Wi-Dia: Data-Driven Wireless Diagnostic Using Context Recognition

Abstract

The recent densification of Wi-Fi networks is exacerbating the effects of well-known pathologies including hidden nodes and flow starvation. This paper provides an automatic diagnostic tool for detecting the source roots of performance impairments by recognizing the wireless operating context. Our tool for Wi-Fi diagnostic, named Wi-Dia, exploits machine learning methods and uses features related to network topology and channel utilization, without impact on regular network operations and working in real-time. Real-time per-link Wi-Fi diagnosis enables recovering actions for context-specific treatments. Wi-Dia classifier recognizes different classes of interference; it is jointly trained using simulated and experimental data, taking advantage of the flexibility of the first and the realistic details of the latter. Wi-Dia has been validated in a large European wireless testbed; it provides the right detection of Wi-Fi pathological conditions in real complex scenarios.