Machine Learning based Radio Environment Map Estimation for Indoor Visible Light Communication

Abstract

This paper presents a machine learning-based methodology for constructing optical Radio Environment Maps (REMs) for indoor visible light communication (VLC) systems. First, REM estimation is evaluated using configurable simulated datasets generated by a physics-based indoor VLC simulator. The selected REM estimation method is then benchmarked against alternative learning- and interpolation-based methods using experimental measurements. Evaluation of Decision Tree and Multi-Layer Perceptron (MLP) models on received signal strength (RSS) datasets generated for different room geometries and transmitter configurations shows that MLPs achieve superior accuracy-performance trade-offs through appropriate tuning of the training sample size, number of epochs, and batch size. Experimental benchmarking, spatial validation, and robustness analysis further demonstrate the predictive accuracy and spatial generalization ability of the MLP. To address sparse measurement scenarios, synthetic data are generated using SMOGN (Synthetic Minority Over-Sampling Technique for Regression with Gaussian Noise) and Gaussian Process Regression (GPR). The results show that GPR-based augmentation can reduce the required number of real measurements by up to 83\% while maintaining prediction accuracy. These findings support the practical viability of data-driven REMs for indoor VLC applications such as network planning, localization, and digital twins.

arXiv commentFinal author manuscript incorporating substantial revisions made during peer review, including expanded experimental validation and additional analysis. Published in IET Optoelectronics, vol. 20, no. 1, e70043 (2026)