Intelligent Routing and Scheduling for the Energy-Delay Dilemma in Future Networks

Abstract

Next-generation networks aiming to support ultra-reliable, low-latency, and high-throughput services often operate in the THz band where high absorption losses necessitate dense gNodeB (gNB) deployment. While this enhances coverage and Quality-of-Service (QoS), it also increases the network energy consumption. A commonly explored solution in this context is the gNB sleep scheduling, which, although effective in conserving energy, introduces additional latency, thereby creating an Energy-Delay Dilemma in time-sensitive, energy-aware communication scenarios. To address this, we propose EnRoute, an intelligent framework for unified routing and scheduling in heterogeneous networks, designed to optimize the route energy efficiency while simultaneously meeting the traffic-specific stringent latency and reliability constraints. We first formulate the Time-Sensitive Energy-Efficient Routing Problem (TSEE-RP), and prove its NP-hardness. Subsequently, EnRoute offers a solution that integrates Deep Q-Network (DQN), $k$ -means clustering, and an adaptive fair scheduling policy to tackle the energy-delay dilemma while preserving the performance of throughput-intensive applications. Our performance analysis shows that EnRoute outperforms state-of-the-art methods with up to 98.4% fewer latency violations, 46.5% higher energy efficiency, and 72.8% lower model energy cost, while ensuring Key Performance Index (KPI) compliance across mixed traffic.