Wi-Fi rate adaptation is a challenging problem due to the complex and dynamic nature of wireless channels. Existing solutions often rely on heuristics or machine learning techniques. In this work, we propose a novel rate adaptation algorithm, WiFi-CUTS, which employs a cascaded unimodal Thompson sampling (CUTS) multi-armed bandit (MAB) approach to optimize the transmission rate for IEEE 802.11ac links. We develop and implement WiFi-CUTS in numerical simulations and real-world testbed experiments, demonstrating its effectiveness in achieving higher throughput than other multi-armed bandit (MAB) algorithms and the default rate adaptation algorithm in the Linux kernel, Minstrel-HT. Our shielding box experiment results show that WiFi-CUTS achieves a throughput gain of at least 7%, on average 15% and up to 37% over Minstrel-HT in our testbed experiments for medium and high signal-to-noise ratio (SNR) channels. Our over-the-air experiments show a throughput gain of at least 18.2% over Minstrel-HT. Moreover, the proposed CUTS-based algorithms achieve at least a 56% higher throughput than Minstrel-HT in the initial 0.7 s, due to their faster convergence speed. Both indicate that CUTS-based approaches can effectively address the growing demands of next-generation Wi-Fi networks and provide significant improvements over Minstrel-HT.

The performance of IEEE 802.11-based wireless networks is determined by the multi-parameter resource control of transmit and receive settings. With today's WiFi transceivers, controllable parameters include data rate, power, frame transmission aggregation, and global receiver settings like sensitivity. Despite its importance, research on IEEE 802.11 MAC-layer resource control over the past two decades has primarily relied on simulations.