Accuracy of Determining the Autocorrelation Function by the Polarity Correlation Function.

The measurement of the normalized autocorrelation function of a continuous stationary random signal such as Gaussian and speech signals may often be replaced by a measurement of the polarity-coincidence function. The accuracy of this simple indirect method is studied in comparison with the direct method of continuous time averaging as well as the method of averaging successive samples. The variances of the estimates of the autocorrelation function obtained by these three methods are calculated as functions of the finite number of samples as well as of the finite integration time. With Gaussian signals the linear correlation between the samples is exactly taken into account. The simplicity of the method and of the required experimental setup together with the reduced rate of flow of information for a given level of precision, establish in all cases a clear advantage of the polarity-coincidence method over the classical methods.