Adaptive spectrum sensing and learning in cognitive radio networks

In this paper, we propose a Primary User (PU) activity detection algorithm for a wideband frequency range which updates spectrum sensing parameters. We assume that the signal of PUs and noise are independent and jointly zero-mean Gaussian processes with unknown variances. We employ a Markov Model (MM) with two states to model the activity of PU which representing the presence and absence of the PU at each subband. By using such a MM, the proposed PU activity detector estimates the probabilities of PU presence in different subbands, recursively, in three steps. Our simulation results show that the proposed algorithm always performs better than the Energy Detector (ED) and despite its simple implementation has slightly better performance than the computationally complex Cyclostationarity Feature Detector (CFD) for practical values of the Signal-to-Noise Ratio (SNR).