Effect of Window Functions on the Sequential Bayesian Filtering Based Frequency Estimation
Nattapol Aunsri · 2018
Window functions are typically used in time-frequency (TF) signal analysis, especially in spectrogram calculation. In frequency estimation problems, most methods for solving such tasks rely on the spectrogram-based approach where the time-series of the signal is considered in the frequency domain via the short-time Fourier transform (STFT). Sequential Bayesian filtering has been one of the most powerful methods for sequential estimation of time-varying parameters where the signal model and observation data are complicated in terms of nonliear/non-Gaussian assumptions. The method integrates mathematical and statistical properties of the signal in order to construct a posterior probability density functions of the interesting parameters. Therefore, when window function is applied in STFT, the statistical property of the transformed domain may be destroy, resulting in the movement of the frequency estimates according to the window type used in spectrogram calculation. Therefore, we investigate in this work the effect of windows on the accuracy of the sequential Bayesian filtering based frequency estimation. Simulation results show that window has some effect only when the noise level is high.