Estimation of Electromagnetic Environment of Wireless Medical Telemetry Services Using Machine Learning

Abstract

In this study, we developed the novel machine learning model to estimate electromagnetic environment of wireless medical telemetry services (WMTS) using In-phase / Quadrature phase data (I/Q data) of normal and interfered its signals. If the occurrence of electromagnetic interference (EMI) sources and carrier to noise ratio (CNR) can be estimated automatically, the management of electromagnetic environment of WMTS can be easier. We, therefore, proposed the machine learning method for estimating an occurrence of EMI and calculating CNR. As the results of the performance evaluation by Stratified K-Fold cross-validation on 74 types of recorded I/Q data, an occurrence of EMI was predicted with 97.6% accuracy by AdaBoost classifier. Whether the CNR was more than 30 dB was predicted with 95.6% accuracy by decision tree classifier. In addition, this method classified 4 CNR levels such as < 20 dB, 20 to 30 dB, 30 to 40 dB, and 40 dB ≦ with 90.5% accuracy by random forest classifier. Moreover, CNR was estimated with 99.5% R-square and 0.661 dB mean absolute error by K-Fold cross-validation using gradient boosting regression tree. Our method is effective for investigating electromagnetic environments in clinical settings.