Resource allocation in mmWave 5G IAB networks: A reinforcement learning approach based on column generation

Millimeter wave (mmWave) communications have been introduced in the 5G standardization process due to their attractive potential to provide a huge capacity extension to traditional sub-6 GHz technologies. However, such high-frequency communications are characterized by harsh propagation conditions, thus requiring base stations to be densely deployed. Integrated access and backhaul (IAB) network architecture proposed by 3GPP is gaining momentum as the most promising and cost-effective solution to this need of network densification. IAB networks' available resources need to be carefully tuned in a complex setting, including directional transmissions, device heterogeneity, and intermittent links with different levels of availability that quickly change over time. It is hard for traditional optimization techniques to provide alone the best performance in these conditions. We believe that Deep Reinforcement Learning (DRL) techniques, especially assisted with Long Short-Term Memory (LSTM), can implicitly capture the regularities of environment dynamics and learn the best resource allocation strategy in networks affected by obstacle blockages. In this article, we propose a DRL based framework based on the Column Generation (CG) that shows remarkable effectiveness in addressing routing and link scheduling in mmWawe 5G IAB networks in realistic scenarios.