Influence of signal processing on estimation of respiratory impedance

Respiratory impedance was estimated between 4 and 30 Hz by spectral analysis of the mouth flow and pressure signals measured in spontaneously breathing subjects when applying a pseudorandom pressure excitation at the mouth. The signals were submitted to antialiasing low-pass filtering followed by digital preprocessing before the calculation of spectra by a fast Fourier transform algorithm. The effectiveness of signal preprocessing in eliminating the leakage error due to breathing noise was illustrated in both a mechanical analogue and a patient. Five preprocessing techniques that combined high-pass filtering and windowing were then compared in 32 randomly selected patients by examining the influence of these techniques on 1) the values of impedance at 5, 10, and 20 Hz, and 2) the parameters of linear models fitting the real (Zr) and imaginary (Zi) parts of impedance for coherence values higher than a preset threshold. The impedance values and derived parameters were either the mean of the estimates separately obtained in the three data recordings (PA) or the single estimate obtained from average spectra (SP). Small but significant differences between filtering and windowing, as well as between SP and PA, were evidenced for the Zr, whereas Zi was only slightly sensitive to the type of averaging technique. We conclude that the signal preprocessing and data averaging techniques selected in this study have similar effects on spectral estimation of respiratory impedance.