Deep Learning based Mobile Network Management for 5G and Beyond

Over the last decade the deployment paradigm of the mobile networks has shifted from complex hierarchical structures towards decoupling of the control and transmission planes. Such transformation enables software defined network configurations that allow improving the scalability and flexibility of the network. However, the process of network management is not trivial, considering the complexity of modern 5G infrastructure, and requires new advanced solutions. In this paper, we propose a new architecture with additional plane, which is responsible for intelligent knowledge generation and decision making. Proposed approach allows to train deep neural networks by using generative adversarial learning for the scenario with limited amount of real data. We study the performance of the proposed approach for the particular use case of small cells coverage optimization considering given spatial and temporal variation of the traffic demand. Simulation results show that proposed approach achieves quasi-optimal small cells coverage in terms of overall network throughput even in conditions when there is a lack of real world statistical data.