Optimization-driven Deep Reinforcement Learning for Sniffer Patrolling in Wireless Networks

Passive traffic monitoring can be used for network diagnosis and management in wireless networks by deploying wireless sniffers to monitor abnormal data traffic on different channels and locations. This motivates the spatial sniffer-channel assignment (SSCA) problem, i.e., assigning each wireless sniffer a proper operating channel and location to detect the target signals or data packets. The existing approaches for SSCA problems are usually designed for the scenarios where the behavior features of the target users are known. In this paper, we focus on a cognitive monitoring system without information about the users' activities. The wireless sniffers can be deployed to patrol different locations and meet a desirable detection probability requirement. Considering a dynamic network environment with a huge state space, we propose a novel deep reinforcement learning (DRL) approach to adapt the patrolling route for each wireless sniffer. Moreover, we employ Bayesian optimization to help explore the action space and thus improve the learning efficiency. Via numerical simulations, we show that the Bayesian optimization enhanced DRL approach can improve the detection performance and fast adapt the wireless sniffers' actions according to the environmental changes.