An AI-based Approach to 5G Networks' Dynamic Resource Allocation
Abstract
The paper refers to and compares dynamic plans of resource distribution in 5G networks under the patronage of artificial intelligence algorithms. A scheme of simulation in the Python programming environment based on the reinforcement learning schemes (DQN, PPO) and load forecasting schemes (LSTM) is put forward considering low-/peak-load and heterogenous users mobility. It is demonstrated that DQN outperforms always the fixed and proportionally fair schemes: reduces the average latency (in URLLC - to standard levels of <2 ms), increases QoE (including in peak load case - >4.0 on a five-point scale), enhances throughput and energy efficiency. The ensuing data bear witness to the viability of AI deployment for self-driving network evolution and open doors towards 6G.