Optimizing 802.11 wireless communications with machine learning

Wireless communication systems are becoming increasingly complex to cope with demands for better performance. The former, combined with the unpredictable behavior of the wireless channel, contribute to the creation of intractable networks that can hardly be characterized by means of accurate yet scalable analytical models. In this work, we discuss on the suitability of machine learning to perform this task. In particular, we present several learning approaches that address relevant performance issues in the context of prominent WLAN systems. With the goal of achieving high throughput 802.11ac defines very wide channels, which increases the perceived frequency variability of the channel and eventually degrades the communication performance. We develop a lightweight learning-based resource allocation scheme that counteracts and exploits the frequency variability. Vehicular communications require reliable message delivery in the context of safety applications. However, we observe that jamming attacks compromise road safety by impairing the communication of 802.11p devices. Motivated by this finding, we develop a jamming detection tool that learns the behavior of commodity devices, in order to later detect jamming attacks in vehicular scenarios. Rate adaptation provides means for the support of infotainment applications. In vehicular environments this is a challenging task, due to the fast changing channel. We develop a learning algorithm that identifies signal propagation patterns buried in empirical data and selects the rate according to predicted future channel conditions.