, the frequency domain network branch feature weight is 0.25, and the continuous wavelet transform time-frequency graph feature weight is 0.4.

[0043] In one embodiment, the feature descriptors of time domain features, frequency domain features and continuous wavelet transform time-frequency graph features are weighted serially concatenated, and after feature fusion, 11 types of modulation modes are classified through a fully connected layer and a softmax layer.

[0044] In one embodiment, since the orthogonal components and in-phase components of the communication signal are used for processing and discrimination, the method is not only applicable to autonomous discrimination of underwater acoustic communication modulation modes, but also to other communication scenarios such as electromagnetic communication and optical communication.

[0045] In a specific embodiment, the present application is further described below in conjunction with the drawings and embodiments. The present application includes but is not limited to the following embodiments.

[0046] 1. Build the dataset: In order to train the multi-domain fusion underwater acoustic communication modulation mode autonomous discrimination network, a rich data set is constructed through a variety of data acquisition methods. The data set includes underwater acoustic communication signals received in a real underwater acoustic environment, simulated signals, and existing public data sets. The simulated transmission signal is obtained by simulating a multipath channel and superimposing Gaussian white noise to obtain a received signal with a signal-to-noise ratio ranging from -20dB to 18dB. The specific simulation signal generation process is as follows Figure 2 As shown in the figure, the existing public datasets include RML2016.10a dataset and RML2016.10b dataset. The total number of samples constructed is 50,000.

[0047] 2. Data Preprocessing Demodulate the measured and simulated signals to baseband, extract the orthogonal and in-phase components of the communication signal, intercept the baseband signal of the same length, and classify and label the data according to the signal-to-noise ratio and the type of communication signal. Normalize each data sample and move each element of each data sample to the range of [0,1] to ensure that each element has the same scale and avoid the influence of different scales on the judgment result. The normalized data is used as the input of the time domain branch network.

[0048] Perform 128-point Fourier transform on each data sample to obtain frequency domain signals of the same length as the input of the frequency domain branch network. Perform continuous wavelet transform on each data and draw the wavelet transform time-frequency graph as the input of the time-frequency graph branch network.

[0049] 3. The flowchart of the autonomous identification method of underwater acoustic communication modulation mode by fusion of multi-domain features is as follows Figure 3 As shown in the figure, the input of the time domain branch network is a 2×128 time domain data signal. The input of the frequency domain branch network is a 128-point frequency domain data signal. The input of the time-frequency graph branch network is a 128×128×3 time-frequency image.

[0050] Build a branch network model: The multi-domain feature fusion network convolutional layer and fully connected layer design are as follows: Figure 4 As shown. The network includes a time domain network branch, a frequency domain branch, and a continuous wavelet transform frequency graph branch. The input of the time domain network branch is a matrix of size 2×128, the size of convolution layer 1 is 2×128×256, the size of pooling layer 1 is 2×64×256, the size of convolution layer 2 is 2×32×128, the size of pooling layer 2 is 2×32×128, the size of convolution layer 3 is 2×32×64, the size of pooling layer 3 is 2×16×64, the size of convolution layer 4 is 2×16×64, the size of pooling layer 4 is 2×8×32, and finally 128 features are output. The frequency domain network branch input matrix size is 1×128, the size of convolution layer 1 is 1×128×256, the size of pooling layer 1 is 1×64×256, the size of convolution layer 1 is 1×32×128, the size of pooling layer 2 is 1×32×128, the size of convolution layer 3 is 1×32×64, the size of pooling layer 3 is 1×16×64, the size of convolution layer 4 is 1×16×64, the size of pooling layer 4 is 1×8×32, and finally 64 features are output. The time-frequency graph is first compressed by dimensionality reduction to reduce network complexity and improve running speed. The input layer size is 128×128×3, the convolution layer 1 outputs 64×64×144, the dense connection block 1 outputs 64×64×144, the pooling layer 1 outputs 32×32×72, the dense connection block 2 outputs 32×32×144, the convolution layer 2 outputs 16×16×72, the pooling layer 2 outputs 16×16×36, the convolution layer 3 outputs 16×16×36, the pooling layer 3 outputs 8×8×18, and finally 64 features are output.

