Deep Learning-Guided Jamming for Cross-Technology Wireless Networks: Attack and Defense

Abstract

Wireless networks of different technologies may interfere with each other when they are deployed at proximity. Such cross-technology interference (CTI) has become prevalent with the surge of IoT devices. In this paper, we exploit CTI in coexisting WiFi-Zigbee networks and propose DeepJam, a new stealthy jamming strategy, to jam Zigbee traffic. DeepJam relies on deep learning techniques to capture the temporal pattern of the past wireless traffic and predict the future wireless traffic. By only jamming the victim’s transmissions that are not disrupted by CTI, DeepJam can significantly reduce the victim’s throughput with far fewer jamming signals and is thus much more stealthy than conventional jamming strategies. Detailed evaluations show that DeepJam can converge within 10 sec and achieve the jamming-efficiency gains of up to 742% and 285% over conventional random and reactive jamming strategies, respectively, in practical scenarios. We also propose a simple yet effective countermeasure against DeepJam.