Use of Machine Learning for the Estimation of Down- and Up-Link Field Exposure in Multi-Source Indoor WiFi Scenarios

Abstract

Overview

A novel Machine Learning (ML) approach based on Neural Networks (NN) is introduced to quantify Radio-Frequency (RF) exposure from WiFi sources in enclosed environments. This advanced model aims to address the complex and variable nature of real-life exposure scenarios.

Methodology

The neural network developed in this study leverages data readily available in typical settings, such as the positioning and types of WiFi equipment and the physical characteristics of the environment, like wall material and penetration loss. This model represents a significant enhancement in the predictability of RF fields, central to health risk assessments concerning electromagnetic exposure.

Findings

• The NN model was calibrated and verified against an additional layout not used during the training phase, ensuring the robustness and reliability of the predictions.
• Prediction accuracy metrics included a median error range from -0.4 to 0.6 dB and a root mean square error between 2.5 and 5.1 dB, offering substantial improvements over traditional indoor network planners utilized for estimating target electric fields.

Conclusion

Given its high accuracy and adaptability to various layouts, this methodology is particularly effective for evaluating RF exposure in indoor settings. By improving the understanding and control of EMF exposure, this research contributes to ongoing discussions about the potential health impacts of electromagnetic fields, promoting safer living and working environments.