Transfer Learning for QoS Aware Topology Management in Energy Efficient 5G Cognitive Radio Networks

Abstract

In this paper, we investigate the use of a transfer learning approach applied to a topology management framework in a 5G heterogeneous aerial-terrestrial broadband access network, to reduce energy consumption and deployment cost, and improve system capacity and QoS. We implement a cognitive engine at the base station (BS), with reinforcement learning algorithms applied at the link level for spectrum assignment, and at the network level for user association. A novel transfer learning algorithm is developed to transfer the expertise knowledge learnt from spectrum assignment to formulate a knowledgebase for user association. Furthermore, a QoS aware base station switching operation algorithm is proposed at a network controller, to dynamically switch BSs between sleep and active modes based on system QoS requirements. System simulations under practical configurations show that the transfer learning based user association algorithm achieves significant energy saving and QoS improvement with optimized load management in a spectrum sharing scenario. The BS switching operation algorithm effectively controls the delay and retransmissions when saving energy from sleep mode.