New Robust LPC-Based Method for Time-resolved Morphology of High-noise Multiple Frequency Signals

This paper introduces a new time-resolved spectral analysis method based on the Linear Prediction Coding (LPC) method that is particularly suited to the study of the dynamics of low Signal-to-noise Ratio (SNR) signals comprising multiple frequency components. One of the challenges of the time-resolved spectral method is that they are limited by the Heisenberg-Gabor uncertainty principle. Consequently, there is a trade-off between the temporal and spectral resolution. Most of the previous studies are time-averaged methods. The proposed method is a parameterisation method which can directly extract the dominant formants. The method is based on a z-plane analysis of the poles of the LPC filter which allows us to identify and to accurately estimate the frequency of the dominant spectral features. We demonstrate how this method can be used to track the temporal variations of the various frequency components in a noisy signal. In particular, the standard LPC method, new proposed LPC method and the Short-time Fourier Transform (STFT) are compared using a noisy Frequency Modulation (FM) signal as a test signal. We show that the proposed method provides the best performance in tracking the frequency changes in real time.