Underwater acoustic communication modulation type identification method based on intelligent signal image identification

Abstract

The invention discloses an underwater acoustic communication modulation type identification method based on intelligent signal image identification, and the method comprises the following steps: (1) signal preprocessing: carrying out the preprocessing of a signal through employing an impulse noise suppression technology for the noise interference in an underwater acoustic channel; (2) time-frequency feature extraction: extracting time-frequency graph features of the preprocessed signals by using short-time Fourier transform STFT so as to effectively distinguish FSK, OFDM and single-carrier signals; (3) first convolutional neural network classification: constructing a first CNN for classification by using depth separable convolution, inputting CNN1 from the time-frequency graph features of the signals, if the output result is an FSK or OFDM signal, completing identification, and if the output result is identified as a single-carrier signal, continuing to extract constellation graph features in the next step; (4) extracting features of the constellation diagram; and (5) carrying out second convolutional neural network classification.

Description

Technical Field

[0001] The invention belongs to the technical field of underwater acoustic communication digital signal processing and relates to an underwater acoustic communication modulation type identification method. Background Art

[0002] Underwater acoustic communication is an indispensable part of modern marine science and technology, especially in underwater exploration, seabed resource exploitation, national defense and security. With the increasing attention paid to marine resources by various countries, the criticality of underwater acoustic communication technology has become more prominent. However, the underwater acoustic communication channel environment is complex and changeable, including multipath effect, frequency selective fading, noise interference and Doppler effect, which poses a huge challenge to the automatic modulation type recognition of underwater acoustic communication signals.

[0003] In underwater acoustic communication, due to the complex channel characteristics and limited channel bandwidth resources, the modulation types currently widely used include frequency shift keying (FSK), phase shift keying (PSK), quadrature amplitude modulation (QAM), orthogonal frequency division multiplexing (OFDM), etc. The underwater acoustic communication modulation type recognition method based on intelligent signal image recognition is used to identify the above commonly used modulation types.

[0004] Traditional modulation recognition methods can be divided into two categories: likelihood ratio-based algorithms (LB methods) and feature extraction-based algorithms (FB methods). The LB method relies on the statistical characteristics of the signal, but has high computational complexity and requires a lot of prior knowledge. The FB method extracts features and inputs them into the classifier for recognition. It has strong performance, but the recognition accuracy is greatly reduced in complex underwater acoustic channel environments.

[0005] In recent years, deep learning (DL) technology has been increasingly used in communication systems. DL methods overcome the reliance of traditional methods on feature extraction and show excellent performance. However, the application of existing DL methods in underwater acoustic communication still faces two major challenges: 1. The strong multipath effect and Doppler frequency shift of underwater communication channels make the signal features blurred and difficult to accurately identify; 2. The limited computing resources of underwater acoustic communication nodes make it difficult to effectively deploy complex DL models. Summary of the invention

[0006] In order to realize automatic modulation type recognition in complex underwater acoustic channel environment, the present invention provides an underwater acoustic communication modulation type recognition based on intelligent signal image recognition. This method converts one-dimensional signals into image features through feature extraction, including time-frequency graph features and constellation graph features, and uses lightweight CNN for classification, which has efficient recognition ability and low computing resource consumption.

[0007] A method for identifying underwater acoustic communication modulation types based on intelligent signal image recognition comprises the following steps:

[0008] (1) Signal preprocessing: To address the noise interference in the underwater acoustic channel, the impulse noise suppression technology is used to preprocess the signal;

[0009] (2) Time-frequency feature extraction: Short-time Fourier transform (STFT) is used to extract the time-frequency features of the preprocessed signal to effectively distinguish FSK, OFDM and single-carrier signals;

[0010] (3) First convolutional neural network classification: The first CNN for classification is constructed using depthwise separable convolution. The time-frequency features of the signal are input into CNN1. If the output result is an FSK or OFDM signal, the recognition is completed. If it is recognized as a single-carrier signal, the next step is to extract the constellation features.

[0011] (4) Extract features from the constellation diagram. The method is as follows:

[0012] For single-carrier signals whose time-frequency characteristics cannot be distinguished, the frequency doubling method is used to estimate the carrier frequency of the single-carrier signal, and the estimated carrier frequency is used to demodulate the baseband signal and draw a constellation diagram to reduce the ambiguity of the constellation diagram caused by carrier offset;

[0013] The exponential decay model is applied to enhance the constellation diagram, reduce the impact of multipath effect and noise on the constellation diagram, and obtain the enhanced constellation diagram characteristics;

[0014] (5) Second convolutional neural network classification: A second CNN for classification is constructed using depthwise separable convolution, and the enhanced constellation features extracted from the signal are input into CNN2 to obtain the classification result.

