Federated learning for resource allocation in vehicular edge computing-enabled moving small cell networks

Moving networks comprising of moving small cells (MoSCs) is an emerging technology that provide ubiquitous connectivity to the cellular users in vehicular environment. MoSCs are the small cells deployed on the top of vehicles (city buses, trains, trams etc.) to support the vehicular users with improved quality-of-service (QoS). However, the deployment of small cells in vehicular environment demands for an efficient resource allocation mechanism. This is due to high mobility of MoSCs resulting in dynamic interference between MoSCs, high computational cost, privacy issues, latency issues, and high data transmission requirement. To overcome these issues, gated recurrent unit (GRU)-based federated learning (FL) model is proposed for resource allocation in moving networks. In our proposed work, we investigate resource allocation in vehicular edge computing (VEC)enabled MoSC network with MoSCs deployed on trams travelling with deterministic mobility. In the proposed MoSC network, road side units (RSUs) equipped with VEC servers use their computational power to train the resource allocation model in a distributed manner, in which each RSU exploits the training data of its associated MoSCs to generate a shared model. The proposed GRU-based FL model enables RSUs to cooperatively train a global model that can predict resource block (RB) allocation to MoSCs without transmitting the historical data to the central server. We have conducted extensive system level simulations to determine key performance comparison between FL and centralized learning-based resource allocation in MoSC network.