Replicated Q-learning Based Sub-band Selection for Wideband Spectrum Sensing in Cognitive Radios

Spectrum sensing is a key basic function in any wideband cognitive radio (CR) for detecting the presence of any spectral activities. However, due to hardware constraints, the instantaneous sensing bandwidth is limited to a single sub-band out of all sub-bands in the spectrum of interest. Hence, sub-band selection is an important step in wideband spectrum sensing. In this paper we develop a partially observable Markov decision process (POMDP) to model the sub-band dynamics and propose an efficient sub-band selection policy based on replicated Q-learning. It is shown through simulations that the proposed selection policy has reasonably low computational complexity and significantly outperforms the random sub-band selection policy.