Preprocessing of PPG and EDA signals for detection of emotional and cognitive states via physiological signals

Abstract

Presented in the current paper is a methodology for approaching the preprocessing of Photoplethysmography and Electrodermal activity for the detection of emotional and cognitive states in humans via physiological signals. Examined closely are the effects of downsampling and segmentation of the PPG, the segmentation and separation of the Skin Conductance Level (SCL), and Skin Conductance Response (SCR) components of the EDA signal with both median and low pass filters. The results from the research indicate that the most appropriate preprocessing with regard to emotions and cognitive load classification is segmentation of 2 minutes which is the recommended length for frequency analysis of heart rate variability. Recommended, furthermore, is the downsampling of the PPG to 64 Hz, which proved to be the lowest sampling frequency that doesn’t introduce errors in the systolic peak detection, neither does it drastically affect the length of the Inter Beat Intervals (IBIs). Proposed, as to the separation of the SCL component of the EDA, is the usage of median filter with window length of 75% of the sampling frequency, which introduces negligible artefacts, mainly at the start of the signal.