Research on Gaussian Mixture Auto-regressive Reverberation Modeling and Whitening Algorithm

For active sonar, reverberation is the main interference. Different from the marine environmental noise, reverberation is non Gaussian in time domain and non white in frequency domain. In view of this, this paper uses Gaussian mixture auto-regressive(GMAR) model to model reverberation, and then uses expectation maximization(EM) algorithm to estimate the model parameters to build “whitening filter”. Simulation results show that EM algorithm can effectively estimate the parameters of reverberation model. The processing results of lake experiment data show that the model has good fitting ability to the probability density function(PDF) of reverberation, and the “whitening filter” constructed can effectively suppress the reverberation band near zero Doppler.