Research On Intelligent Recognition Method Of Underwater Acoustic Communication Modulation Based On Deep Fusion Neural Network

Marine information technology is the top priority in the development of marine industry.It plays a vital role in the progress of national marine territorial security,marine resources development,marine climate early warning,fishery and ship manufacturing industries.Moreover,underwater acoustic communication technology is a key component of marine information technology.The working methods of underwater acoustic communication are divided into cooperative and non-cooperative.In the cooperative underwater acoustic communication system,the modulation intelligent recognition technology enables the receiver to automatically identify the modulation mode of the received signal,so as to ensure the correct demodulation method to restore the data and improve the efficiency and reliability of the data transmission in the underwater acoustic communication system.In non-cooperative underwater acoustic communication system,modulation intelligent recognition technology can improve the recognition speed and accuracy,and meet the needs of real-time modulation and recognition accuracy in various fields.However,due to the complex underwater environment,limited performance of communication transceiver and insufficient underwater acoustic signal,the development of intelligent recognition method of underwater acoustic communication modulation mode faces great challenges,including: first,the performance of modulation mode recognition algorithm depends on the effectiveness of feature extraction,and the complex marine environment will seriously interfere with the characteristics of underwater acoustic signal;Second,the high complexity underwater acoustic signal modulation recognition algorithm is difficult to deploy in underwater communication nodes with limited computing power and energy;Thirdly,the underwater acoustic signal modulation recognition algorithm based on simulation data or simulation channel research and design is not suitable for the actual marine communication scene.To solve the above problems,this paper proposes a modulation recognition method of underwater acoustic signal based on deep fusion neural network,which can effectively extract the effective characteristics of underwater acoustic signal,and reduce the complexity of the algorithm by optimizing the network structure of the model.Finally,the performance advantages of the algorithm are verified based on the underwater acoustic signal data sets of the South China Sea and the Yellow Sea obtained by the research group in the early stage.The main work and innovations of this paper include:(1)Aiming at the effectiveness of feature extraction,the signal features are effectively extracted by reasonably designing the vertical structure of the network to ensure the accuracy of the model.Firstly,the cyclic layer with the feature extraction ability of time series data constitutes the shallow layer and convolution layer of the model constitute the deep layer of the model,so as to make up for the deficiency of the feature extraction ability of cyclic neural network layer,and automatically extract the most descriptive features of underwater acoustic signal modulation mode based on fusion network model.Secondly,the proposed model removes the pooling layer of neural network,avoids the problem of signal feature loss caused by pooling,and ensures the accuracy of feature extraction of network model.(2)Under the premise of reasonable network structure,the accuracy of the model is reduced.Firstly,the convolution layer design is improved,and the one-dimensional convolution kernel is used to replace the two-dimensional convolution kernel to form the convolution layer of the model,which can effectively reduce the network parameters.Secondly,optimize the overall structure of the network and design the convolution layer of the model with multi branch architecture,which widens the network width and further improves the learning ability of the model.(3)Compared with underwater acoustic signal modulation recognition based on various machine learning algorithms,the performance advantages of the proposed algorithm are verified.In this paper,the common features of identifying signal modulation methods in existing research are extracted.Based on various machine learning algorithms such as Decision Tree,K-Nearest Neighbor and Support Vector Machine,the grid search method is used to determine the optimal parameter setting of various machine learning algorithms to identify underwater acoustic signal modulation methods.The results show that the recognition accuracy of various machine learning algorithms is not higher than 60% based on the actual underwater acoustic signal data verification in the South China Sea and the Yellow Sea.(4)Compared with the underwater acoustic signal modulation recognition based on various deep learning algorithms and the model proposed in this paper,the performance advantages of the proposed algorithm are verified.The classical Le Net5,Alex Net8,LSTM,CNN-LSTM network models in deep learning and the neural network model proposed in this paper are used to identify the modulation mode of underwater acoustic signal.Under the same parameter setting,the performance of various neural network models is verified based on the underwater acoustic signal data of the South China Sea and the Yellow Sea.The results show that the proposed deep fusion neural network model is better than the above deep learning and machine learning algorithms,the recognition accuracy is more than 99%,and the average recognition time of signal modulation is only about 7ms.(5)Based on the neural network model proposed in this paper,a set of underwater acoustic signal modulation recognition system is developed and designed by using Raspberry Zero 2W embedded device,and the model is transplanted to the embedded device.The actual engineering verification shows that the system can effectively identify the underwater acoustic signals of 8 modulation modes,such as BFSK,QPSK,BPSK,QFSK,16 QAM,64QAM,OFDM and DSSS.To sum up,this paper has carried out relevant work in the field of underwater acoustic signal modulation recognition,carried out research on underwater acoustic signal modulation recognition based on machine learning algorithm and deep learning algorithm respectively,proposed a deep fusion neural network model,constructed an underwater acoustic signal modulation recognition system,and tested and verified that the proposed model has better recognition performance than the above algorithm based on actual sea area data.This work has important theoretical significance for guiding the research of underwater acoustic signal modulation recognition algorithm,promoting the development of adaptive modulation and coding technology,promoting the progress of marine information technology,and has a huge application space in military and civil fields in the future.