Experimental Evaluation of a Wireless Local Area Network (WLAN) Impersonator Detection System Using Deep Learning

Abstract

We present an experimental evaluation of a radio impersonator detection system using deep learning in an indoor environment. RF fingerprinting is used to detect a wireless local area network (WLAN) impersonator that transmits beacon frames to WLAN users with a known medium access control address. The RF impersonator detection was performed by training a neural network with real-time beacon frame data captured from the trusted routers in a multifloor building in a campus environment. In addition to the beacon frames from the trusted routers, the beacon frames were captured from an unknown router at five different locations. The neural network was trained using RF fingerprinting from the trusted routers and the unknown router. The confusion matrices about the classification of the trusted routers and the unknown router are presented. Finally, a software-defined radio (SDR) over-the-air (OTA) test was performed in the same multifloor indoor environment to detect router impersonator. Our SDR-based approach shows that the trained neural network can classify the WLAN router impersonator with high accuracy.