The effect of sampling frequency on EMG measures of occupational mechanical exposure

The purpose of the present study was to determine how much it is possible to downsample EMG without loosing the accuracy of some EMG measures of occupational mechanical exposure frequently used in the literature. The EMG signals of four muscles (two trapezius, right deltoid and right extensor digitorum) were collected (sampling frequency: 1024 Hz) from 20 subjects while performing a 20 min computer work task. The EMG RMS amplitude was computed from 0.125 s successive time-windows for the original (1024 Hz) and four additional digitally resampled (at 512, 256, 128 and 64 Hz) EMG signals. Three of the most frequently used data reduction methods (1. gaps analysis, 2. amplitude probability distribution function and 3. exposure variation analysis) were used to summarise the 20 min EMG activation profiles. The results from the exposure variation analysis were summarized into three variables using a new method detailed here. In general, the ANOVA for repeated measures demonstrated that a decrease of the sampling frequency significantly changed the EMG measures (relative to the 1024 Hz sampling condition) most of the time at 64 and 128 Hz, occasionally at 256 Hz, but practically never at 512 Hz. An analysis of the maximal errors (relative to the 1024 Hz condition) across all subjects supported these findings. Consequently, it was concluded that 512 Hz is quite conservative and should practically never lead to invalid EMG parameters estimations. Conversely, 256 Hz represents the lowest limit tolerable for some EMG parameters (gaps analysis, amplitude probability distribution function) while it is unacceptable for others (modified exposure variation analysis).