True random number generation via sampling from band-limited Gaussian processes

Abstract

A true random number generator topology based on regular sampling of an "irregular" process is considered, which is obtained via thresholding a continuoustime Gaussian (normal) process, of which spectrum is assumed to be at between two known frequencies and zero everywhere else. Per-sample joint entropy of the resulting bit sequence is introduced as the main gure of merit. Employing an approach based on statistical signal processing and information theory, novel analytical results on the optimum choice of the sampling period is presented that ensure maximal randomness of the resulting bit sequence together with asymptotic analysis and numerical experiments. In addition, new results that fully characterize the autocorrelation behavior (equivalently spectral properties) of the resulting bit sequence is presented and a related metric, termed "spectral correlation" is introduced to quantify the "uncorrelatedness" of the binary bit sequence output.