Sequential Bayesian Filtering with Particle Smoother for Improving Frequency Estimation in Frequency Domain Approach

Abstract

In signal processing, frequency estimation is one of the most important tasks for enormous number of applications. Particle filtering has been implemented intensively for the purpose of frequency estimation and the results were found to be excellent for many cases, under different time-frequency representation of the signal and different particle filter implementations. This work presents an enhancement of frequency estimation by using a particle smoother (backward particle filter) to enhance the frequency estimates as compared to the forward particle filter (PF). Calculated from the short-time Fourier transforms (STFTs), the signal was analyzed in the frequency domain, acting as a measurement model of the PF framework. Simulation results exhibit the advantage of the particle smoother over the forward PF. Demonstrated via the frequency estimates from both filters, particle smoother delivers better tracking results than the forward PF under low noise levels.