[0015] Furthermore, the exponential decay model is applied to enhance the constellation diagram. The calculation process is as follows:

[0016]

[0017] Among them, E i,j represents the influence of the i-th signal sample on the j-th pixel, P i represents the power of the i-th signal sample, μ represents the exponential decay rate, and d i,j Represents the distance between the i-th signal sample and the center of the j-th pixel. By calculating the influence of each signal sample on each pixel and summing them up, the enhanced constellation diagram feature is obtained.

[0018] Furthermore, the first CNN and the second CNN for classification have the same structure.

[0019] Furthermore, both the first CNN and the second CNN are lightweight convolutional neural networks. The input feature map is extracted through DepthWise-PointWise Block, CoordinateAttention Block, and pooling operations, and finally the classification results are output through the fully connected layer.

[0020] The present invention proposes a method for identifying the modulation type of underwater acoustic communication based on intelligent signal image recognition. First, the time-frequency features of the signal are extracted and classified. For signals that cannot be identified by time-frequency features, the constellation features are further extracted for classification. In a complex underwater environment, efficient modulation type identification of underwater acoustic communication signals is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of the algorithm for identifying the modulation type of underwater acoustic communication based on the intelligent signal image recognition technology.

[0022] Figure 2 This is a diagram of the structure of a lightweight convolutional neural network based on deep separable convolution.

[0023] Figure 3 It is a method for identifying the modulation type of underwater acoustic communication based on intelligent signal image recognition and a confusion matrix was tested.

[0024] Figure 4 It is a sea trial confusion matrix of the underwater acoustic communication modulation type recognition method based on intelligent signal image recognition. DETAILED DESCRIPTION

[0025] The present invention will be described below in conjunction with the accompanying drawings.

[0026] The method for identifying the modulation type of underwater acoustic communication based on intelligent signal image recognition proposed in the present invention comprehensively considers the variable multipath effect and Doppler effect of the underwater acoustic channel, extracts the time-frequency diagram features and constellation diagram features of the signal, and uses a lightweight convolutional neural network to realize the recognition of the modulation type of underwater acoustic communication signals, with a high recognition rate.

[0027] Take the system composed of two underwater communication nodes A and B as an example, assuming that node A is the transmitting node and node B is the receiving node. After node A sends a signal, node B receives the signal, first preprocesses the signal, then extracts the time-frequency features of the signal, and then inputs the time-frequency features into CNN1. If the output is an FSK or OFDM signal, the recognition ends. If the output is a single carrier signal, the constellation features are extracted and input into CNN2 to obtain the classification result. CNN2 has the same structure as CNN1.

[0028] The specific operation process of the present invention is as follows:

[0029] 1. Signal preprocessing: After receiving the underwater acoustic signal y(n), the impulse noise suppression technology is first used to preprocess the signal to reduce the interference of impulse noise on subsequent feature extraction, and the signal is normalized to obtain y'(n).

[0030] The specific calculation process of impulse noise suppression technology is as follows:

[0031]

[0032] Wherein, y(n) is the received signal, β=(1+2β0)Q2(n), β0 is a constant, and Q2(n) is the median of |y(n)|.

[0033] The received signal after impulse noise suppression is normalized to facilitate feature extraction. The specific calculation process of normalization is as follows:

[0034]

[0035] Wherein, y(n) is the signal after impulse noise suppression processing.

[0036] 2. Time-frequency graph feature extraction: STFT is used to extract the time-frequency graph features of the signal y'(n). STFT divides the signal into segments and adds windows, and performs fast Fourier transform on each segment to obtain the spectrum information of the signal in each time period.

[0037] The calculation process of STFT is as follows:

[0038]

[0039] Where Y(f,m) is the DFT coefficient at discrete frequency index f and discrete time frame index m, and ω = 2πf / J is the discrete frequency variable at frequency f. The time-frequency diagram is given by Conclude.

[0040] The resolution of the time-frequency graph feature is 128*128*3, which can effectively distinguish FSK, OFDM and single-carrier signals.

