On the Exploitation of CEEMDAN for PPG Synthetic Data Generation

The Photoplethysmogram (PPG) is commonly used in wearable sensors as it provides a reliable measure of the heart rate of a subject. This measure can be useful to detect a person’s health state and is strongly correlated to arousal, permitting to monitor a person’s level of stress. The major limitations of PPG signal adoption are related to the presence of different types of noise, especially when wearable devices are adopted in real-word environment, and to the difficulties in acquiring a huge amount of data to properly train machine learning models. In this paper, the adoption of Complete Ensemble Empirical Mode De-composition with Adaptive Noise (CEEMDAN) for PPG synthetic data generation is exploited, considering also the capability of the Intrinsic Mode Functions (IMFs) to reveal the presence of noise. Preliminary results are presented considering a dataset available in the literature related to PPG signals acquired during cognitive tasks. An in depth analysis on these results permits to underline the advantages and potential limits of the proposed strategy.