Rule-Based QoS Governors versus Built-in Constrained Reinforcement Learning for Wireless Resource Management: A Review and Design Perspective

Abstract: Deep reinforcement learning (DRL) has become a promising approach for wireless resource management because of its ability to adapt scheduling, power control, and allocation decisions under dynamic network conditions. However, many DRL-based wireless solutions optimize performance through reward shaping alone, which can make Quality-of-Service (QoS) constraints difficult to interpret, tune, or guarantee in practical deployments. This paper presents a review and design perspective on constraint-handling mechanisms for learning-based wireless resource management, with a particular focus on the comparison between built-in constrained reinforcement learning methods and external rule-based QoS governors. We discuss four major approaches: reward-penalty design, constrained Markov decision process formulations, safety-layer or shielding methods, and runtime QoS governor mechanisms. The analysis highlights that while constrained RL methods provide a principled mathematical framework, they may introduce additional training complexity and sensitivity to hyperparameter selection. In contrast, rule-based QoS governors offer a practical and interpretable way to monitor service degradation, enforce conservative recovery actions, and support deployment around legacy schedulers such as proportional fair scheduling. Based on this comparison, the paper argues that external QoS governors can serve as a useful middle layer between fully heuristic scheduling and fully autonomous learning-based control. Finally, we identify open challenges related to multi-cell coordination, real-time inference, traffic generalization, O-RAN integration, and standardized benchmarking for safe DRL-based wireless systems. Keywords: Deep reinforcement learning, wireless resource management, QoS constraints, safe reinforcement learning, proportional fair scheduling, runtime governor, O-RAN. Title: Rule-Based QoS Governors versus Built-in Constrained Reinforcement Learning for Wireless Resource Management: A Review and Design Perspective Author: Aouari Abdelhamid, Yuting Li International Journal of Recent Research in Mathematics Computer Science and Information Technology ISSN 2350-1022 Vol. 13, Issue 1, April 2026 - September 2026 Page No: 18-24 Paper Publications Website: www.paperpublications.org Published Date: 04-May-2026 DOI: https://doi.org/10.5281/zenodo.20024605 Paper Download link (Source) https://www.paperpublications.org/upload/book/Rule-Based%20QoS%20Governors%20versus%20Built-in-04052026-4.pdf