Compressive sensing based energy efficient wideband cognitive radio network

Abstract

In this paper, we present a cooperative spectrum sensing technique to identify the spectral location of primary users (PUs) in a wideband spectrum. Here we first divide the wideband into narrow bands of equal bandwidth and deploy arbitrarily located cognitive radio (CR) sensors to sense the narrowbands. The CR sensors operate on conventional energy detection principle and we assume that their spectrum sensing range overlap over each other so as to exploit diversity and mitigate deep fade problem. In order to have energy efficiency, we introduce probabilistic active and sleep state for individual CR sensors. CR sensors in active state compute energy in the specified sensing range and communicate it to a fusion center. Assuming sparse occupancy of PUs in the wideband and by Parseval's theorem, we represent energies of sub-bands in form of sparse vector. Next, we exploit concept of compressive sensing (CS) at fusion center to reconstruct the vector representing energies in the sub-bands. Since individual CR sensors randomly take active or sleep state in a time epoch, the sensing matrix for reconstruction is identified as random matrix. Extending the analysis, we also investigate the value of probability of active/sleep state for which sensing matrix satisfies Restricted Isometry Property (RIP). Finally, we compare the reconstructed energies of sub-band with a specified threshold to make decision on presence or absence of PU in the particular sub-bands. We validate our approach via simulations in which we show performance with (i) variation in probability of sleep state; (ii) variation in number of time epochs or measurements; (iii) varying degree of overlap in spectrum sensing range.