Online Heart Rate Variability Evaluation

Type: Semester project
Students: Stefan Scheidegger
Advisors: Martin Kusserow, Johannes Schumm

In this project algorithms for heart rate variability (HRV) were discussed in terms of complexity, on-line capability and frequency and time resolution. Fast Fourier transform (FFT), Lomb-Scargle periodogram (LSP) and continuous time wavelet transform (CWT) were implemented and evaluated using Matlab. Using a sinusoidal testsignal LSP showed results similar to FFT and had low computation times. CWT had higher computation times than FFT and LSP and a lower frequency resolution, the time resolution however was higher. Because of its high time resolution and its ability to handle non-stationary data the CWT was afterwards implemented in the context recognition network (CRN) toolbox. Finally the results of HRV feature computation of the different PSD estimation algorithms in Matlab and by using CWT in Matlab and the CRN toolbox were compared using data acquired in an experiment.

The experiment consisted of 5 phases each lasting 10 minutes: lying supine, sitting, standing, walking (1 km/h) and walking (4 km/h). In the results of the HRV analysis these 5 phases may be distinguished using mean and variance values of the HF features.

It has been shown that the error in results using different methods for PSD estimation are not negligible. It was however shown on experimental data that different activities may be distinguished using these methods. The advantages of the CWT - foremost the ability of handling non-stationary data and a tremendous time resolution compared to FFT and LSP - indicate considerable improvements for using on-line frequency analysis in heart rate signals.