Research On Modulation Recognition Method Of Underwater Acoustic Signal Based On Deep Learning

The country’s military and economic development is currently increasingly centered on the ocean field,and these developments are inextricably linked to underwater acoustic communication.Modulation recognition technology is a particularly critical communication technology in underwater acoustic communication systems.Underwater channels are complex and changeable,and it is necessary to select the most suitable communication modulation scheme for the current channel environment according to the current channel conditions.However,in a communication system based on multiple modulation modes,the communication parties need to determine the current modulation mode through a handshake signal.Underwater complex noise interference and changing channel environment can easily lead to errors in the handshake signal,causing the receiving end to adopt an unmatched demodulation method,which in turn leads to serious errors in the demodulated data.The use of modulation mode intelligent identification technology enables the receiving end to automatically identify the modulation mode of the received signal,which can ensure the accuracy of data demodulation and improve the efficiency and reliability of underwater data transmission.However,there are still many challenges in the automatic recognition technology of the modulation mode of underwater acoustic signals in terms of recognition accuracy and recognition speed.(1)Underwater acoustic communication systems frequently have strict real-time performance requirements,which means that the model can not be too complex.The number of network layers is typically raised in order to gain better recognition accuracy,but more time is lost in the process.Aiming at communication scenarios with high realtime requirements of the system,this paper designs a lightweight neural network that combines Long Short-Term Memory and Convolutional Neural Networks.In the longshort-term memory network layer,the multi-channel design improves the recognition rate and controls the depth of the network.In the convolutional network layer,the idea of sub-channel convolution is used to control the parameters of the network.In order to further manage parameters and prevent the issue of feature loss brought on by pooling operation,one-dimensional convolution operation and elimination pooling operation are simultaneously implemented.Finally,on the Qingdao and Sanya datasets,the recognition accuracy of the underwater acoustic signal modulation method is more than93% in only about 6ms,which can meet most communication systems that require recognition speed.(2)Aiming at the problems of low effectiveness of feature extraction and insufficient recognition accuracy,a double-branch parallel network structure is designed to extract more representative and richer feature information of underwater acoustic signals.One of the branches is designed based on the convolution operation,which fully extracts the local features of the underwater acoustic signal by taking advantage of the advantages of the convolution operation in extracting local features of the data;the other branch is based on the attention mechanism,which can extract more global information of the signal.This model can better solve the feature distortion in the real underwater acoustic communication environment.At the same time,in order to improve the recognition speed of the system,the positional encoding and decoder part in the Transformer structure were removed through reasonable analysis.The network structure was greatly simplified,and the recognition accuracy rate of more than 98% for 8 kinds of modulation signals was achieved.