Effect of record length on power spectral estimation of the heart rate variability signal

An empirical investigation was carried out to assess how record length selection might affect the accuracy of estimation of some common parameters on the autoregressive power spectrum of the heart rate variability signal. The parameters analyzed were the power associated with very low frequency (VLFP), with low frequency (LFP), and with high frequency (HFP), and also the ration between LFP and HFP (LFP/HFP). The accuracies of estimates were measured as percentage absolute bias and percentage variation over 14 normal subjects. Four record lengths were considered: 900, 600, 300, and 150 samples. The authors found that the estimates of the spectral parameters are unreliable for records shorter than 300 samples, while only a slight improvement in accuracy was obtained moving from a sample size of 600 to 900.<
>