Spectrum Sensing of Correlated Subbands With Colored Noise in Cognitive Radios

In this paper, we consider the problem of wideband spectrum sensing by using the correlation among the observation samples in different subbands. The Primary User (PU) signal samples in occupied subbands are assumed to be zero-mean correlated Gaussian random variables and additive noise is modeled as colored zero-mean Gaussian random variables independent of the PU signal. It is also assumed that there is at least a minimum given number of subbands that are vacant of PU signals. First we derive the optimal detector and the Generalized Likelihood Ratio (GLR) detector for the case that the covariance matrix of PUs signal samples is unknown and the noise variance in the different subbands is known. Then, we propose an iterative algorithm for GLR test when both the covariance matrix of the PUs signal samples and the noise variances in the different subbands, are unknown. For analytical performance evaluation, we derive some closed-form expressions for detection and false alarm probabilities of the proposed detectors in low Signal to Noise Ratio (SNR) regime. The simulation results are further presented to compare the performance of the proposed detectors.