Base station prediction and proactive mobility management in virtual cells using recurrent neural networks

Abstract

Multicell cooperation in 5G next-generation wireless networks is essential to increasing multiuser channel capacity. Multiple base stations need to coherently process their transmitted (or received) data streams to mitigate inter-cell interference and achieve significant diversity gains. This is only possible if the correct base stations are selected. As users demand higher data rates at higher mobility, the time required to predict the optimal set of base stations to create a virtual cell is significantly reduced. In this paper, a method based on a Recurrent Neural Network (RNN) is presented to rapidly predict the next base station that a mobile node will associate with. RNNs have been used in machine learning to identify sequential data patterns such as required in protein sequence classification. In this research a RNN, trained using sequences of Received Signal Strength (RSS) values, is used to predict base station association. Simulation results demonstrate that the proposed machine learning method achieves an accuracy of over 98% to predict the optimal virtual cell topology in the time required based on the mobility of users.