A computationally efficient algorithm for online spectral analysis of beat-to-beat signals

Abstract

A new algorithm is proposed for the online Fourier analysis of unevenly sampled data. The method is based on the theoretical evaluation of the Fourier Transform of a function linearly interpolating the data, and does not require actual interpolation and re-sampling. The method is particularly suitable for the running evaluation of power spectra. In fact, when a new sample is available, the spectrum can be updated simply by performing calculations on the last sample, without the need to calculate the Fourier Transform again over the whole data record. Applications with simulated and real data show the capability of the algorithm to efficiently estimate the Fourier transform of unevenly sampled cardiovascular data, beat after beat.