Mobile Operator Collaboration Using Cooperative Multi-Agent Deep Reinforcement Learning

Next generation mobile networks will provide connectivity services for an unprecedented number of mobile devices, used in industry-vertical applications with diverse network requirements as to throughput, latency, and geographical area of coverage. In order to satisfy these requirements at scale, collaboration between multiple operator is oftentimes essential. Current collaborations between network operators are long-term, reactive, and established with human involvement. The dynamic nature of mobile network traffic demands contradicts these collaborations’ rigid nature; thus, resulting in sub-optimal network resource allocation to connectivity service. In this work, we introduce an agent-based architecture for automating collaboration of mobile network operators based on predicted future demand. The architecture uses a multi-agent deep reinforcement learning algorithm, wherein every operator has its own agent suggesting future collaborations, which can change dynamically. Using simulation, we show that our approach outperforms the state of the art operator collaboration approaches and leads to more sustainable growth of mobile networks by reducing capital and operational expenses for mobile network operators.