Andriana Kusuma Dewi, Dyah Lestari
The absence of applying back electromotive force (Back EMF) to compact drivers such as earphones motivated us to develop a back EMF dataset for in-ear wired earbuds. This study aims to assess the feasibility of using back EMF to distinguish common user activities with earphones, including putting on, taking off, and scratching. The experiment involved recording back EMF waves from wired earphones using a feedback controller circuit, and feature extraction in the time domain was performed, covering fundamental statistical features, waveform-derived features, distribution features, and wave energy-related features. Datasets were obtained from ten users with predefined conditions. After the extraction process was completed, a normality assessment using the Shapiro-Wilk method showed that all data exhibited a non-normal distribution. We assessed the significance of the features using Kruskal-Wallis, Mutual Information (MI), Random Forest, and SHapley Additive explanation (SHAP) techniques. The test results indicate that zero crossing, zero crossing rate (ZCR), and the number of prominent peaks and valleys are critical features in shaping user interaction. These findings suggest that back EMF could serve as a low-cost alternative method for earphone detection, replacing capacitive sensors in detecting user activities and enabling practical applications without requiring additional sensors. © 2025 IEEE.
State University of Malang, Dept. Electrical and Informatics Eng., Malang, Indonesia
