ABSTRACT
Cognitive radio (CR) is an enabling technology in the forthcoming generations of wireless networks. A major challenge in CR networks is the efficient utilization of invaluable radio spectrum. This chapter addresses the issue of wideband spectrum sensing (WSS) in CRs using machine learning techniques. It begins with a detailed discussion of classical spectrum-sensing techniques. These techniques cannot be applied directly for WSS efficiently. Hence, the wideband is first divided into equal, nonoverlapping, narrow sub-bands and then for improved detection, the detector is optimized for each narrow sub-band with the help of some optimization technique. In the WSS context, problem formulation of two frameworks, namely multiband joint detection (MJD) and multiband sensing-time-adaptive joint detection (MSJD), is considered. Machine learning techniques are applied for optimization purposes that can solve convex regions as well as non-convex regions (a feature missing in other implemented optimization techniques). Genetic algorithms (GA) have been previously proposed as an optimization technique for MJD, but after simulating the problem formulation of MJD with GA and particle swarm optimization (PSO), it becomes evident that PSO is a better technique than GA. This chapter proposes PSO as a more suitable optimization technique for CR because it takes less time for execution than GA. Since a CR has to perform tasks in a real-time scenario, therefore the minimum the time it takes the better it is. MSJD is also implemented with PSO, and the simulation results show that PSO is the ultimate choice.
