ระบบจัดเก็บและค้นหางานวิจัยอัจฉริยะ

AutoPPGEncoder: Autoencoder Training With Two-Stage Peak-Aware Loss for Resource-Constrained PPG Signal Compression

Suvichak Santiwongkarn | Autoencoder photoplethysmograph resource-constrained

จำนวนผู้เข้าชม

AutoPPGEncoder: Autoencoder Training With Two-Stage Peak-Aware Loss for Resource-Constrained PPG Signal Compression

Collection	RMUTK Research Repository (RMUTK IR) วารสารวิชาการ — e-Journal Articles
ID RMUTK Digital	RMUTK000118
Title	AutoPPGEncoder: Autoencoder Training With Two-Stage Peak-Aware Loss for Resource-Constrained PPG Signal Compression
Alternative title	AutoPPGEncoder: Autoencoder Training With Two-Stage Peak-Aware Loss for Resource-Constrained PPG Signal Compression
Authors	อ.ดร.อภิวัฒน์ ดิษฐาพร Suvichak Santiwongkarn ผู้แต่งหลัก Chatdanai Hutchaleelaha Tanut Chokchatchawathi Phairot Autthasan Prapun Suksompong Corresponding Theerawit Wilaiprasitporn Corresponding
Faculty	คณะบริหารธุรกิจ
Journal Title	IEEE Sensors Journal
ISSN	1558-1748
Volume / Issue / Pages	Vol.26 \| No.12 \| pp.19343-19355
Published	2026-04-29
Year	2569
Level	ระดับนานาชาติ (SCOPUS)
Quartile	Q1
DOI	10.1109/JSEN.2026.3686963
Abstract	Wireless wearable devices enable convenient acquisition of photoplethysmograph signals. A limitation of continual measurement is the resource-constrained processing and limited battery capacity. To solve the transmission power consumption problem, we propose an information reduction method, AutoPPGEncoder. The AutoPPGEncoder is an autoencoder trained with conventional mean squared error (mse) loss and mse systolic peak amplitude loss. The scheduling of impact from two losses improves the raw signal reconstruction by preserving the systolic peak in the reconstructed signal. Our system reduces the amount of transmitted BLE data from the PPG data by a ratio of 1:7.07. In terms of signal quality, the AutoPPGEncoder performs better than the recent compression technique, deep compressive sensing network (DCSNET) compression in both percentage root-mean-square difference (PRD) and peak amplitude PRD on BIDMC, SensAI, and VitalDB datasets. The AutoPPGEncoder’s power consumption was verified on an NRF54L15-DK board, which reduced the total energy consumption by 0.4 mJ per segment. This demonstrated the capability of the AutoPPGEncoder in IoT applications with power efficiency and the quality of PPG signal transmission.
Abstract (EN)	Wireless wearable devices enable convenient acquisition of photoplethysmograph signals. A limitation of continual measurement is the resource-constrained processing and limited battery capacity. To solve the transmission power consumption problem, we propose an information reduction method, AutoPPGEncoder. The AutoPPGEncoder is an autoencoder trained with conventional mean squared error (mse) loss and mse systolic peak amplitude loss. The scheduling of impact from two losses improves the raw signal reconstruction by preserving the systolic peak in the reconstructed signal. Our system reduces the amount of transmitted BLE data from the PPG data by a ratio of 1:7.07. In terms of signal quality, the AutoPPGEncoder performs better than the recent compression technique, deep compressive sensing network (DCSNET) compression in both percentage root-mean-square difference (PRD) and peak amplitude PRD on BIDMC, SensAI, and VitalDB datasets. The AutoPPGEncoder’s power consumption was verified on an NRF54L15-DK board, which reduced the total energy consumption by 0.4 mJ per segment. This demonstrated the capability of the AutoPPGEncoder in IoT applications with power efficiency and the quality of PPG signal transmission.
Keywords (TH)	Autoencoder photoplethysmograph resource-constrained signal compression systolic peak
Keywords (EN)	Autoencoder photoplethysmograph resource-constrained signal compression systolic peak
Access Level	Open Access (เปิดสาธารณะ)

ลิงก์แหล่งเผยแพร่

เอกสารฉบับเต็ม (ลิงก์ภายนอก)

https://ieeexplore.ieee.org/document/11501190

เปิดลิงก์