Title: Expectation Maximization Aided Modified Weighted Sequential Energy Detector for Distributed Cooperative Spectrum Sensing

Abstract: Distributed cooperative spectrum sensing usually involves a group of unlicensed secondary users (SUs) collaborating to detect the primary user (PU) in the channel, and thereby opportunistically utilize it without causing interference to the PU. The conventional energy detector (ED) based spectrum sensing ignores the dynamic nature of the PU by using energy statistic only from the present sensing interval for the PU detection. However, for a dynamic PU, previous studies have shown that improved detection capabilities can be achieved by aggregating both present and past energy samples in a test statistic. To this end, a weighted sequential energy detector (WSED) has been proposed, but it is based on aggregating all the collected energy samples over an observation window. For a highly dynamic PU, that involves also combining the outdated samples in the test statistic. In this paper, we propose a modified WSED (mWSED) that uses the primary user states information over the window to aggregate only the highly correlated energy samples in its test statistic. In practice, since the PU states are a priori unknown, we also develop a joint expectation-maximization and Viterbi (EM-Viterbi) algorithm based scheme to iteratively estimate the states by using the energy samples collected over the window. The estimated states are then used in mWSED to compute its test statistics, and the algorithm is referred to here as EM-mWSED. Simulation results are presented to demonstrate the states estimation performance of EM-Viterbi and the PU detection performance of EM-mWSED. The results show that, for both highly dynamic as well as slowly time-varying PU, these algorithms outperform the ED and WSED at PU detection, and their performances improve by either increasing the average number of neighbors per SU in the network, or by increasing the SNR or the number of samples per energy statistic.