End-to-End Learning of Proactive Handover Policy for Camera-Assisted mmWave Networks Using Deep Reinforcement Learning

For mmWave networks, this paper proposes an image-to-decision proactive handover framework, which directly maps camera images to a handover decision. With the help of camera images, the proposed framework enables a proactive handover, i.e., a handover is triggered before a temporal variation in the received power induced by obstacles even if the variation is extremely rapid such that it cannot be predicted from a time series of received power. Furthermore, direct mapping allows scalability for the number of obstacles. This paper shows that optimal mapping is learned via deep reinforcement learning (RL) by proving that the decision process in our proposed framework is a Markov decision process. While performing deep RL, this paper designs a neural network (NN) architecture for a network controller to successfully learn the use of lower-dimensional observations in state information and higher-dimensional image observations. The evaluations based on experimentally obtained camera images and received powers indicate that the learned handover policy in the proposed framework outperforms the learned policy in a received power-based handover framework.