Abstract

Heart rate variability (HRV) is the temporal variation of the interval between consecutive heartbeats. It can be analyzed by numerousmethods, including spectral analysis. However, HRV time series naturally consist of unevenly spaced data. Several methods emerged to counteract this problem, usually yielding different results. Therefore, in this work three spectral analysis methods were investigated: the Welch, Lomb-Scargle, and Burg method. Their properties were analyzed by theoretical considerations and verified in simulations. Using an oscillator networkmodel with the integral pulse frequency modulation model, artificial HRV time series were generated. Their power spectral densities and their dominant frequencies were evaluated and compared to the nominal values of the simulated time series. The results of these experiments and the theoretical considerations suggest that the Lomb-Scargle method is the most exact in reproducing power spectral densities and dominant frequencies of unevenly spaced HRV data.