On learning and exploiting time domain traffic patterns in cellular radio access networks

Abstract

This paper presents a vision of how the different management procedures of future Fifth Generation (5G) wireless networks can be built upon the pillar of artificial intelligence concepts. After a general description of a cellular network and its management functionalities, highlighting the trends towards automatization, the paper focuses on the particular case of extracting knowledge about the time domain traffic pattern of the cells deployed by an operator. A general methodology for supervised classification of this traffic pattern is presented and it is particularized in two applicability use cases. The first use case addresses the reduction of energy consumption in the cellular network by automatically identifying cells that are candidates to be switched-off when they serve low traffic. The second use case focuses on the spectrum planning and identifies the cells whose capacity can be boosted through additional unlicensed spectrum. In both cases the outcomes of different classification tools are assessed. This capability to automatically classify cells according to some expert guidance is fundamental in future networks, where an operator deploys tenths of thousands of cells, so manual intervention of the expert is unfeasible.