Beam search assisted DRX mechanism in 5G networks: A machine learning and Semi-Markov modeling integration

Abstract

The integration of 5G New Radio into unlicensed spectrum (NR-U) offers potential to expand capacity and enhance user experience. However, the shared environment is highly dynamic due to coexistence with technologies such as Wi-Fi, making conventional static Discontinuous Reception (DRX) mechanisms insufficient. These limitations lead to unnecessary wake-ups, increased delay, and reduced energy efficiency for User Equipment (UE). To address this challenge, this work proposes a beam search assisted adaptive DRX framework that enables reliable directional access under unlicensed operation. The framework integrates Machine Learning (ML) based channel status prediction to determine whether the channel is available or congested by analyzing Wi-Fi traffic patterns including Inter-Frame Space statistics and collision ratios. The DRX mechanism is further modeled using a Semi-Markov model to capture realistic state timing dynamics and non-exponential holding times. Multiple ML classifiers are evaluated on a dataset of 20,000 samples. Simulation results show that the LSTM model achieves 96.8% prediction accuracy and improves Power Saving Factor from 83.2% to 87.8%, throughput from 30.6 Mbps to 60.4 Mbps, and reduces delay from 125.3 ms to 50.9 ms compared with non-predictive DRX operation. These results demonstrate that prediction-driven DRX mechanism significantly improves energy efficiency and latency performance in NR-U deployments.