MagnifiSense: inferring device interaction using wrist-worn passive magneto-inductive sensors

摘要: The different electronic devices we use on a daily basis produce distinct electromagnetic radiation due to differences in their underlying electrical components. We present MagnifiSense, low-power wearable system that uses three passive magneto-inductive sensors and minimal ADC setup identify the device person is operating. MagnifiSense achieves this by analyzing near-field from common components such as motors, rectifiers, modulators. conducted staged, in-the-wild evaluation where an instrumented participant used set of variety settings home cooking outdoors commuting vehicle. classification accuracy 82.6% using model-agnostic classifier 94.0% model-specific classifier. In 24-hour naturalistic deployment, correctly identified 25 total 29 events, while achieving low false positive rate 0.65% during 20.5 hours non-activity.