Search

Despite Wireless Sensor Networks (WSNs) significantly developing over the past decade, these networks, like most wireless networks, remain susceptible to malicious interference and spectrum coexistence. Other vulnerabilities arise as WSN applications adopt open standards and typically resource and energy-constrained commercial-off-the-shelf equipment. Deployments include safety-critical applications such as the internet of things, medical, aerospace and space and deep-sea exploration. To manage safety and privacy requirements across such a diverse wireless landscape, security on wireless edge devices needs improvement while maintaining low complexity. This paper improves wireless edge device security by developing a novel intelligent interference diagnostic framework. Received in-phase (I) and quadrature-phase (Q) samples are exclusively utilized to detect modern, subtle and traditional crude jamming attacks. This I/Q sample utilization inherently enables decentralized decision-making, where the low-order features were extracted in a previous study focused on classifying typical 2.4–2.5 GHz wireless signals. The associated optimal intelligent models are leveraged as the foundation for this paper's work. Initially, Matlab Monte Carlo simulations investigate the ideal case, which incorporates no hardware limitations, identifies the required data type of signal interactions and motivates a hardware investigation. Software-defined radios (SDRs) collect the required live over-the-air I/Q data and transmit matched signal (ZigBee) and continuous-wave interference in developed ZigBee wireless testbeds. Low complexity supervised machine learning models are developed based exclusively on the low-order features and achieve an average accuracy among the developed models above 98%. The designed methodology involves examining ZigBee over-the-air data for artificial jamming and SDR jamming of ZigBee signals transmitted from SDR and commercial (XBee) sources. This approach expands to a legitimate node classification technique and an overall algorithm for wireless edge device interference diagnostic tools. The investigation includes developing Support Vector Machine, XGBoost and Deep Neural Network (DNN) models, where XGBoost is optimal. Adapting the optimized models to global positioning system signals establishes the transferability of the designed methodology. Implementing the designed approaches on a Raspberry Pi embedded device examines a relatively resource-constrained deployment. The primary contribution is the real experimentally validated interference diagnostic framework that enables independent device operation, as no channel assumptions, network-level information or spectral images are required. Developed models exclusively use I/Q data low-order features and achieve high accuracy and generalization to unseen data. [ABSTRACT FROM AUTHOR]