A DEEP REINFORCEMENT LEARNING APPROACH TO JOINT CODEBOOK SELECTION AND UE SCHEDULING FOR NR-U/WIGIG COEXISTENCE IN UNLICENSED MMWAVE BANDS

This paper introduces an intelligent method to enhance communication in unlicensed millimetre-wave (mmWave) networks for New Radio Unlicensed (NR-U) and Wireless Gigabit (WiGig) systems. Since both networks share the same frequency band, they often interfere with each other, reducing performance and fairness. The challenge lies in ensuring smooth coexistence without harming the efficiency of either system. NR-U plays a crucial role in 5G networks by meeting the growing demand for faster wireless communication. To tackle this problem, the authors propose a novel method that integrates two essential processes: codebook selection and user equipment (UE) scheduling. Codebook selection optimizes beam patterns for communication, while UE scheduling determines which users access the network and when. These two processes operate at different speeds, making optimization complex. The researchers use Deep Reinforcement Learning (DRL) to solve this challenge dynamically and intelligently. The proposed system, DeepCBU, is based on a Layered Deep Q-Network (L-DQN) framework. It learns from past experiences to make better decisions over time. DeepCBU adjusts dynamically, balancing the need for high data rates while minimizing interference between NR-U and WiGig. Additionally, it ensures fairness among users by distributing network access efficiently. Simulation results show that DeepCBU outperforms traditional methods like DRL-dirLBT, TS-dirLBT, and TS-DRL. It improves data rates for NR-U, reduces WiGig interference, and better satisfies user Quality of Service (QoS) requirements. Unlike conventional approaches, DeepCBU does not require prior network knowledge, making it highly adaptable. In conclusion, DeepCBU is a powerful DRL-based system that enhances NR-U and WiGig coexistence. It optimizes both codebook selection and UE scheduling, ensuring better performance and fairness in future wireless networks.