Averaged multiple L-spectrogram for analysis of noisy nonstationary signals

Abstract

A time–frequency approach for improved instantaneous frequency estimation of noisy signals has been proposed. This approach is based on the averaged multiple time–frequency distribution obtained by using L-spectrograms with different values of the scaling parameter L. The L- spectrogram with low value of L is characterized by large bias and low variance, while higher L decreases the bias and increases the variance of the instantaneous frequency estimator. The averaged L-spectrogram is considered to provide a good bias-variance compromise, leading to low MSE of the instantaneous frequency estimation in the presence of noise. The proposed approach has been tested in the experiments with different values of signal to noise ratio, showing its superior performance when compared to individual L-spectrograms.