Use of Machine Learning for the Estimation of Down- and Up-Link Field Exposure in Multi-Source Indoor WiFi Scenarios
Abstract
Abstract Overview
A novel Machine Learning (ML) method based on Neural Networks (NN) is introduced for assessing radio-frequency (RF) exposure in indoor scenarios involving WiFi sources.
Study Aim
The study aims to develop an NN equipped to handle the complexity and variability inherent to real-life exposure setups, accommodating the effects of both down-link and up-link transmissions from various devices.
Methodology
The NN utilized readily available data such as the location and types of WiFi sources and the building's physical characteristics.
Performance Evaluation
- The NN was tested using a new layout, showing high field prediction accuracy under various exposure conditions.
- The median prediction error ranged between -0.4 to 0.6 dB, with a root mean square error of 2.5-5.1 dB.
Conclusion
This versatile model performs effectively across different layout configurations, demonstrating its general applicability in assessing RF exposure in indoor environments.