Effect of FFT based algorithms on estimation of myoelectric signal spectral parameters

The effects of epoch duration, time windowing, and an algorithm upon the estimates of spectral parameters of surface myoelectric signals are discussed. Simulation with computer-generated signals shows that: (a) rectangular window and linear interpolation between contiguous spectral values are acceptable for epoch durations as short as 0.25 s; (b) for the raw periodogram and epoch durations of 1-2 s, coefficients of variation near 3-4% for f_mean and 4-6% for f _med should be expected; and (c) for epochs of 0.25 s these values double, and for shorter epochs the estimates become meaningless because of high variance. These conclusions should be taken into account when selecting windows and epoch durations for myoelectric signal analysis