[Retracted] Edge-Enabled Heart Rate Estimation from Multisensor PPG Signals

Table 3

Comparison of our proposed method with several existing methods on computation time.

Nos.	Devices and methods	Signal type	Sampling rate (Hz)	Loading data length (s)	Processing platform configuration	Computation time
1	PC, Zhang [17]	PPG	125	3600	Intel Core-i7 4790 at 3.60 GHz, 8 GB RAM, windows 7 64 bit	3.5 h
2	PC, Khan et al. [21]	PPG	125	3600	Intel Core-i7 [email protected] GHz, 8 GB RAM, windows 7 64 bit	668 s
3	PC, Han et al. [45]	PPG	500	30	Intel Core i7 [email protected] GHz, RAM 16.0 GB	0.15 s
4	Wearable device, Burrello et al. [49]	PPG	100	3	STM32H743ZIT6 with480 MHz, 1 MB RAM, and 2 MB FLASH	1.27 s
5	Edge device, our method	PPG	200	30	Raspberry Pi 3B+, ARM [email protected] GHz 2 GB SDRAM	15.25 s
6	Edge device, our method	PPG	200	30	Raspberry Pi 4B, ARM [email protected] GHz, 4 GB SDRAM	4.24 s

Note. There is a special case in this experiment, which aims at subjects 3 and 7. For the case of subject 3, the MAEP of the EeHRA-right method is more than that of the EeHRA-left and EeHRA-bilateral methods. For a similar situation, subject 7 gets the highest MAEP using the EeHRA-left. These problems result in higher MAEP of the EeHRA-bilateral method. Furthermore, in these situations, the abnormal IHR mechanism fails because the scores of some unuseful IHR values in the HRP are not close to 1. According to the principle of the AntIHub, these IHRs are deemed to be the normal data. So, in this condition, the MAEP of the EeHRA-bilateral increases.