Optimal Lomb transform variant in HRV assessment

Abstract

This study used to compare the performance of Lomb, fast Lomb and averaged Lomb on PSD estimates to examine the systematic differences for: 1 frequency and amplitudes distortions 2 size of RR-interval series 3 computational time. The PSD estimates realised using non-uniformly sampled RR-interval series are found to be within an error range of 1%, 3% and 40% for averaged Lomb, Lomb and fast Lomb transforms respectively, for sine wave model, while 30%, 60% and 80% when simulations were carried using IPFM model, with respect to FFT-based estimates of uniformly sampled data. The computational time for PSD using averaged Lomb transform is quite higher (over ten times in case of sine wave model and 100 times in case of IPFM model) to that of fast Lomb transform and lies in approximately same range, but slightly higher than that for Lomb transform. © 2010 Inderscience Enterprises Ltd.