[0051] Construct feature fusion network: The three branch networks output a total of 256 features. In order to make the three types of feature descriptors cooperate with each other to achieve better results, they need to be fused. Since the features output by different branch networks have different relevance to the final communication judgment, weighted fusion of different network features is adopted to reflect the characterization ability of each feature for the communication modulation mode. In terms of weight value calculation, the Pearson correlation coefficient method is selected, and the expression is as follows:

[0052]

[0053]

[0054]

[0055] in, is the Pearson correlation coefficient; cov(X, Y) is the covariance of two variables X and Y, which is used to reflect the relationship between variables X and Y; is the standard deviation of the X variable. is the standard deviation of the Y variable.

[0056] After the weight value calculation is completed, the feature descriptors of time domain features, frequency domain features and continuous wavelet transform time-frequency graph features are weighted serially spliced, and the fused feature descriptors are expressed as:

[0057] According to the analysis of the correlation coefficient in the experiment, the feature weight value of the time domain network branch is 0.35, the feature weight of the frequency domain network branch is 0.25, and the feature weight of the continuous wavelet transform time-frequency graph is 0.4. Finally, through the fully connected layer and the softmax classification layer, the output size is 11 modulation mode classifications.

[0058] Model training: Each time the model is trained, the data set is randomly rearranged. 490,000 data are extracted as the training set and 10,000 data are extracted as the validation set. The multi-domain feature fusion underwater acoustic communication modulation mode network model is fully trained using the training signal data set until the detection error loss value of the network model is reduced to below the preset threshold, the training is terminated and the network model is saved.

[0059] Hydroacoustic pattern recognition: Based on the trained network model, the validation data set is verified. The recognition accuracy of multi-domain feature fusion communication modulation mode is as follows: Figure 5 The average recognition accuracy of the multi-domain feature fusion communication modulation mode is shown in Figure 6 As shown. The results show that the recognition accuracy gradually increases with the increase of the signal-to-noise ratio, and the average recognition rate is 84% ​​when the signal-to-noise ratio is greater than 15dB, which has better recognition ability than the traditional method. This application effectively realizes the recognition of the data set.

[0060] The overall confusion matrix of the multi-domain feature fusion communication modulation mode recognition network is as follows Figure 7 The confusion matrix of the multi-domain feature fusion communication modulation mode recognition network under different signal-to-noise ratios is shown in Figure 8The results show that when the signal-to-noise ratio is 18dB, the present application has a recognition rate of more than 92% for modulation modes such as 8PSK, AM-DSB, AM-SSB, and BPSK. It is easier to confuse the WBFM and 16-QAM modulation modes, so the present application is more suitable for the recognition of modulation modes such as 8PSK and AM-DSB.

[0061] Through the above-mentioned multi-domain feature fusion underwater acoustic communication modulation mode autonomous discrimination method, the data is first preprocessed to extract the orthogonal components and in-phase components of the communication signal and input into the time domain network branch. The frequency domain waveform after Fourier transform of the signal is input into the frequency domain network branch. The time-frequency graph obtained by continuous wavelet transform of the signal is input into the two-dimensional time-frequency graph network branch. The Pearson correlation method is used to calculate the correlation of the time domain network branch, frequency domain network branch, and continuous wavelet transform time-frequency graph network branch proposed in this paper, and the features obtained by weighted fusion and splicing network branches are obtained. After completing the feature fusion, the modulation mode of the underwater acoustic communication signal is classified through the fully connected layer and the softmax layer.

