Respiratory frequency estimation from heart rate variability signals in non-stationary conditions based on the Wigner-Ville distribution

A method for respiratory frequency estimation from the high frequency (HF) component of heart rate variability (HRV) by means of the smoothed pseudo Wigner-Ville (SPWVD) distribution is presented. The method is based on maxima SPWVD detection with time-varying frequency smoothing window length, which reduces the estimation error, specially when the respiratory frequency is a nonlinear function of time. Evaluation is performed over HRV simulated signals with time-varying amplitude, nonlinear HF frequency, and 20dB SNR, obtaining a mean frequency estimation error of 0.22±2.04% (0.10±5.96 mHz). The method has been tested on a database of ECG and respiratory signals simultaneously recorded during the listening of different musical stimuli, obtaining a median respiratory frequency estimation error of 0.02±1.90% (0.00±0.98 mHz) during musical stimuli and of 1.98±7.21% (35.41±33.20 mHz) during transitions between stimuli.