Heart rate variability (HRV) have been used to show the variations of the autonomic nervous system (ANS) activity. In practice, short-time Fourier transform (STFT) and wavelet transform (WT) are used as time and frequency technique to estimate changes in the power spectra of HRV. As a problem, however, it is that these methods are limited that they use fixed basis function which required a prior knowledge about the data. On the other hand, independent component analysis (ICA) learns basis functions based on the statistical characteristics of the data. In this paper, we examine whether HRV basis functions trained by ICA are useful or not to evaluate the ANS variations. In the psychological experiment, we applied learned basis functions to analyze HRV while watching a video by 2D and 3D display. We obtained the results that suggest basis functions learned by ICA are an effective way to represent changes in LF/HF of HRV power spectra comparing to STFT.