QaSAL: QoS-aware State-Augmented Learnable Algorithms for Coexistence of 5G NR-U/Wi-Fi

TL;DR

The paper tackles QoS-aware coexistence of NR-U and Wi-Fi in unlicensed spectrum by formulating the problem as a constrained MDP and introducing QaSAL, a state-augmented learnable framework that embeds dual variables into the environment to enforce QoS constraints during learning. It develops a Lagrangian-based CPM approach and demonstrates how state augmentation enables dynamic, constraint-aware policy adaptation without extensive manual tuning. Applying QaSAL to NR-U/Wi-Fi coexistence with two priority classes, the authors show improved high-priority delay guarantees and fairness (Jain's index) across varying traffic loads, using a SimPy-based MAC simulation. These results highlight QaSAL’s potential for robust, adaptive spectrum sharing in next-generation wireless systems.

Abstract

With the increasing demand for wireless connectivity, ensuring the efficient coexistence of multiple radio access technologies in shared unlicensed spectrum has become an important issue. This paper focuses on optimizing Medium Access Control (MAC) parameters to enhance the coexistence of 5G New Radio in Unlicensed Spectrum (NR-U) and Wi-Fi networks operating in unlicensed spectrum with multiple priority classes of traffic that may have varying quality-of-service (QoS) requirements. In this context, we tackle the coexistence parameter management problem by introducing a QoS-aware State-Augmented Learnable (QaSAL) framework, designed to improve network performance under various traffic conditions. Our approach augments the state representation with constraint information, enabling dynamic policy adjustments to enforce QoS requirements effectively. Simulation results validate the effectiveness of QaSAL in managing NR-U and Wi-Fi coexistence, demonstrating improved channel access fairness while satisfying a latency constraint for high-priority traffic.

Figures (8)

• Figure 1: Parameterized optimization of CPM policy.
• Figure 2: The flowchart of QaSAL algorithm for CPM: a) Training phase b) Execution phase.
• Figure 3: Training evolution of transmission delay of gNB PC1 transmitter ($D_{\text{th,PC1}}=2~\text{ms}$).
• Figure 4: Execution evolution of transmission delay of gNB PC1 transmitter ($D_{\text{th,PC1}}=2~\text{ms}$).
• Figure 5: Dual variable evolution for delay constraint of gNB PC1 transmitter ($D_{\text{th,PC1}}=2~\text{ms}$).