Cooperatively Learning Footprints of Multiple Incumbent Transmitters by Using Cognitive Radio Networks

Abstract

Energy measurements have conventionally been used for detecting the presence of signal energy but not for distinguishing incumbent users (IUs). In this paper, soft and hard reports Gaussian mixture model learning algorithms are proposed to distinguish intermittently transmitting IUs and to find the footprints for each IU, i.e., the cognitive radio (CRs) that receive signals from each IU. Unlike existing methods for distinguishing IUs using single antenna CRs, the proposed methods do not require CR locations, channel models, or prior knowledge of the number of IUs or their protocol. The soft reports algorithm uses Mahalanobis distance to separate components while a two stage process learns components corresponding to individual IUs. The hard reports algorithm reduces computational complexity by learning unidimensional mixtures at each CR and fusing results using a novel algorithm for finding the maximum weight dominating set in a directed graph. MATLAB simulations of slotted ALOHA IU networks are used to evaluate the algorithms' performance as the number of CRs, IUs, and average activity are varied. In frequent collision scenarios, the soft reports algorithm has the best performance. However, NS3 simulations of 802.11n IU networks show that methods proposed to reduce false positives deteriorate detection performance due to channel capture effects.