Multi-agent Reinforcement Learning Based Distributed Bandwidth Resource Allocation for Multipath Transmission

In intelligent office environments with multiple users and diverse service requirements, the concurrent emergence of differentiated service demands often leads to bandwidth resource limitations and frequent network disruptions. These issues significantly compromise the Quality of Experience (QoE) for users. Multipath transmission technology offers a viable solution to this challenge, not only enhancing the overall network capacity but also improving network stability. The key to multipath transmission lies in the optimal bandwidth allocation among various paths and service types, as well as the optimization of redundant transmission across paths. This paper presents a multipath distributed bandwidth resource allocation algorithm based on multi-agent reinforcement learning(MARL), which learns optimal bandwidth allocation strategies and FEC redundancy ratio settings through direct QoE feedback from transmission clients, thereby adapting to network dynamics. We built a simulation platform on Mininet and conducted transmission experiments for video streaming and file download services in dynamic network scenarios with both two-path and multi-path configurations. Experimental results demonstrate that, in multi-user and multi-service scenarios, the proposed multipath bandwidth allocation method exhibits significant advantages across various QoE metrics for both video streaming and file download services.