WideRate: Reinforcement Learning Rate Adaptation for Mobile Wide Area Networks
Karyn Doke, Elham Sadeghi, Vaasu Taneja, Habib Affinnih, Petko Bogdanov, Mariya Zhivkova Zheleva · 2024
Mobile wireless networks revolutionize our lives and livelihoods. Yet, rural areas, characterized with sparse populations and rugged terrain, consistently lag behind in mobile connectivity compared to their urban counterparts. As a result, community-owned networks realized through fixed wireless tech-nologies, have become an increasingly viable Internet option for otherwise disconnected areas. Fixed wireless, however, is inherently designed for residential/stationary access and is not readily applicable for the use of mobile agents that might travel through a rural community. In this paper we explore the extension of fixed wireless networks for mobile access. A key factor for continuous mobile access is efficient rate adaptation. To that end, we develop WideRate- a reinforcement learning framework that employs signal strength measurements for optimal rate adaptation. We showcase WideRate in the context of wide-area Television White Space networks, whereby we design a vehicular mobile unit and carry out an extensive measurement campaign in a real community network. We use the collected traces to motivate the need for rate adaptation and implement a realistic network simulator that aids in our evaluation. We demonstrate that WideRate significantly outperforms counterparts from the literature including a reinforcement learning model.