Cross-Layer Rate Adaptation in IEEE 802.11ax WLANs using Hybrid Feedback and Bayesian Learning Mechanisms

Abstract

The rapid growth in wireless data demand has led to the evolution of IEEE 802.11 standards, with 802.11ax (Wi-Fi 6) aiming to improve spectral efficiency, throughput, and user experience in dense environments. A key aspect of this performance enhancement lies in intelligent rate adaptation (RA), that is, dynamically selecting the optimal Modulation and Coding Scheme (MCS) based on changing channel conditions. Conventional rate adaptation algorithms (RAAs) generally operate at either the MAC or PHY layer, depending solely on implicit feedback (such as ACKs) or explicit feedback mechanisms. These methods often face drawbacks like delayed adaptation, stale channel quality information (CQI), or excessive feedback overhead. In contrast, cross-layer rate adaptation presents a principled framework for handling uncertainties while jointly leveraging PHY and MAC layer observations. High Efficiency (HE) 802.11ax includes the HE sounding protocol to determine the channel quality information (CQI). The HE sounding protocol provides an explicit feedback mechanism where the STA sends an array of per-RU average SNRs in the HE CQI report field. Traditional RAAs, such as Minstrel, SampleRate, and ARF, are predominantly reactive or rely on historical statistics, often struggling to adapt swiftly to dynamic channel conditions. While most explicit feedback RA algorithms suffer from drawbacks such as modifying ACK frames, which violates the standard, introducing additional overhead through special feedback frames, and consuming valuable bandwidth. By utilizing the explicit CQI feedback available in IEEE 802.11ax, we propose a hybrid (cross-layer) rate adaptation approach that integrates MAC-layer information, derived from transmission history at the transmitter, with PHY-layer information from receiver-provided CQI feedback. This combined strategy enables more accurate channel assessment and choosing the appropriate transmission rate. In this thesis, we focus on designing cross-layer rate adaptation techniques without altering standard frames and protocol, ensuring suitability for practical deployment. The solutions provided do not require modifications to standard 802.11ax frames or protocols, making them easily deployable in real-world systems. We have designed and evaluated all the proposed and benchmarked algorithms using a standards-compliant MATLAB WLAN Toolbox, to model an end-to-end link-level SISO transmit-receive link with IEEE standard-defined channel models. All our rate adaptation simulations account for realistic PHY-layer impairments, including carrier frequency offset (CFO) and symbol timing offset. In the first part of our work, we present a Hybrid Channel-Dependent Rate Adaptation (HCDRA) scheme that maps the per-RU average SNR to an MCS value that maximizes the throughput while keeping the Packet Error Rate (PER) below a threshold value. We propose a delayed CQI feedback that offers a more efficient alternative to instantaneous CQI feedback by reducing airtime and protocol overhead. We compare the performance of HCDRA with well-known RAAs – Automatic Rate Fallback (ARF), Adaptive ARF, Minstrel, and MutFed. To minimize the overhead associated with frequent explicit CQI feedback, we adopt a robust Bayesian learning-based approach that allows the transmitter to statistically infer channel quality (per-RU average SNR) from less frequent CQI feedback. An online learning mechanism is integrated within the Bayesian framework to continuously update the parameters of the SNR distribution as new feedback is received. To this end, the second part of our work presents a novel Bayesian learning-based channel feedback framework that utilizes Bayesian updates to refine the probabilistic model of the SNR. Building on this, we propose a Bayesian Learning-based Rate Adaptation (BLbRA) algorithm designed to maintain a target PER. Further, we observed that the original hyperparameter update rule used in BLbRA struggles to adapt to sudden changes in the wireless environment. In practical WLAN scenarios, channel conditions can change, particularly in the presence of user mobility or dynamic surroundings, such as movement between rooms or dense deployments. We aim to design an RA algorithm that can respond swiftly to these changes. This objective signals a need for improvement in the learning mechanism or hyperparameter update strategy. To address this, we maintain the Bayesian learning-based channel feedback framework while revising the objective to maximize instantaneous expected throughput by adopting a different MCS selection strategy. Based on these new objectives, in our third work, we propose Maximum Expected Throughput BLbRA (MET-BLbRA). In MET-BLbRA, the posterior updates are performed online using efficient recursive rules, enabling the algorithm to track time-varying channel conditions without incurring high computational overhead. We have benchmarked the performance of MET-BLbRA with the Multi-armed Bandit (MAB) based and Reinforcement learning (RL) based RA algorithms. To evaluate the robustness and adaptability, we performed experiments under varying environmental conditions. MET-BLbRA demonstrates enhanced throughput performance and effectively tracks and adapts to dynamic changes in the wireless environment. The Bayesian paradigm allows for principled decision-making under uncertainty, facilitates exploration-exploitation, and enables online adaptation to changing channel conditions.