Domain Compliant Recommendation of Remote Electrical Tilt Using ML Approach

Abstract

Due to highly complex and unpredictable nature of telecom network dynamics, optimal adjustment (daily/weekly or as per need) of antenna tilt regularly at large scale (for thousands of cells) is necessary to maintain acceptable quality of service(QoS). Optimal tilt prediction using standard data and algorithm driven approaches like simulation or modelling ( based on ML/DL/Statistical modelling etc.) fundamentally lack to address the critical aspect of legitimate tilt prediction, namely, antenna tilt and coverage are inversely related as governed by physical laws in telecommunication science. This fundamental lack produces inconsistent tilt prediction resulting poor cell coverage which accumulates over multiple cells in the network and reduces network level efficiency. In this paper we propose a synthetic sampling scheme which can enforce any model to learn this domain principal of tilt-range relation through generated smart training sample. This enables legitimate tilt prediction at large scale which can improved cell and also network level performance. The proposed approach has been tested in field with observed improvement in cell and network level performance.