Tutorial --- 5/11/98 Morning (2)

Simulation-Based Computational Methods for Statistical Signal Processing

by Patrick Duvaut

In many applications such as RADAR, SONAR, Digital Communications, Non-destructive Testing, Geophysical Data Analysis, Speech Processing, the search for Bayesian estimators struggles with non-gaussianity, non-stationarity, and non-linearity. Under those constraints, analytical derivation of Bayesian solutions is almost impossible. Moreover, numerical approximations lead to inaccurate estimates. This tutorial presents to newcomers recently developed, powerful, effective simulation- based methods e.g. Monte Carlo (Gibbs sampler, Metropolis-Hastings); Reversible Jump Model selection; Importance Sampling. Many illustrations will be given, of Equalization, Array Processing, Spectral Analysis, Statistical Analysis of physical data and Audio Signal Processing.

NOTE: This tutorial is prepared in collaboration with C. ANDRIEU, A. DOUCET, B.FITZGERALD and S.GODSILL.