Exploring the application of various condenser microphones for wrist pulse measurement using machine learning models

Area of Science:

• Biomedical Engineering
• Cardiovascular Health Monitoring
• Sensor Technology

Background:

• Wrist pulse measurement is vital for cardiovascular health assessment.
• Existing sensors (optical, pressure, image, ultrasonic) face limitations like environmental interference and system complexity.
• Condenser microphones are emerging as a viable alternative for pulse detection.

Purpose of the Study:

• To evaluate the effectiveness of different condenser microphone types for wrist pulse measurement.
• To compare the accuracy of omnidirectional versus unidirectional condenser microphones.
• To assess the performance of machine learning models in predicting pulse rate using microphone data.

Main Methods:

• Three condenser microphones (one omnidirectional, two unidirectional) were tested.
• A mechanical pulse acquisition system was developed.
• Data from 27 healthy subjects were analyzed using four regression models (Partial Least Square Regression, Ridge Regression, Principal Component Regression, Nu-Support Vector Regression).

Main Results:

• Unidirectional condenser microphones demonstrated higher accuracy than the omnidirectional type.
• A unidirectional microphone with a sensitivity of -50 to -44 dB outperformed one with -40 to -34 dB.
• Nu-Support Vector Regression achieved the lowest prediction errors, showing superior performance.

Conclusions:

• Unidirectional condenser microphones are more suitable for wrist pulse measurement.
• Microphone sensitivity plays a crucial role in accuracy.
• Machine learning, especially Nu-Support Vector Regression, significantly improves pulse rate prediction accuracy, guiding future sensor development.