An online learning approach for auto link-Configuration in IEEE 802.11ac wireless networks

Abstract

High throughput wireless standards based on IEEE 802.11, such as IEEE 802.11ac, pose a significant challenge in selection of link configuration parameters in an automatic approach. These high throughput wireless standards have a large pool of link configuration parameters at the both physical (PHY) and media access control (MAC) layers, which include channel bonding, multiple input multiple output (MIMO) technology, short guard interval, advanced modulation and coding schemes (MCS), frame aggregation, block acknowledgement (BA) etc. In time-varying wireless channel, each wireless station must tune such multiple link configuration parameters dynamically and adaptively according to the channel and network conditions, to achieve high throughput in practical scenarios. This dynamic adjustment of link configuration parameters is known as auto link-configuration. In this paper, we design an online learning-based mechanism, BanditLink, for auto link-configuration in high throughput wireless networks. To tackle auto link-configuration based on both network load and channel conditions, BanditLink implements multi-armed bandit based adaptive learning methodology along with fuzzy logic. We analyze the performance of BanditLink from both simulation and testbed results, and observe that it can improve the network performance compared to other link adaptation methodologies.