Online learning-based optimal primary user emulation attacks in cognitive radio networks.

In a cognitive radio (CR) network, a secondary user learns the spectrum environment and dynamically accesses the channel where the primary user is inactive. At the same time, a primary user emulation (PUE) attacker can send falsified primary user signals and prevent the secondary user from utilizing the available channel. Although there is a large body of work on PUE attack detection and defending strategies, the best attacking strategies that an attacker can apply have not been well studied. In this paper, for the first time, we study the optimal PUE attack strategies without any prior knowledge on the primary user activity characteristics and the secondary user access strategies. We formulate the problem as a non-stochastic online learning problem where the attacker needs to dynamically decide the attacking channel in each time slot based on its attacking experience in previous slots. The challenge in our problem is that the PUE attacker cannot observe the reward on the attacked channel because it never knows if a secondary user ever tries to access it. To solve this challenge, we propose an attack-but-observe- another (ABOA) scheme, in which the attacker attacks one channel in the spectrum sensing phase, but observes at least one other channel in the data transmission phase. We propose two non-stochastic online learning-based attacking algorithms, EXP3-DO and OPT-RO, which select the observing channel deterministically based on the attacking channel and uniform randomly, respectively. EXP3-DO employs an existing theoretical framework and is suboptimal. OPT-RO is based on the new proposed theoretical framework and is optimal. They achieve regret in the order of O (T-2/3) and O (root T), respectively. T is the number of slots the CR network operates. We also generalize OPT-RO to multichannel observation cases. We show consistency between simulation and analytical results under various system parameters.