BACKGROUND
As the demand to acquire bio-signals and use them for health management in daily life increases, it is becoming more common to mount photoplethysmography on mobile devices such as smartphones and smart watches. However, though it is very important to know the optimized signal measurement conditions due to resource limitations such as computing power and battery life in the mobile environment, the optimal signal acquisition conditions required to derive clinically meaningful results has not been proposed yet.
OBJECTIVE
This study aims to suggest appropriate criteria for measurement photoplethysmogram with clinical utility by identifying the changes in photoplethysmogram waveform with decreasing sampling frequency and quantization bit depth.
METHODS
Photoplethysmograms recorded at a 1-kHz sampling frequency and 16-bit quantization bit depth were converted to signals with sampling frequencies of 500-, 250-, 100-, 50-, 25-, and 10-Hz by means of down-sampling, and then we did re-quantization to convert the quantization bit depth into 16, 14, 12, 10, 8, and 6 bits for each down-sampled signal. Degradation of the signal was quantified in terms of morphological change using normalized root mean square error and feature-point deviation using mean absolute error at representative photoplethysmogram features such as pulse onset or systolic peak.
RESULTS
Although there were differences according to pulse onset and systolic peak, the sampling frequency of ≥ 250-Hz and 16-bit quantization bits are required in order to have ≤ 1 ms of timing error and a normalized amplitude error of ≤ 1%. In addition, a sampling frequency ≥ 100-Hz and a 12-bit quantization bits are recommended to have feature-point time errors and amplitude errors of < 10 ms and a normalized amplitude error of < 10%.
CONCLUSIONS
Our results highly recommend that 16-bit quantization bit depth and ≥ 250-Hz sampling to secure < 1% of error, and ≥ 25-Hz sampling and ≥ 10-bit quantization bit depth for minimal use compared with photoplethysmogram obtained by 1-kHz sampling and 16-bit quantization bit depth.