Sotiris Skaperas

Muralikrishnan Srinivasan, Sotiris Skaperas, Arsenia Chorti

This paper presents a systematic approach to use channel state information for authentication and secret key distillation for physical layer security (PLS). We use popular machine learning (ML) methods and signal processing-based approaches to disentangle the large scale fading and be used as a source of uniqueness, from the small scale fading, to be treated as a source of shared entropy secret key generation (SKG). The ML-based approaches are completely unsupervised and hence avoid exhaustive measurement campaigns. We also propose using the Hilbert Schmidt independence criterion (HSIC); our simulation results demonstrate that the extracted stochastic part of the channel state information (CSI) vectors are statistically independent.

Gustavo A. Nunez Segura, Sotiris Skaperas, Arsenia Chorti et al.

Software-defined networking (SDN) is a promising technology to overcome many challenges in wireless sensor networks (WSN), particularly with respect to flexibility and reuse. Conversely, the centralization and the planes' separation turn SDNs vulnerable to new security threats in the general context of distributed denial of service (DDoS) attacks. State-of-the-art approaches to identify DDoS do not always take into consideration restrictions in typical WSNs e.g., computational complexity and power constraints, while further performance improvement is always a target. The objective of this work is to propose a lightweight but very efficient DDoS attack detection approach using change point analysis. Our approach has a high detection rate and linear complexity, so that it is suitable for WSNs. We demonstrate the performance of our detector in software-defined WSNs of 36 and 100 nodes with varying attack intensity (the number of attackers ranges from 5% to 20% of nodes). We use change point detectors to monitor anomalies in two metrics: the data packets delivery rate and the control packets overhead. Our results show that with increasing intensity of attack, our approach can achieve a detection rate close to100% and that the type of attack can also be inferred.