February Fourier Talks 2007

On the Use of Sequential Monte Carlo Methods For Estimating Time-Varying Spectral Signal Parameters

In waveform-agile sensing applications, environment characterization can provide important information in increasing system performance. As a significant aspect of studying environments is their effect on waveform signature, we present a new approach of instantaneous frequency estimation of nonstationary signals using sequential Bayesian techniques. These techniques are based on combining particle filtering and Markov Chain Monte Carlo (MCMC) methods to sequentially estimate highly nonlinear time-varying frequency variations as piecewise linear or power functions. Simultaneously applying parameter estimation and model selection, the new techniques are extended to the instantaneous estimation of multicomponent signals.