On windowing effects in estimating averaged periodograms of noisy signals

Abstract

Due to the unequal weighting of samples of measured signals, there is considerable uncertainty in the variance estimate based on a spectral density estimate computed by the method of averaged periodograms. Especially, if the signals contain non-stationary components randomly distributed along the data, careful design of a computation procedure is needed. In this paper the effect of windowing on the variance estimate computed based on a power spectral density estimate is analysed experimentally. Four different window functions are studied in order to find out their effect on periodogram-based variance estimates. One of these windows has been newly developed for the use of overlapped-segment averaged periodogram computations. The performance evaluation of the windows is made, and the effect of windows in the function of overlapping is studied.