UC Irvine

A photoplethysmography (PPG) is a unsophisticated and reasonable-cost optical technique that is frequently utilized in the healthcare field to extract useful health-related data such as heart rate variability, blood pressure, and respiration rate. With the use of portable wearable devices, PPG signals can be captured constantly and remotely. These measuring devices, however, are susceptible to motion artifacts induced by everyday activities. Using extra accelerometer sensors is the most frequent technique to minimize motion artifacts, but they have two drawbacks: (1) Excessive power consumption; (2) the requirement for an ac- celerometer sensor in a wearable device (which is not required in certain wearables). In this thesis, we provide a non-accelerometer-based PPG motion artifacts reduction method that outperforms previous methods in terms of accuracy. To rebuild clean PPG signals from noisy PPG signals, we employ a Cycle Generative Adversarial Network (CycleGAN). In compar- ison to the state-of-the-art, our remarkable machine-learning-based technique provides 9.5 times improvement in motion artifact removal without the use of additional sensors like an accelerometer.