Consensus-based decentralized clustering for cooperative spectrum sensing in cognitive radio networks

A large number of previous works have demonstrated that cooperative spectrum sensing (CSS) among multiple users can greatly improve detection performance. However, when the number of secondary users (SUs; i.e., spectrum sensors) is large, the sensing overheads (e.g., time and energy consumption) will likely be intolerable if all SUs participate in CSS. In this paper, we proposed a fully decentralized CSS scheme based on recent advances in consensus theory and unsupervised learning technology. Relying only on iteratively information exchanges among one-hop neighbors, the SUs with potentially best detection performance form a cluster in an ad hoc manner. These SUs take charge of CSS according to an average consensus protocol and other SUs outside the cluster simply overhear the sensing outcomes. For comparison, we also provide a decentralized implementation of the existing centralized optimal soft combination (OSC) scheme. Numerical results show that the proposed scheme has detection performance comparable to that of the OSC scheme and outperforms the equal gain combination scheme and location-awareness scheme. Meanwhile, compared with the OSC scheme, the proposed scheme significantly reduces the sensing overheads and does not require a priori knowledge of the local received signal-to-noise ratio at each SU.