A reversible jump sampler for polynomial-phase signals

Abstract

We use reversible jump Markov chain Monte Carlo (MCMC) methods to address the problem of order and parameters estimation of noisy polynomial-phase signals within a Bayesian framework. As posterior distributions of the parameters are not tractable, MCMC methods are used to simulate them. Efficient model jumping is achieved by proposing the model space moves from the conditional density of the polynomial coefficients, estimated with the one variable at a time Metropolis Hasting algorithm. This algorithm provides simultaneous order and parameters estimation from simulated marginal posterior distributions. Results on simulated data are given and discussed.