Non-cooperative state tracking of a cognitive radio network with multiple primary users via multiple hypothesis testing

This paper proposes an algorithm to enable a secondary user in a cognitive radio network to be aware of whether each primary transmitter is active or passive, namely network state estimation. The number of transmitters is unknown apriori. A secondary user needs to perform a multiple-hypothesis testing, which can be achieved via clustering the received observations such that observations sharing the same cluster are declared to have been generated from the same hypothesis. However, most of the clustering algorithms assume that data is available offline as a batch. We have earlier proposed the LOC algorithm, a Large-scale Online hierarchical Clustering algorithm for unsupervised sequential numerical data. Unlike most of clustering algorithms, hierarchical algorithms do not assume the number of clusters to be known apriori. In this paper, the LOC algorithm is applied in the context of multiple-hypothesis testing to enable each secondary user to track the network state. The received samples at a secondary user are fed to a proposed filter then to the LOC algorithm to obtain a hierarchical tree. We study the choice of an adequate cutting level and evaluate the spectrum sensing performance. We report a 93% average probability of correctly declaring a true hypothesis in a scenario where number of true hypotheses in the network is five.