Comparison of curve fitting methods for synthetic generation of PPG

With the advancement of wearable sensors, Photoplethysmography (PPG) signals are widely used for heart rate measurements. But the characteristics of PPG signal with variation in heart rate is not studied much due to lack of publicly available PPG databases, for which data augmentation is a possible solution. This work aims to synthetically generate PPG signals by incorporating the characteristic changes in the Heart Rate Variation (HRV) and the morphological changes as present in a real PPG signal using curve fitting methods. Measures to reduce error of synthetic generation of PPG signals has been carried out in this study. Two curve fitting methods, SCF and GCF were proposed in this work to reduce error- a combination of the two curve fitting methods and a method which uses four gaussians for synthetic generation of PPG signals. Compared to the most commonly used curve fitting method, the combined method could reduce the error by 93.4%, where as the four guassian method could reduce by 57.7%. The former could best replicate the HRV changes present in the signal but requires a reference PPG signal, where as the latter could generate different templates of PPG signal without using any reference signal for curve fitting.