[0062] The basic principle and implementation scheme of the method proposed in this application have been verified by computer numerical simulation, and the results show that in the data set constructed in this application and in a typical underwater acoustic channel environment, the method proposed in this application can effectively classify 11 types of underwater acoustic communication signal modulation modes. In the typical embodiment given, the method proposed in this application (compared with the traditional method) can reduce the modulation mode classification accuracy from 70% to 98%.

[0063] It should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc. in the above description indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present disclosure and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the embodiments of the present disclosure.

[0064] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0065] In the embodiments of the present disclosure, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present disclosure can be understood according to specific circumstances.

[0066] In the embodiments of the present disclosure, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features not being in direct contact but being in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes the first feature being directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes the first feature being directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.

[0067] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0068] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A multi-domain feature fusion underwater acoustic communication modulation mode autonomous discrimination method, characterized in that: The method includes: Constructing an underwater acoustic communication modulation mode data set, and preprocessing the underwater acoustic communication modulation mode data set; Dividing the preprocessed underwater acoustic communication modulation mode data set into a training set and a test set according to a preset ratio; Constructing a multi-domain fusion communication pattern recognition network; wherein the multi-domain fusion communication pattern recognition network includes a time domain branch network, a frequency domain branch network, a time-frequency graph branch network and a feature fusion network, and the time domain branch network, the frequency domain branch network and the time-frequency graph branch network are in parallel and are all connected to the feature fusion network; Using the training set and the test set to train the multi-domain fusion communication pattern recognition network respectively to obtain the trained multi-domain fusion communication pattern recognition network; The data to be judged is input into the trained multi-domain fusion communication pattern recognition network to obtain a judgment result.

2. According to the multi-domain feature fusion underwater acoustic communication modulation mode autonomous identification method of claim 1, it is characterized by: The step of constructing an underwater acoustic communication modulation mode data set and preprocessing the underwater acoustic communication modulation mode data set includes: Based on the measured signals received in the underwater acoustic environment, the simulated signals generated by simulation and the underwater acoustic data set in the database, a data set of underwater acoustic communication modulation patterns is constructed; Demodulating the measured signal and the simulated signal to baseband, extracting the orthogonal components and in-phase components of the measured signal and the simulated signal, intercepting baseband signals of the same length, and classifying and labeling the measured signal and the simulated signal according to the signal-to-noise ratio and the type of the communication signal; For all data samples in the underwater acoustic communication modulation pattern dataset After normalization, we get , and each data sample Each element of is moved to the range of [0,1]; the normalized expression is: In the formula, is the normalized underwater acoustic communication modulation pattern dataset, is the maximum amplitude value in the original data set, is the minimum amplitude value in the original data set; For each of the data samples Perform a 128-point Fourier transform to obtain a frequency domain signal of the same length as the input of the frequency domain branch network; wherein the expression of the Fourier transform is: In the formula, is the nth element of a single underwater acoustic communication data after normalization, is the kth element of the frequency domain signal vector, N is the number of Fourier transform points, n is the coordinate of the time domain signal element, k is the coordinate of the frequency domain signal element, and j is an imaginary number; For each of the data samples Perform continuous wavelet transform and draw a wavelet transform time-frequency diagram as the input of the time-frequency diagram branch network; wherein the expression of wavelet transform is: In the formula, is the nth element of a single underwater acoustic communication data after normalization, is the wavelet function, a is the ath row of the matrix after wavelet transformation, b is the bth column of the matrix after wavelet transformation, c o is a constant, d o is a constant.

3. According to claim 2, the method for autonomously distinguishing the modulation mode of underwater acoustic communication by fusion of multi-domain features is characterized in that: The step of dividing the preprocessed underwater acoustic communication modulation mode data set into a training set and a test set according to a preset ratio includes: All the data samples in the preprocessed underwater acoustic communication modulation mode data set are randomly shuffled, and the training set and the test set are randomly extracted according to a preset ratio.

