Autoregressive Modeling of Nonstationary Multibeam Sonar Reverberation

This work describes a technique for simulating nonstationary multibeam Gaussian sonar reverberation sequences. The primary objective is to simulate quadrature-demodulated complex digital vector sequences that have a prescribed time-variant covariance function. We use a first-order integral scattering model to relate geometrical, environmental, and sonar parameters to the assumed locally stationary multivariate reverberation power spectrum for each of a set of ranges spanning the total observation range of interest. At each range we compute the multivariate autoregressive (AR) canonical factorization of the power spectrum. The resultant autoregressive models are used to generate multivariate correlated stationary overlapping realizations for each range. Next we combine stationary realizations adjacent in range with an overlapped windowing scheme. This scheme allows the spectral characteristics of the reverberation to change smoothly with time. Finally, a time-dependent gain is applied to the entire realization to adjust the average intensity.