Conclusions

Lomb and FT spectra are derived using $O(N \log N)$ algorithms, and AR spectra are derived using an algorithm that is only slightly slower for reasonable choices of model order. The essential similarity of the Lomb, FT, and AR spectra given ideal inputs, considered in light of their similar computational demands, suggests that there may be little reason to choose one over any other. When considering the less-than-ideal inputs endemic to HRV studies, however, only the Lomb method produces robust PSD estimates in the presence of noise and ectopy. The Lomb method avoids all of the complications and pitfalls of resampling and replacement of outliers, and introduces no drawbacks of its own; in consequence, it is the method of choice for PSD estimation of heart rate.