Sampling rates, aliasing, and the analysis of electrophysiological signals

Abstract

Low analog to digital (A-D) sampling rates, including the Nyquist rate, are inadequate for direct display and analysis of many electrophysiological signals (i.e., without inverse Fourier reconstruction). Recent empirical studies have reported some results that do not follow from sampling theory, such as increasing spike frequency with increasing sampling rate, and thus require explanation. This study addresses the effects of A-D sampling rate on the frequency and amplitude of known artificial signals and of electromyograms. A-D sampling rate need not always be a multiple of the upper band limit because electromyograms, and many other electrophysiological signals, do not always contain frequency components near the upper band limit. Rather, sampling rate must be matched to particular signal frequencies. Furthermore, A-D sampling rate has different effects on frequency and amplitude. Higher sampling rates are required for accurate amplitude reproduction than for accurate frequency reproduction. Very high sampling rates significantly bias quantitative results by detecting low-level noise in signals. This bias is exacerbated in taped signals sampled at reduced tape speeds.