류형록 | Engineering Portfolio · 전체 프로젝트

<aside> 📌

PPG·STM32·HRV·CNN+Transformer / CICS’25 제1저자(1/3)

발행 2025.10.22 · 발표 2025.10.23 · pp.291–292

논문 초록·결론 AUC 0.99·F1 0.992 / 코드 5-fold AUC 0.9988·F1 0.9775

회로·펌웨어·모델 · 논문·포스터 · PDF 원문 · 코드

</aside>

<aside> 💚

PPG–HRV Cognitive Load · Paper View

Website Case Study · GitHub Case Study · Repository

</aside>

flowchart LR
    A["Earlobe PPG · AFE"] --> B["STM32F411 acquisition"]
    B --> C["Peak · IBI · HRV features"]
    C --> D["CNN + Transformer"]
    D --> E["AUC · F1 · error analysis"]

귀불 PPG sensor부터 STM32F411 real-time signal processing, HRV feature extraction, CNN+Transformer classification까지 연결한 embedded biomedical AI project입니다.

1. Problem

인지부하를 단순 questionnaire가 아니라 physiological signal에서 추정하려면 sensor quality, peak timing, IBI consistency, HRV windowing, model evaluation이 하나의 chain으로 맞아야 합니다. 이 프로젝트는 analog circuit, MCU firmware, Python dataset pipeline, model output을 분리하지 않고 추적 가능한 형태로 정리했습니다.

2. System Architecture

Overall PPG HRV architecture

Overall PPG HRV architecture

Signal flow

  1. Earlobe PPG sensor와 analog amplification/filtering
  2. STM32F411 ADC acquisition
  3. Moving average, IIR band-pass, adaptive threshold, derivative FSM
  4. Peak interval에서 IBI/BPM 생성 후 UART stream
  5. Windowed HRV feature extraction and StandardScaler
  6. CNN local features + Transformer dependencies
  7. High / Low cognitive-load classification

3. Hardware and Experiment

PPG hardware overview

PPG hardware overview