4. According to claim 3, the method for autonomously distinguishing the modulation mode of underwater acoustic communication by fusion of multi-domain features is characterized in that: The time domain branch network comprises: Time domain input layer, time domain convolution layer 1, time domain pooling layer 1, time domain convolution layer 2, time domain pooling layer 2, time domain convolution layer 3, time domain pooling layer 3, time domain convolution layer 4, time domain pooling layer 4 and time domain output layer; among them, The input of the time domain input layer of the time domain branch network is a matrix size of 2×128, the time domain convolution layer 1 is 2×128×256, the time domain pooling layer 1 is 2×64×256, the time domain convolution layer 2 is 2×32×128, the time domain pooling layer 2 is 2×32×128, the time domain convolution layer 3 is 2×32×64, the time domain pooling layer 3 is 2×16×64, the time domain convolution layer 4 is 2×16×64, the time domain pooling layer 4 is 2×8×32, and the time domain output layer outputs 128 time domain features.

5. According to claim 4, the method for autonomously distinguishing the modulation mode of underwater acoustic communication by fusion of multi-domain features is characterized in that: The frequency domain branch network comprises: Frequency domain input layer, frequency domain convolution layer 1, frequency domain pooling layer 1, frequency domain convolution layer 1, frequency domain pooling layer 2, frequency domain convolution layer 3, frequency domain pooling layer 3, frequency domain convolution layer 4, frequency domain pooling layer 4 and frequency domain output layer; among them, The input of the frequency domain input layer of the frequency domain branch network is a matrix size of 1×128, the frequency domain convolution layer 1 is of size 1×128×256, the frequency domain pooling layer 1 is of size 1×64×256, the frequency domain convolution layer 1 is of size 1×32×128, the frequency domain pooling layer 2 is of size 1×32×128, the frequency domain convolution layer 3 is of size 1×32×64, the frequency domain pooling layer 3 is of size 1×16×64, the frequency domain convolution layer 4 is of size 1×16×64, the frequency domain pooling layer 4 is of size 1×8×32, and the output layer outputs 64 frequency domain features.

6. According to claim 5, the method for autonomously distinguishing the modulation mode of underwater acoustic communication by fusion of multi-domain features is characterized in that: The frequency spectrum branch network comprises: Time-frequency graph input, time-frequency graph convolution layer 1, dense connection block 1, time-frequency graph pooling layer 1, dense connection block 2, time-frequency graph convolution layer 2, time-frequency graph pooling layer 2, time-frequency graph convolution layer 3, time-frequency graph pooling layer 3 and time-frequency graph output layer; among them, The frequency graph input layer size of the frequency graph branch network is 128×128×3, the frequency graph convolution layer 1 outputs 64×64×144, the dense connection block 1 outputs a size of 64×64×144, the frequency graph pooling layer 1 outputs 32×32×72, the dense connection block 2 outputs a size of 32×32×144, the frequency graph convolution layer 2 outputs 16×16×72, the frequency graph pooling layer 2 outputs 16×16×36, the frequency graph convolution layer 3 outputs 16×16×36, the frequency graph pooling layer 3 outputs 8×8×18, and the frequency graph output layer outputs 64 frequency graph features.

7. The method for autonomously distinguishing modulation modes of underwater acoustic communication by fusion of multi-domain features according to claim 6 is characterized in that: The training step of the multi-domain fusion communication pattern recognition network includes: Inputting the time domain signal in the data sample into the time domain branch network to obtain the time domain feature; Inputting the frequency domain signal in the data sample into the frequency domain branch network to obtain the frequency domain feature; Inputting the time domain graph in the data sample into the time domain graph branch network to obtain the time domain graph feature; Calculating weight values ​​of the time domain feature, the frequency domain feature and the time domain graph feature respectively; Based on the weight values ​​of the respective features, weighted serial concatenation is performed on the feature descriptors of the time domain feature, the frequency domain feature and the time-frequency graph feature to obtain a fusion feature; The fused features are passed through a fully connected layer and a softmax layer to complete the classification of all modulation modes.

Images