Mobility-Aware Resource Allocation for mmWave IAB Networks via Multi-Agent RL

MmWave communications are expected to provide huge wireless access data rates. However, mmWave signals are strongly affected by high path losses and blockages, which can only be partially alleviated by directional phased-array antennas. This makes mmWave networks coverage-limited, thus requiring network densification. 3GPP has introduced Integrated Access and Backhaul (IAB) architecture as a cost-effective solution. Resource allocation in IAB networks is complicated because it has to cope with directional transmissions, device heterogeneity, intermittent links, and mobile users. While traditional optimization techniques usually struggle in these scenarios, we believe Reinforcement Learning (RL) techniques, especially Multi-Agent RL (MARL), can implicitly capture environment dynamics and lead to interference coordination among nodes. In this paper, we propose an MARL-based framework that shows remarkable effectiveness in addressing flow allocation and link scheduling for mmWave 5G IAB networks in scenarios with random obstacles and mobile users.