Online Learning for Unreliable Passive Monitoring in Multi-Channel Wireless Networks

Passive network monitoring is important for the critical applications of network diagnosis and criminal investigation. As in multi-channel wireless networks, the sniffer-channel assignment problem faces a tradeoff between exploitation and exploration. In this paper, we investigate this problem in a practical scenario. Different from the existing literature, we assume that the knowledge of the users' activities is not known a priori, and there exists capture uncertainty due to unreliable monitoring conditions. Furthermore, we consider the case of sniffer redundancy deployment, which enables multiple sniffers to monitor one channel to enhance capture reliability. Our problem is then formulated as a combinatorial multi-arm bandit problem. We propose an online learning policy, in which sniffer-channel assignment is dynamically decided based on the learning results of the users' activities. We further develop a greedy algorithm to achieve the channel assignment decision in polynomial time. Our solution is evaluated by both theoretical analysis and numerical simulations. Simulation results show that our policy achieves logarithmic regret in time and outperforms the learning policy without consideration of sniffer redundancy deployment.