Entropy-based spectrum sensing in cognitive radio

In this paper, we present a simple technique for detection of primary users in cognitive radio networks with unknown noise and interference levels. We will show that the likelihood ratio test for detecting the primary user can be approximated to a formulation that compares the estimated entropy of the received signal to a suitable threshold. This formulation is also intuitive since for a given variance, the entropy of a stochastic signal is maximized if it is Gaussian. If the received signal contains the primary user's digitally modulated component, the entropy is reduced. Although the proposed approach is applicable under any scenario, we will specifically consider matched-filter-based detection in this paper, with its underlying assumption that the cognitive radio knows the primary user signaling waveform. We will consider the case where the Gaussian noise and interference levels in the region are unknown, which renders traditional matched-filtering and energy-based detection approaches unfeasible. The probabilities of successful detection and false alarm are characterized for both classical and Bayesian scenarios.