Deep Reinforcement Learning for Wireless Resource Allocation Using\n Buffer State Information

2021/08/27 by Eike-Manuel Bansbach, Bansbach, Eike-Manuel, Victor Eliachevitch +3
Engineering · Computer Science · #Advanced Wireless Network Optimization #Wireless Networks and Protocols #Advanced MIMO Systems Optimization

paper · pdf · doi:10.48550/arxiv.2108.12198

Abstract

As the number of user equipments (UEs) with various data rate and latency\nrequirements increases in wireless networks, the resource allocation problem\nfor orthogonal frequency-division multiple access (OFDMA) becomes challenging.\nIn particular, varying requirements lead to a non-convex optimization problem\nwhen maximizing the systems data rate while preserving fairness between UEs. In\nthis paper, we solve the non-convex optimization problem using deep\nreinforcement learning (DRL). We outline, train and evaluate a DRL agent, which\nperforms the task of media access control scheduling for a downlink OFDMA\nscenario. To kickstart training of our agent, we introduce mimicking learning.\nFor improvement of scheduling performance, full buffer state information at the\nbase station (e.g. packet age, packet size) is taken into account. Techniques\nlike input feature compression, packet shuffling and age capping further\nimprove the performance of the agent. We train and evaluate our agents using\nNokia's wireless suite and evaluate against different benchmark agents. We show\nthat our agents clearly outperform the benchmark agents.\n

Related

• Deep Reinforcement Learning for Dynamic Resource Allocation in Wireless Networks
• Deep Reinforcement Learning for Uplink Multi-Carrier Non-Orthogonal Multiple Access Resource Allocation Using Buffer State Information
• DRL-based Joint Resource Scheduling of eMBB and URLLC in O-RAN
• When Multiple Agents Learn to Schedule: A Distributed Radio Resource Management Framework
• Buffer-aware Wireless Scheduling based on Deep Reinforcement Learning