Machine Learning-based Jamming Detection in Wireless IoT Networks
Bikalpa Upadhyaya, Sumei Sun, Biplab Sikdar · 2019
Jamming is a well known threat to the wireless community which is becoming a crucial issue with the rise of security-critical applications. The design of a jamming detection mechanism is paramount to current networks to counteract smart jamming attacks, where the attacker stealthily jams the network at different time intervals to cause a significant performance degradation in wireless networks. Current detection mechanisms generally employ the nodes in the network to collect information about the network, hence incurring overhead and communication cost on the nodes. In this paper, we propose to address this issue by using machine learning algorithms to develop a passive, non node-centric, low-overhead network jamming detection mechanism. A number of dedicated nodes, referred as anchor nodes are deployed to collect the network information for jamming detection. We first consider the radio signal strength information for jamming detection and validate with simulated and real network data. With 5 anchor nodes, the proposed detection algorithms report accuracy of 98% and 89.7% respectively for simulated and real network data. We then propose to use the multi-path profile information for jamming detection. With more features, the accuracy is improved significantly over the signal strength-based detection, with 99.01% achieved with simulated data with 5 anchor nodes. Future work will include testing and enhancing the performance of multipath profile-based detection using real network data.