A Note on Machine Learning-based EMI Spectrum Prediction for Signals’ Waveform on the Characteristic and Frequency Modulation

Electromagnetic interference (EMI), also known as radio frequency interference, is noise or disruption in an electrical circuit caused by external sources. It can degrade the performance of electronic devices, lead to malfunctions, or cause complete failure. EMI can come from both natural and artificial sources. Therefore, it is essential to use machine learning for predicting and minimizing its effects. In this study, we present our findings from investigating the use of a machine learning (ML) technique to predict EMI based on the parameters of signal characteristic and frequency. The characteristics of signal waves and frequency modulation are well-known parameters that significantly affect to the EMI spectrum. To understand their effects on EMI spectrum, five types of signals, including sine, ampalt, gauss, lorentz, and noise, are analyzed. The main features of each signal are extracted and used to train a machine learning model that can accurately predict EMI spectrum. The results show that linear regression method exhibits a higher error when applied to lorentz signal, whereas it achieves a lower root mean square error (RMSE) when modeling noise signals, 10.29% and 7.14%, respectively. These results open up new opportunities to predict the EMI spectrum with insufficient dataset more resistant, as well as for improving the performance of electronic systems in high-density environments.