Doa Estimation Of Underwater Acoustic Array Signal Based On Deep Learning

The "14th Five-Year plan" for Marine Economic Development proposes to speed up the building of a strong maritime country with Chinese characteristics.Marine information technology is the premise and foundation of the strategy.In order to better observe the ocean,a large number of marine equipment and instruments are used for underwater information collection and transmission.Compared with the single underwater acoustic transducer,the array of hydrophones can effectively improve the quality of the received signal.As one of the main research contents of array signal processing,the direction of arrival(DOA)estimation of underwater acoustic array signal plays an important role in the field of ocean exploration,and is the core foundation of beamforming and location.However,in underwater signal transmission,due to the channel complexity,a large signal transmission loss,and heavy noise interference,the received signal is seriously distorted,and the DOA estimation is poor.Therefore,we propose a DOA estimation method of underwater acoustic array signal based on continuous wavelet transform combined with double branch convolutional neural network(CWT-CNN)for complex marine acoustic channels.This algorithm can not only improve the accuracy and efficiency of DOA estimation,but also ensure the robustness.The main innovations of this paper are as follows:(1)In complex marine environment,aiming at the problem that signal features cannot be effectively extracted under the influence of environmental noise and transmission loss,this work proposes a time-frequency fusion feature extraction method based on continuous wavelet transform,which enhances the connection between signal features and classification tags to improves the accuracy of DOA estimation.Firstly,we introduce the continuous wavelet transform to analyze the received signal of underwater acoustic array,and improve the wavelet coefficient by the linear factor.Meanwhile the improved time-frequency array signal model is built,which can effectively enhance the useful signal and suppress the noise interference.Then,the complex covariance matrix is reconstructed by virtual and real segmentation to form the improved time-frequency covariance matrix,which solves the problem that the complex covariance matrix cannot be input into the convolutional neural network.At the same time,singular value decomposition(SVD)was used to extract the improved time-frequency energy features and large singular value.Finally,we construct the multi-angle time-frequency fusion feature by the covariance matrix feature,time-frequency energy dimension reduction feature and large singular value matrix to improve the completeness of features and enhance the connection between signal and DOA tags.(2)Rows and columns of fusion features have a strong correlation.Aiming at this characteristic,a double branch convolutional neural network model is designed to reduce the complexity of network structure and improve the accuracy of DOA estimation.The network model firstly performs different convolution operations through parallel branches to enhance the row and column associations of features respectively.Then,we integrate the output results of double branches to enhance the association of parallel branches.Finally,the fully connected network outputs the estimation results.In addition,the pooling layer in convolutional neural network is removed to avoid the loss of effective features.(3)Finally,the effectiveness of proposed CWT-CNN algorithm in the field of underwater acoustic array signals DOA estimation is verified by simulation experiments.In this paper,BELLHOP toolbox is used to construct underwater acoustic channel model and carry out simulation experiments.The estimation accuracy and average running time are used as evaluation indexes.As compared with other DOA algorithms,the proposed algorithm shows a 5%–20% higher accuracy at 7 different values of SNR.In addition,the proposed algorithm is less computationally complex.This work greatly improve the DOA estimation accuracy and speed of underwater acoustic array signal under the influence of complex noise interference and transmission loss,and provide theoretical guidance and technical support for improving the reliability of underwater acoustic communication,which has both theoretical value and practical application value.