Multitask Spectrum Sensing in Cognitive Radio Networks via Spatiotemporal Data Mining

Recently, compressive sensing (CS) and spectrum sensing have been two hot topics in the signal processing and cognitive radio network (CRN) fields, respectively. Due to the sampling rate limitation of the analog-to-digital converter in spectrum-sensing circuits, some works have proposed integrating these two techniques to achieve low-overhead spectrum sensing in CRNs. These works aim to minimize spectrum reconstruction errors based on linear regression methods, and l(1)-norm is typically used to make a tradeoff between spectrum sparseness and reconstruction accuracy. However, since the interference range of primary users is limited, multiple clusters in the CRN may not share a common sparse spectrum, and thus, the l(1)-norm may not be appropriate to handle all clusters in CS inversion. Hence, we propose a novel multitask spectrum-sensing method based on spatiotemporal data mining methods. In each cluster, we assume that the spectrum sensing is executed in a synchronized way. The cluster head (CH) manages the operations, and a common sparseness hyperparameter is used to make a consensus decision. Among multiple clusters, synchronized CS sampling is not required in our scheme; instead, the Dirichlet process prior is employed to make an automatic grouping of the spectrum-sensing results among different clusters with a common sparseness hyperparameter shared inside each group. To exploit the time-domain relevance among consecutive CS observations, a hidden Markov model is employed to describe the relationship between the hidden subcarrier states and the consecutive CS observations, and the Viterbi algorithm is used to make an accurate spectrum decision for each secondary user. Simulation results show that our proposed algorithm can successfully exploit the spatiotemporal relationship to achieve higher spectrum-sensing performance in terms of normalized mean square error, probability of correct detection, and probability of false alarm, compared with a few other related works.