Energy-Aware Dynamic TDD Scheduling in 5G RAN Using Deep Reinforcement Learning

The rapid expansion of 5G networks has increased Radio Access Network (RAN) energy consumption, leading to higher costs and environmental concerns. Static Time Division Duplex (TDD) configurations fail to adapt to fluctuating traffic demands, resulting in spectrum and power wastage. This paper proposes an AI-driven D-TDD framework utilizing Deep Reinforcement Learning (DRL) within an O-RAN architecture for real-time, energy-aware slot adaptation. The proposed system automatically adjusts uplink and downlink slot ratios, predicts idle transmission opportunities, and activates O-RU smart sleep modes to conserve power while maintaining Quality of Service (QoS). FlexRIC integration, OpenAirInterface (OAI), and USRP B210 hardware testing were employed to validate the framework. Long-term energy savings of up to 65 kWh were observed in realistic macrocell simulations, particularly under low traffic conditions (0-40% load), without compromising throughput. The results demonstrate that AI-driven dynamic TDD scheduling is scalable and robust for energy-efficient 5G and beyond-6G RAN deployments.