Identifying statistical mimicry attacks in distributed spectrum sensing

In this work, we consider the spectrum sensing problem in cognitive radio applications where a fusion center collects reports from secondary users (SUs) and fuses them to estimate spectrum occupancy. Some SUs may be malicious and provide false reports. In particular, instead of sensing the spectrum, a malicious SU may use another SU´s report in order to reduce their power consumption, or hide their identity and location. We prove that when the identity of mimic SUs is known, the sufficient test statistic for the optimal fusion rule ignores the mimics´ reports. We show that the joint distribution of the SU reports can be represented by a graphical model. Based on the structural properties of this graphical model, we design an algorithm to learn its structure and thus identify the mimic SUs in the system. We derive an approximation to the probability of misclassification for our proposed algorithm. Simulation results are provided for evaluating performance.