An Intelligent Q-Learning Approach for Energy-Efficient Channel Occupancy in NR-U Cellular Networks for Fair Unlicensed Spectrum Access

The growing demand for mobile data strains the licensed spectrum, prompting the need for efficient offloading strategies. New Radio in Unlicensed Spectrum (NR-U) leverages 5 GHz and 6 GHz bands but faces coexistence challenges with WiFi and Bluetooth due to interference. This paper proposes three reinforcement learning-based models to optimize Channel Occupancy Time (COT) in NR-U networks. The first uses Qlearning for fair coexistence with WiFi under Listen-BeforeTalk (LBT) rules. The second enhances energy efficiency by jointly optimizing COT and power control. The third applies Dyna-Q+ to accelerate learning and adapt to dynamic traffic loads. Simulations over 500 scenarios show all models achieve Jain’s Fairness Index above 0.9, ensuring equitable spectrum sharing. The energy-efficient model reduces power consumption by up to 20% without compromising performance. These results demonstrate the potential of intelligent learning-based approaches for robust and efficient NR-U and WiFi coexistence in shared spectrum environments.