Energy efficient sleep mode strategies for communication and computing devices in cellular networks with edge computing

Abstract

Over the years, to minimize the energy consumption of ultra-dense networks (UDNs), numerous sleep strategies for base stations (BSs) have been proposed. However, more fine-grained sleep patterns remain underexplored. Additionally, the integration of edge computing into UDNs introduces further complexity to the energy consumption minimizing problem. This paper aims to propose a multi-agent deep reinforcement learning (DRL)-based framework to significantly reduce energy consumption in cellular networks with edge computing. By leveraging the gate recurrent unit (GRU) to predict unobserved local state information for each agent, including channel gain values of outgoing and incoming links, minimum transmission rate requirements for user response messages and task offloading packets, as well as minimum computing power requirements for offloaded tasks, the proposed scheme enables more precise control over the sleep modes of both radio frequency (RF) transceivers of BSs and edge servers. Consequently, this scheme achieves a substantial reduction in cumulative energy consumption. Extensive simulations confirm that the proposed scheme not only reduces communication energy but also decreases the computing energy required by edge servers.