Research On Anti-reverberation Technology Based On Deep Learning

Active sonar is often used to explore the marine environment and resources due to its advantages of long detection distance and detection of buried objects on the seabed.However,active sonar is often accompanied by much interference due to its working mode.Reverberation is one of a kind.Reverberation is one of the most critical factors affecting the detection capability of active sonar.In order to explore the reverberation formation principle and reverberation suppression algorithm,this paper proposes a deep learning-based method based on in-depth theoretical research on reverberation signals.The anti-reverberation algorithm combines the reverberation simulation method based on the adversarial neural network,the reverberation area autonomous identification technology,the beamforming technology,and the time-frequency domain anti-reverberation processing technology on the support vector machine.In the case of a low signal-to-mixing ratio,the reverberation is effectively suppressed,and signal detection is performed.The main research contents and achievements of this paper are as follows:1.The principle of reverberation formation and reverberation characteristics are intensely studied.The point scattering model simulation algorithm is used to simulate the reverberation,and the reverberation characteristics of the simulated data are studied.In addition,this paper proposes a reverberation simulation method based on an adversarial neural network.Compared with the point scattering model,this method has better computational complexity and universality advantages.2.The reverberation beamforming algorithm is studied and analyzed.In this paper,the conventional CBF beamforming,LMS adaptive beamforming,and MVDR adaptive beamforming are deeply studied,and three methods are used to process the reverberation simulation data under the background of white noise and reverberation.Whether it is in the background of white noise or the background of reverberation,the three methods can accurately find the direction of the incoming wave of the signal,which can improve the signal mixing ratio,but the LMS adaptive beamforming method has the best effect,and its main lobe is narrower,with lower side lobes.3.The self-adaptive judgment of the reverberation zone is studied and analyzed.In order to judge the reverberation zone after beamforming,this paper proposes an adaptive judgment algorithm for the reverberation zone based on the calculation of the energy gradient characteristics.According to the difference in the energy gradient between the reverberation signal and the environmental noise,the corresponding threshold is set to achieve the purpose of adaptively judging the reverberation zone.4.The time-frequency domain anti-reverberation algorithm is researched and analyzed.In this paper,the traditional principal component inversion(PCI)method,AR pre-whitening,and other methods are studied and based on this research,an anti-reverberation method based on PCI-SVM is proposed.The method combines Principal Component Inverse(PCI)and Support Vector Machine(SVM)to realize the subspace decomposition of the received signal of the sonar array.In this paper,this method is used to process the simulation and sea trial data,and the data processing results are compared with the direct matched filtering method.The comparison results show that the PCI-SVM method used in this paper has good robustness and anti-reverberation ability.It can improve the detection performance of active sonar and has excellent engineering application value.5.The adversarial reverberation model is validated.First,the reverberation data generated based on the adversarial neural network is used to simulate the array signal.Then,the LMS adaptive beamforming method and the PCI-SVM-based time-frequency domain anti-reverberation algorithm are used to suppress the reverberation.Finally,the reverberation is realized.Echo signal detection function under the signal.