A Unified Framework for Joint Energy and AoI Optimization via Deep Reinforcement Learning for NOMA MEC-based Networks

TL;DR

This paper introduces a deep reinforcement learning-based framework for joint energy and Age of Information optimization in NOMA-enabled MEC networks, outperforming existing methods significantly.

Contribution

It proposes a unified metric for optimizing energy efficiency and AoI simultaneously, integrating NOMA and deep Q-learning for enhanced network performance.

Findings

Achieves up to 64% improvement in the objective function.

Reduces AAoI by up to 51%.

Fast convergence of the proposed reward function.

Abstract

In this paper, we design a novel scheduling and resource allocation algorithm for a smart mobile edge computing (MEC) assisted radio access network. Different from previous energy efficiency (EE) based or the average age of information (AAoI)-based network designs, we propose a unified metric for simultaneously optimizing ESE and AAoI of the network. To further improve the system capacity, non-orthogonal multiple access (NOMA) is proposed as a candidate for multiple access schemes for future cellular networks. Our main aim is to maximize the long-term objective function under AoI, NOMA, and resource capacity constraints using stochastic optimization. To overcome the complexities and unknown dynamics of the network parameters (e.g., wireless channel and interference), we apply the concept of reinforcement learning and implement a deep Q-network (DQN). Simulation results illustrate the effectiveness of the proposed framework and analyze different parameters impact on network performance. Based on the results, our proposed reward function converges fast with negligible loss value. Also, they illustrate our work outperforms the existing state of the art baselines up to 64% in the objective function and 51% in AAoI, which are stated as examples.