비접촉 레이다 기반 비정상 호흡 시 산소포화도 추정 연구

다국어 초록 (Multilingual Abstract)

Sleep-disordered breathing, such as obstructive sleep apnea (OSA), is a major medical condition that induces intermittent hypoxia, fragmented sleep, and elevated cardiovascular risks. Continuous monitoring of blood oxygen saturation (SpO2) plays an essential role in diagnosing and evaluating such disorders. However, conventional photoplethysmography (PPG)-based oximetry requires direct skin contact and is highly vulnerable to motion artifacts, temperature variations, and blood perfusion changes, limiting its use for long-term and comfortable monitoring during sleep. In recent years, camera-based remote photoplethysmography (rPPG) has emerged as a non-contact alternative, using optical color variations to infer physiological rhythms. Yet, its performance is strongly affected by illumination, skin tone, distance, and movement, resulting in unstable readings under low-light or uncontrolled environments. These limitations emphasize the need for a robust, illumination-independent, and contact-free sensing system capable of continuously capturing oxygen saturation dynamics without the constraints of optical or physical contact.
This thesis presents a non-contact SpO2 estimation framework based on a 60.2 GHz frequency-modulated continuous-wave (FMCW) radar sensor, which overcomes the limitations of optical sensing. Radar offers distinct advantages, including insensitivity to ambient illumination, skin reflectivity, and body motion, while enabling sub-millimeter detection of chest wall displacement caused by respiration and cardiac activity. The received in-phase (I) and quadrature (Q) radar signals were processed through principal component analysis (PCA) to isolate the dominant respiratory component among range bins, followed by phase unwrapping to reconstruct a continuous chest displacement waveform. This radar-derived signal was temporally synchronized with medical-grade SpO2 data obtained under identical conditions. Experiments were conducted in a controlled environment replicating real sleep settings stable posture, low ambient light, and minimal external interference to analyze normal breathing, hypopnea, and apnea episodes across multiple subjects with diverse physiological characteristics.
To investigate the physiological relationship between respiration and oxygen saturation, three temporal indices were defined: desaturation delay (τ), resaturation delay (τ’), and saturation slope (∂SpO2/∂t). The delay τ represents the time interval between an abnormal breathing event and the onset of SpO2 decline, while τ’ describes the latency between breathing resumption and SpO2 recovery. The negative and positive slopes during desaturation and resaturation phases reflect oxygen depletion and reoxygenation efficiency. Building upon these observations, this study developed an empirical linear correction framework that reconstructs SpO2 dynamics based on respiratory patterns without relying on complex regression or machine learning algorithms. The framework consists of three core steps: (1) global delay alignment to synchronize radar respiration with delayed SpO2 response, (2) event-based smoothing to stabilize saturation transitions, and (3) empirical linear blending correction to model physiologically consistent desaturation–resaturation waveforms. This structure ensures transparent signal processing while maintaining strong physiological interpretability.
The proposed method successfully reproduced stable and physiologically consistent SpO2 variations across all breathing conditions. Unlike optical PPG and rPPG systems, the radar-based framework remained robust to lighting changes, motion, and skin reflectivity, maintaining reliable operation even in complete darkness. The temporal delays and slopes also reflected subject-specific physiological traits such as lung capacity, body composition, and ventilatory responsiveness, revealing that radar signals inherently encode meaningful information about oxygen exchange dynamics. These findings demonstrate that radar sensing can go beyond respiratory rate estimation to represent the complex interaction between ventilation and oxygen transport. Ultimately, this study establishes a non-contact, illumination-independent, and physiologically interpretable framework for SpO2 estimation. The approach improves the stability and reliability of oxygen monitoring while eliminating the constraints of optical sensing. Furthermore, it lays the groundwork for future extensions toward multi-parameter physiological estimation—integrating radar-derived respiration, heart rate, and ventilation—to enable personalized and continuous healthcare monitoring in both clinical and home environments.

목차 (Table of Contents)

• Ⅰ. 서 론 1
• 1.1. 연구 배경 및 필요성 2
• 1.2. 기존 측정 방식의 한계와 기술적 과제 6
• 1.3. 비접촉 레이다 기반 생체신호 모니터링의 필요성 11
• 1.4. 기존 연구 동향 및 한계 분석 14
• Ⅱ. 이론적 배경 19
• 2.1. SpO2와 호흡 생리학 개요 19
• 2.1.1. 혈중 SpO2와 산소분압의 관계 19
• 2.1.2. 호흡 환기와 관련된 지표 24
• 2.1.3. 산소 해리 곡선의 이해 29
• 2.2. 레이다 기반 생체신호 측정 원리 32
• 2.2.1. FMCW 레이다 신호 모델 32
• 2.2.2. I/Q 데이터와 위상 38
• 2.2.3. PCA 기반 가슴 변위 신호 추출 41
• 2.3. 기존 SpO2 추정 기법 44
• 2.3.1. 주파수 도메인 기반 방법 45
• 2.3.2. 복합 생리학적 모델링 46
• 2.3.3. 기존 기법의 한계 49
• Ⅲ. 연구 방법론 51
• 3.1. 연구 전체 흐름도 52
• 3.2. 실험 환경 및 데이터 수집 56
• 3.2.1. 피험자 구성 및 실험 환경 구축 56
• 3.2.2. 레이다 센서 및 SpO2 센서 장비 구성 60
• 3.2.3. 데이터 동기화 63
• 3.2.4. 정상 및 비정상 호흡 구간 정의 65
• 3.3. 데이터 처리 및 특징 추출 67
• 3.3.1. 원 I/Q 데이터 전처리 67
• 3.3.2. PCA 기반 가슴 변위 신호 도출 68
• 3.3.3. 호흡수 및 심박수 계산 70
• 3.4. 상관관계 분석 방법 72
• 3.4.1. 호흡 구간 기반 분리 73
• 3.4.2. 지연 시간 추정 73
• 3.4.3. SpO2 하강 회복 기울기 산출 74
• 3.5. 경험적 선형 보정 모델 75
• 3.5.1. 선형 보정 개념 및 설계 배경 76
• 3.5.2. 전역 지역 정렬 78
• 3.5.3. 비정상 호흡 기반 평활화 80
• 3.5.4. 선형 블렌딩 83
• Ⅳ. 연구 결과 86
• 4.1. 상관관계 분석 결과 87
• 4.1.1. 지연 시간 분석 87
• 4.1.2. SpO2 변화율 분석 90
• 4.1.3. 통계적 상관 분석 92
• 4.2. 경험적 선형 보정 결과 94
• 4.2.1. 보정 전후 SpO2 파형 비교 96
• 4.2.2. 성능 평가 97
• Ⅴ. 결론 102
• 참고문헌 104