Coordinated Machine Learning for Handover in Mobile Networks with Transparent Relaying UAVs

The Unmanned Aerial Vehicles (UAVs) acting as relays in the mobile networks are usually energy constrained. To improve the energy efficiency of such networks, UAVs should operate in a transparent relaying mode. In such mode, however, the channel quality between users and UAVs cannot be measured, since the transparent relays do not transmit own signaling. A lack of information on the quality of channel between users and UAVs limits practical implementation and is serious restraint for mobility management. To overcome this limitation, we develop a novel concept of coordinated machine learning for handover of users and UAVs in the mobile networks with transparent relaying UAVs. First, we predict the channel quality from other known information in the network via deep neural network (DNN). Such predicted channel quality is then fed into deep reinforcement learning (DRL) for an adjustment of handover parameter – cell individual offset (CIO). Unfortunately, a simple concatenation of the DNN and the DRL leads to a notable performance degradation. Hence, we propose a coordination of the DNN for channel quality prediction and the DRL for CIO setting. The coordination consists in a mutual exchange of performance-related information and an update of DNN according to a reward of DRL. The proposal increases the sum capacity by up to 12.7% while reducing the number of user and UAV handovers by up to 12.9% and 16.4%, respectively, compared to related works.