Comparison of methods for harmonic wavelet analysis of heart rate variability

Two different methods for the spectral analysis of heart rate variability data are compared: the discrete Fourier transform and the nonequispaced Fourier transform. The methods are used to analyse test signals that are constructed to mimic R-R interval data, with graded levels of noise, generated using the integral pulse frequency modulation model. It is found that the nonequispaced Fourier transform is the better method for determining the frequency coefficients of the test signals as the noise level is increased. Both methods are compared when they are used as the first step in a time-frequency analysis of the test signals using the discrete harmonic wavelet transform, and a quantitative comparison is made on a set of randomly generated test signals (this shows that the nonequispaced Fourier transform is the better method). A further study shows the ability of the discrete harmonic wavelet transform to detect frequencies close to the boundary between wavelet levels. Since each harmonic wavelet represents a distinct frequency band, careful choice of sampling frequency means that such wavelets can be used to identify spectral bands associated with physiological causes. A clinical example, using the nonequispaced Fourier transform, shows how the discrete harmonic wavelet transform could be developed for use in detecting brief alterations in autonomic tone.