[0041] 3.CNN1 classification; The present invention constructs a lightweight convolutional neural network as a classifier based on deep separable convolution. The structure of the convolutional neural network is as follows: Figure 2As shown in the figure, the input of the neural network is the feature map, which is extracted through operations such as DepthWise-PointWise Block (DPBlock), CoordinateAttention Block (CABlock), and pooling, and finally the classification result is output through the fully connected layer. The classifier used to distinguish the time-frequency map features is CNN1. If the classification result is FSK or OFDM, the process ends. If it is a single-carrier signal, the constellation map features are extracted.

[0042] 4. Constellation feature extraction: For single-carrier signals whose time-frequency features cannot be distinguished, the frequency doubling method is first used to estimate the carrier frequency of the single-carrier signal. After obtaining the estimated carrier, the estimated carrier frequency is used to demodulate the baseband signal and draw the constellation diagram. Secondly, the multipath effect and noise of the underwater acoustic channel will cause the constellation diagram to be blurred. The exponential decay model is used to enhance the constellation diagram. The calculation process is as follows:

[0043]

[0044] Among them, E i,j represents the influence of the i-th signal sample on the j-th pixel, P i represents the power of the i-th signal sample, μ represents the exponential decay rate, and d i,j Indicates the distance between the i-th signal sample and the center of the j-th pixel. The enhanced constellation diagram is obtained by calculating the impact of each signal sample on each pixel and summing them. The resolution of the enhanced constellation diagram feature is 128*128*3. This feature can effectively distinguish single-carrier signals such as 2PSK, 4PSK, 16QAM, and 64QAM.

[0045] 5. CNN2 classification: The enhanced constellation features are input into CNN2, where CNN2 has the same structure as CNN1. The network is used to distinguish different single-carrier signals, and the output result is the category of different single-carrier signals.

[0046] pass Figure 3 and Figure 4 From the confusion matrices of the lake test and the sea test, it can be seen that the underwater acoustic communication modulation type recognition method based on intelligent signal image recognition can realize the accurate recognition of 2FSK, 4FSK, 8FSK, 16FSK, OFDM, 2PSK, 4PSK, 16QAM and 64QAM in the complex and changeable underwater acoustic channel environment.

Claims

1. A method for identifying underwater acoustic communication modulation types based on intelligent signal image recognition, comprising the following steps: (1) Signal preprocessing: To address the noise interference in the underwater acoustic channel, the impulse noise suppression technology is used to preprocess the signal; (2) Time-frequency feature extraction: Short-time Fourier transform (STFT) is used to extract the time-frequency features of the preprocessed signal to effectively distinguish FSK, OFDM and single-carrier signals; (3) First convolutional neural network classification: The first CNN for classification is constructed using depthwise separable convolution. The time-frequency features of the signal are input into CNN1. If the output result is an FSK or OFDM signal, the recognition is completed. If it is recognized as a single-carrier signal, the next step is to extract the constellation features. (4) Extract features from the constellation diagram. The method is as follows: For single-carrier signals whose time-frequency characteristics cannot be distinguished, the frequency doubling method is used to estimate the carrier frequency of the single-carrier signal, and the estimated carrier frequency is used to demodulate the baseband signal and draw a constellation diagram to reduce the ambiguity of the constellation diagram caused by carrier offset; The exponential decay model is applied to enhance the constellation diagram, reduce the impact of multipath effect and noise on the constellation diagram, and obtain the enhanced constellation diagram characteristics; (5) Second convolutional neural network classification: A second CNN for classification is constructed using deep separable convolution. The enhanced constellation features extracted from the signal are input into the second CNN to obtain the classification result.

2. The method for identifying the modulation type of underwater acoustic communication according to claim 1, characterized in that: The exponential decay model is applied to enhance the constellation diagram. The calculation process is as follows: Among them, E i,j represents the influence of the i-th signal sample on the j-th pixel, P i represents the power of the i-th signal sample, μ represents the exponential decay rate, and d i,j Represents the distance between the i-th signal sample and the center of the j-th pixel. By calculating the influence of each signal sample on each pixel and summing them up, the enhanced constellation diagram feature is obtained.

3. The method for identifying the modulation type of underwater acoustic communication according to claim 1, characterized in that: The first CNN and the second CNN for classification have the same structure.

4. The method for identifying the modulation type of underwater acoustic communication according to claim 1, characterized in that: Both the first CNN and the second CNN are lightweight convolutional neural networks. The input feature map is extracted through DepthWise-PointWise Block, CoordinateAttentionBlock, and pooling operations, and finally the classification results are output through the fully connected layer.

Images