Weighted joint LRTs for cooperative spectrum sensing using K-means clustering

In this paper, a blind centralized cooperative spectrum sensing (CSS) with soft decision fusion is considered. The fusion center (FC) constructs the energy vector from the collaborating secondary nodes. The energy feature is assumed to be an efficient measure, as it is widely used and directly correlates with the strength of the received signal. The K-means clustering algorithm is employed to extract descriptive statistical features about the distributions of the absence and presence of the primary user (PU) signal, such as the mean and non-centrality parameter. In the framework of this statistical analysis, the signal-to-noise ratio (SNR) at each node is easily estimated and examined. Equal, selective, and weighted combining techniques are applied to develop three joint likelihood ratio test (JLRT)-based algorithms. These algorithms are implemented in the context of simple hypothesis testing, where the distribution of the data is fully specified. Furthermore, they are justified by the Neyman-Pearson theorem, which constructs the most powerful test for a given significance level. The proposed selective and weighted JLRT approaches are based on the estimated SNRs at each sensor, reflecting their reliability. Several comparison scenarios between the proposed algorithms, K-means, fuzzy c-means (FCM), and ORrule are simulated over Rayleigh fading channel with low average SNR and few samples. The simulation results reveal that the proposed tests outperform other CSS techniques. Additionally, asymptotic theoretical expressions for probability of detection and the probability of false alarm are derived, which show high agreement with the simulated results.