Photoplethysmography-Based Distance Estimation for True Wireless Stereo.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Wearable Technology

Background:

• Wearable devices are increasingly used for healthcare, but True Wireless Stereo (TWS) devices have limitations in space and power.
• Photoplethysmography (PPG) sensors are vital for health monitoring but susceptible to noise from motion and individual characteristics.
• Effective noise reduction is crucial for accurate PPG signal analysis in TWS devices.

Purpose of the Study:

• To propose a distance estimation algorithm for PPG signals specifically designed for TWS devices.
• To address the challenges of noise reduction and simultaneous multi-function implementation in resource-constrained TWS.

Main Methods:

• Development of a waveform adjustment (WA) filter to minimize noise while preserving signal integrity.
• Implementation of MobileNet, a lightweight deep learning model, for efficient PPG signal processing.
• Establishment of a PPG monitoring testbed for algorithm verification with three distance estimation criteria.

Main Results:

• The proposed algorithm achieved high performance metrics: 92.5% accuracy, 92.6% precision, 92.8% recall, and 0.927 F1 score with a signal length of 15.
• While other algorithms showed superior metrics in specific cases, the proposed model demonstrated the fastest inference time.
• The algorithm effectively reduces noise in PPG signals for TWS applications.

Conclusions:

• The developed distance estimation algorithm offers a viable solution for PPG signal processing in TWS devices.
• The combination of the WA filter and MobileNet provides an efficient and accurate method for wearable health monitoring.
• The proposed approach balances performance with the computational constraints of TWS, enabling advanced healthcare functionalities.