SEM blind identification of ARMA models application to seismic data

In this paper, we address blind identification of an ARMA model convolved with an impulse sequence via Maximum Likelihood (ML) approach. A Stochastic Expectation Maximization (SEM) implementation of the criterion is considered. The problem of ARMA models with long impulse response is addressed as well as the SEM initialization problem. The model estimation is performed in two steps: First, a truncated estimate of the wavelet is obtained from a SEM algorithm. Then improved wavelet estimation is achieved by fitting an ARMA model to the initial MA wavelet using the Prony algorithm. Simulation results show the significant improvement brought by this approach in situations corresponding to seismic data deconvolution.