Interpolation-Based Densification of Sparse Measurement Datasets for 5G+ Systems

Abstract

The process of obtaining dense datasets from heterogeneous cellular networks is a difficult task that requires extensive measurement campaigns, considerable time, and financial resources. Therefore, we propose an interpolation-based approach to densify sparse datasets acquired from a limited number of measurements. We demonstrate that the proposed approach can accurately model the signal levels of the serving base station (BS) and predict the coverage by neighboring cells. With the best performing linear and Kriging interpolation algorithms, we achieved values of mean absolute error under 3 dB. Specifically, both methods achieved a prediction error of less than 5 dB for more than 80 % of the measurement points. The prediction accuracy was nearly constant over the entire range of separation distances between the BSs and end devices. Owing to the minimal need for parameterization of the input dataset, the proposed solution is significantly less computationally demanding than machine learning (ML)-based approaches, and does not require training-validation cycles.