Computationally-Efficient Pulse Rate Estimation From Compressed PPG Measurements for Continuous Vital Signs Monitoring

Energy-constrained vital monitoring devices highly demand energy-efficient pulse rate (PR) estimation from photoplethysmogram (PPG) signal. In this paper, we propose a computationally-efficient PR estimation method using compressed sensing (CS) based PPG measurements. The proposed method includes preprocessing, CS measurement generation, Fourier magnitude computation, and maximum spectral peak based PR computation. Among different sensing matrices such as random Gaussian, random Bernoulli, random sparse binary and deterministic block binary diagonal (DBBD), the DBBD sensing matrix based measurements can be directly used for the PR estimation. The proposed CS-PPG based PR estimation method is evaluated using three standard databases (MIMIC, BIDMC and CapnoBase) for signal lengths of 10 s and 60 s. On the standard metrics, the performances of the CS-PPG and original-PPG based PR estimation methods are comparable, but the CS-PPG based method requires only one fourth (25%) of the computational operations compared to original-PPG based method. Further, failure cases of Fourier spectrum based PR estimation method are highlighted under pathological PPG signals.