Performance Analysis for Virtual-Cell Based CoMP 5G Networks Using Deep Recurrent Neural Nets

Abstract

Providing high date rates that are independent of user location in the network is one of the Fifth Generation (5G) wireless network goals. This goal becomes even more challenging when the mobility of users is taken into account. The handovers that happens as the user crosses from cell to another can cause a sever degradation in the user's perceived data rate. The concept of Virtual Cell (VC) that is based on Coordinated Multipoint (CoMP) transmission is a promising solution for providing high data rates independent of user location in the network, and in particular for cell edge users. Multiple base-stations (BS) can coordinate with each other creating a Virtual Cell (VC). Users can roam within a Virtual Cell (VC) without the need to perform a handover. This diminishes the number of handovers a user will encounter, and also enhances the data rate for cell edge user by mitigating the Inter-Cell Interference (ICI) within a virtual cell. However, to enable the concept of Virtual Cells (VCs) rapid decisions need to be taken about when to enable / disable the VC mode and which Base Stations should be joining/leaving the VC as the user roams in the network. In this paper, the performance analysis of a novel algorithm based on a modification in the hidden layer of the Recurrent Neural Network (RNN), referred to as Gated Recurrent Units (GRUs). The RNN-GRU model is used for predicting the triggering conditions on enabling / disabling the VC mode. Sequences of the Received Signal Strength (RSS) values for different users in the network, were used for training the RNN-GRU model. After training, the RNN-GRU was used to predict the future RSS values, which is then used for making proactive decisions on enabling/disabling VC mode. Simulation results demonstrates that the proposed GRU-RNN model achieves an accuracy of 92% to predict the triggering conditions for enabling and disabling the CoMP mode as required based on the mobility of users.