Pulse rate estimation from forehead photoplethysmograph signal using RLS adaptive filtering with dynamical reference signal

Abstract

Objective: Pulse rate (PR) extracted from Photoplethysmograph (PPG) is a significant feature in emotion recognition. However, PPG signal can be easily affected by the motion artifacts (MA) produced from the body movement and the random head swing of the subjects who are in Immersive Virtual Environments (IVE). The objective is to estimate PR from MA corrupted PPG signal.Methods: In order to reduce the influence of random MA, we propose a novel adaptive noise cancellation (ANC) method. This method could attenuate MA and reduce the execution time by adopting RLS adaptive filtering with dynamical reference signal (DRS-RLS). Then the spectral peak tracking with verification based on the Fast Fourier Transform (FFT) method is used to estimate PR value.Results: In this paper, the two-channel PPG signals and tri-axis acceleration signals were recorded synchronously from forehead by the virtual reality (VR) physiological helmet platform developed in our research center. Seventeen participants were able to observe and move randomly in four IVEs. The mean average absolute error (μAAE) using our proposed PR estimation method is 2.40 beats per minute (BPM). Furthermore, verification result of the proposed method on a public database of 22 PPG recordings showed the μAAE (2.01 BPM) and runtime (17.2 ms) of each time window.Conclusion: The proposed PR estimation method has achieved the lowest error compared with RLS adaptive filtering using fixed axis as reference signal or cascade RLS adaptive filtering. And the result has also shown a lower μAAE and less runtime than other adaptive filtering methods proposed in recent years.Significance: The proposed method in this paper has potential to be implemented in wearable devices due to its low error and low computation cost. It also provides a new way for reference signal selection of adaptive filtering.