Energy detection of unknown deterministic signals

TLDR

Prior studies by Marcum and Kaplan on energy detection are reviewed. The detection problem is modeled as a chi‑square test on the energy of a sampled signal, with the statistic following a central chi‑square under noise‑only and a noncentral chi‑square under signal presence; an approximate modified chi‑square is used to facilitate calculation. ROC curves for various time‑bandwidth products and a nomogram for chi‑square probabilities are provided to quickly compute false‑alarm and detection rates.

Abstract

By using Shannon's sampling formula, the problem of the detection of a deterministic signal in white Gaussian noise, by means of an energy-measuring device, reduces to the consideration of the sum of the squares of statistically independent Gaussian variates. When the signal is absent, the decision statistic has a central chi-square distribution with the number of degrees of freedom equal to twice the time-bandwidth product of the input. When the signal is present, the decision statistic has a noncentral chi-square distribution with the same number of degrees of freedom and a noncentrality parameter λ equal to the ratio of signal energy to two-sided noise spectral density. Since the noncentral chi-square distribution has not been tabulated extensively enough for our purpose, an approximate form was used. This form replaces the noncentral chi-square with a modified chi-square whose degrees of freedom and threshold are determined by the noncentrality parameter and the previous degrees of freedom. Sets of receiver operating characteristic (ROC) curves are drawn for several time-bandwidth products, as well as an extended nomogram of the chi-square cumulative probability which can be used for rapid calculation of false alarm and detection probabilities. Related work in energy detection by J. I. Marcum and E. L Kaplan is discussed.