DOA Estimation Of Vector Hydrophone Based On Improved BP Neural Network And Maximum Likelihood Method

The development of vector hydrophone is of great significance to the understanding of ocean information and the study of underwater acoustic signal.Vector hydrophone is a hot research topic after scalar hydrophone,and is widely used in detection,communication,positioning,guidance,navigation,measurement and so on.Array signal is an important branch of signal processing.The main research problems are as follows: beamforming technology,spatial spectrum estimation,signal source location and source separation.Among them,the spatial spectrum estimation is superresolution estimation of the distribution of the spatial signal's arrival direction.There are two ways to estimate the DOA: one is to build a pure mathematical model and get the result through a lot of calculation,such as the traditional multiple signal classification algorithm(MUSIC),rotation invariant subspace algorithm(ESPRIT),maximum likelihood algorithm(ML),etc.Second,intelligent learning is used for DOA(Direction of Arrival Estimation)Estimation,such as BP neural network,radial basis neural network,etc.This paper mainly studies the DOA estimation based on the improved BP neural network and the maximum likelihood method,and verifies the effectiveness of the proposed method by comparing the experimental results.The main contents of this paper are as follows:1.When traditional BP neural network is doing DOA estimation,it often falls into local optimum because of the improper selection of weight and threshold of BP neural network.Therefore,we use the positive cochaotic double-string locust optimization algorithm to select the weight and threshold of BP neural network,which can effectively find the optimal value in the global scope.Grasshopper optimization algorithm(GOA)is the result of simulation of nature swarm in feeding behavior,not only from its current position and the global best position update position,but also by other locusts location update location,requires all individuals involved in the optimization process,the search more efficient,but into local extremum problems.In this paper,sine and cosine chaotic mapping(SCA)is introduced.After finding the optimal solution in each iteration of the group optimization algorithm,further chaotic iterative optimization is carried out,which can well combine the characteristics of each algorithm.Using the weight and threshold of BP neural network optimized by the positive and chaotic double-string locust optimization algorithm(SCAGOA)to do DOA estimation,the SCAGOA-BP model can make up for the deficiency of improper selection in local minimum,so as to significantly improve the accuracy and the ability to generalize the SNR.The simulation results show that SCAGOA-BP model is superior to BP,GWO-BP,PSO-BP,GOA-BP and SCAGOA-BP models,and not only has better estimation accuracy,better convergence speed and better optimization performance.2.The hybrid mutation grasshopper optimization algorithm(GOA)was obtained by combining the concept of reverse barycenter solution and Cauchy mutation operator,and the maximum likelihood estimation was optimized to estimate the DOA of the vector hydrophone.In this paper,we carry out DOA estimation simulation experiments of the grasshopper optimization algorithm(GOA),the seeker optimization algorithm(SOA),the squirrel search algorithm(SSA),the sine and cosine optimization algorithm(SCA)and the hybrid mutation locust optimization algorithm to optimize the maximum likelihood estimation algorithm under the dual source.It is observed that the mixed variation GOA-ML method proposed in this paper has better goodness of fit and higher stability than other optimization algorithms,which can ensure that the error between the estimated value and the input value is within.On the premise of ensuring the prediction accuracy,the running time has a strong competitiveness,and the SNR generalization ability is also strong.The two improved methods proposed in this paper have some advantages over the previous methods in DOA estimation accuracy,SNR and the generalization ability of the number of sources.It can be seen from the simulation experiment and the measured experiment that they are practical to a certain extent.We hope to provide some new ideas for DOA estimation.