Throughput Optimized Non-Contiguous Wideband Spectrum Sensing via Online Learning and Sub-Nyquist Sampling

In this letter, we consider non-contiguous wideband spectrum sensing (WSS) using the sub-Nyquist sampling approach. Compared to contiguous WSS which senses the entire spectrum, non-contiguous WSS has an additional task of determining the number and location of frequency bands for digitization and sensing. Since throughput (i.e., the number of sensed vacant bands) increases while the probability of successful sensing decreases with a decrease in the sparsity of digitized bands, we develop exploration-exploitation-based online learning algorithm to learn the spectrum statistics. We provide a lower bound on the number of time slots required to learn spectrum statistics after which the proposed algorithm intelligently selects a maximum possible number of frequency bands which are more likely to be vacant and hence, it is named as throughput optimized non-contiguous WSS. Simulation and experimental results using USRP testbed validate the efficacy of the proposed approach compared to the Myopic approach which has prior knowledge of spectrum statistics.