Entropy-based Spectrum Sensing under Symmetric Alpha Stable Impulsive Noise

We propose a novel spectrum sensing algorithm based on the differential entropy under symmetric alpha stable impulsive noise, in contrary to the several existing algorithms where the Gaussian noise is assumed. The need for a nonGaussian noise model for spectrum sensing stems from the observation that the tails of probability density function of receiver noise are typically heavier that that of Gaussian. The proposed method employs a goodness-of-fit test based on an estimate of the entropy in the observations, compares this decision statistic to a threshold chosen based on a given level on the probability of false-alarm, and decides in the favor of the presence of the incumbent when this statistic exceeds a threshold. Through Monte Carlo simulations, we study the performance of the proposed method and show that it outperforms the matched filter-based detector and the energy detector in practical scenarios.