A Machine Learning Based Proactive Security Framework for Wireless Networks

Physical layer security is one of the most crucial aspect of security in wireless networks. Due to continued sharing of resources, wireless networks often come under security attacks, most common of which are eavesdropping attacks. In the case of eavesdropping attacks, deliberately designed random eavesdropping data is added to the channel. These eavesdropping along with noise result in packet losses and low throughput, degrading the overall performance of the cognitive network. In this work, a security aware eavesdropping rejection mechanism is proposed which detects suspicious signals in the channel frequency response and employs discrete equalization to recover transmitted data. This paper presents a machine learning assisted security aware channel assignment protocol against possible adversarial eavesdropping attacks. The ML parameters such as gradient, iterations to convergence and cost function have been computed and presented. The final error rate with and without the proposed system under adversarial attacks is also presented. It can be observed that the proposed approach is close to the no adversarial attack condition clearly indicating the efficacy of the approach to proactively thwart potential attacks. The error rates are also significantly lower than existing approach in the domain.