Statistical measures of clipped random signals

It is shown that substantial and yet difficult to detect errors may occur if random data are clipped and then filtered. For stationary random processes formulas are presented for the spectral density (power spectrum) of the clipped and filtered random process in terms of the spectral density of the original random process, the clipping level, and the gain characteristics of the filter. A nonstationary random process that is formed by summing components, each of which is a stationary, band-limited random process modulated by a deterministic function of time, is also analyzed. Estimates of the mean square output (the clipped and filtered input) are given. Examples are given that illustrate the problem of difficult to detect errors for both stationary and nonstationary cases. An overload detector is suggested which could significantly reduce the probability of this error. 5 refs., 13 figs.