With the accelerated rollout of 5G networks comes a dramatic explosion of data rates for numerous applications, accompanied by explosive energy consumption in User Equipment (UE). This paper proposes AI-DRX: an artificial intelligence (AI) powered Dynamic Discontinuous Reception (DRX) mechanism for energy saving in 5G networks. We present a data-driven approach that utilizes Convolutional Neural Networks (CNNs) specifically optimized for real-time operation on mobile devices to model the complex relationship between DRX parameters, network conditions, and user behavior patterns. The proposed AI-DRX mechanism optimizes DRX parameters according to real-time conditions, enabling significant energy savings with minimal QoS degradation. Using comprehensive evaluation with real wireless traffic traces from the MONROE dataset, the AI-DRX mechanism demonstrates energy savings of 69.2% and 55.8% compared to traditional LTE-DRX across different network scenarios, and 70% improvement over Poisson packet arrival models. The CNN model achieves 91.8% prediction accuracy with inference latency of only 2.3ms on modern mobile hardware, making real-time deployment feasible. These results emphasize the promising impact of AI-DRX on dramatically increasing energy efficiency in 5G-powered devices, marking it as an integral component for ensuring sustainable performance of next-generation mobile networks.

Fifth-generation (5G) networks represent a paradigm shift in telecommunications, offering ultra-reliable low-latency communication, massive connectivity of devices, and unparalleled data rates. While these advantages also present significant complications surrounding energy consumption and cybersecurity, requiring new approaches to maintain operational effectiveness and network fidelity. This study proposes a new hybrid artificial intelligence (AI) framework consisting of explainable AI (XAI) for transparent resource allocation, convolutional neural networks (CNNs) for real-time anomaly detection, and recurrent neural networks (RNNs) for predictive energy optimization. Experiments and real-world case studies illustrate this framework’s scalability and efficiency by achieving improved network resource management, a detection accuracy of 99.7% for anomalies, and energy savings of up to 65%.