ML-driven Radio Resource Management in Wireless Local Area Networks

点击翻译按钮获取中文摘要

Wireless local area networks (WLANs) empowered by IEEE 802.11 (Wi-Fi) hold a dominant position in providing Internet access due to freedom of deployment and configuration (thanks to operating in unlicensed bands) and affordable and highly interoperable devices. However, the unplanned deployment and distributed management of WLANs operating in shared unlicensed spectrum is becoming an emerging challenge with the densification of these networks. Another challenge is related to technical innovations, which are making the next-generation of this technology exceedingly complex. Specifically, each new mechanism, designed to improve network performance, comes with a plethora of parameters which have to be properly configured to achieve the best results (and this configuration is left out of the standard). In most cases, multiple parameters have to be tuned together, which is a non-trivial task as the dependencies between parameters and their joint optimization have a highly non-linear impact on network performance. The level of complexity is further increased in the case of coexisting networks, where diverse parameters have to be set across multiple nodes that serve various applications with different QoS requirements. Indeed, future WLAN generations are anticipated to accommodate not only high throughput but also low latency and high-reliability traffic. In summary, the problem of next-generation WLANs is that traditional radio resource management (RRM) algorithms fail to guarantee a reasonable level of performance across a range of scenarios characterized by unplanned deployments and distributed management under the increasing number of configuration options.All these factors make ML algorithms a perfect fit for modern networking, i.e., it can provide estimated models with tunable accuracy, help in tackling existing problems, and encourage new solutions potentially leading to breakthroughs. However, the application of ML algorithms to solve the problems of modern wireless networks poses certain challenges. For example, it requires the definition of the environment state (i.e., observation space), the action space, as well as the reward function for the RRM problem, which is not an obvious task but has a critical impact on the learning process and network performance. This and other ML-related challenges are being slowly overcome in 5G networks operating in licensed bands. Nevertheless, solutions designed for 5G will not be directly applicable to WLANs due to their characteristic differences. First, 5G uses a centralized management approach with carefully planned deployment. Meanwhile, WLANs use a distributed management approach where the deployment is in most cases unplanned and chaotic. Second, 5G operates in licensed bands, with no outside interference, whereas WLANs operate in shared bands where they interfere with each other as well as with devices of other technologies. The ML4WIFI project will address those issues.