Developing Self-Optimising 5G Communication Frameworks Using a Deep Reinforcement Learning Approach

Abstract

This study introduces a self-optimising fifth-generation  (5G) communication architecture that utilises Deep Reinforcement Learning (DRL) to meet the increasing demands of ultra-reliable low-latency communication (URLLC) and extensive device connectivity. A DRL agent built on the PPO algorithm is designed to autonomously orchestrate resource allocation within a virtual environment comprising 30,000 nodes, with the dual objectives of minimising end-to-end latency and enhancing overall network effectiveness. The agent learns to adjust resource assignment in response to varying traffic fluctuations and interference patterns, rendering traditional, static, heuristic strategies obsolete. Simulation experiments reveal a 51% decline in average latency, decreasing from 88.7 ms to 43.6 ms, thus assuring compliance with URLLC strictures. The architecture concurrently produces a 39.9% uplift in throughput and a 49.3% rise in the Quality of Service (QoS) satisfaction rate. A comparative evaluation validates the framework’s dominance over conventional benchmarks, underscoring its viability for expansive, intelligent, and self-segregating 5G and forthcoming 6G networks.