Feed-Forward Neural Network Based Mode Selection for Moving D2D-Enabled Heterogeneous Ultra-Dense Network

Abstract

Device-to-device (D2D) communications have been proposed as a promising technology to improve system capacity and user experiences. In moving D2D-enabled heterogeneous ultra-dense networks (H-UDNs), it will cause heavy system overhead from the frequent mode selection between D2D mode and cellular mode, which is also belong to handover strategies. Thus, the optimization of mode selection strategy is needed urgently. In this paper, for the mode selection occurring from cellular communication mode to D2D communication mode (C2D), we propose a feed-forward neural network (FFNN) based multi-attribute D2D transmitter choosing strategy. The proposed strategy implements FFNN model, meanwhile combine the stochastic geometry based long-term analytical results with instant parameters involved in mode selection process. As a result, our proposed strategy brings improvements to the mode selection performance, which can be observed in reducing the mode selection probability and increasing the D2D mode dwell time. Moreover, the system overhead is further reduced on the basis of achieving full-spectrum reuse.