Time-frequency analysis based on the phase-rectified signal averaging method

Phase rectified signal averaging (PRSA) is a technique recently introduced that outperforms the classical Fourier analysis when applied to nonstationary signals corrupted by impulsive noise. Indeed, the PRSA helps enhance quasi-periodic components in nonstationary signals while artifacts, intermittent components and high level noise are canceled. Thus the frequency estimation is improved. In this paper, we introduce a new time-frequency representation (PRSA-TFR) which is obtained by applying the PRSA to sliding windows along the signal. This PRSA-TFR keeps the advantages of the PRSA and displays the time evolution of significant frequency components. The comparison with classical time-frequency representations, such as the spectrogram and the Smoothed Pseudo-Wigner-Ville (SPWV) distribution, illustrates the potential of the PRSA-TFR to better reduce the noise level and to make the interpretation of the time-frequency features of the signal easier. The interest of the PRSA-TFR for real signals is illustrated on an electroencephalogram signal.