| 摘要: | 本研究旨在建立一套以 Wi-Fi 通道狀態資訊（Channel State Information, CSI）為核心的非接觸式動態生理訊號監測系統，透過深入的理論推導與臨床級驗證，實現對呼吸與心跳等微小人體運動的即時估測。隨著睡眠障礙與慢性呼吸疾病在全球日益普遍，現有的穿戴式或接觸式監測裝置常因侵入性高、不適合長期使用而受限。因此，本研究提出一種利用環境中即可取得的 Wi-Fi 訊號進行生理監測的方法，期望提供高隱私、低成本且可長期部署的健康監測解決方案。
研究核心基於 Wi-Fi 的 OFDM 與 MIMO 架構，從實體層(Physical Layer)提取 CSI，再以 Ray Tracing 模型描述訊號多路徑與動態反射特性。為使 CSI 相位資訊可用於解析胸腔與腹腔因呼吸造成的毫米級位移，本研究提出一套包含 Ratio Model 相位誤差消除、主成分分析（PCA）訊號強化、DC compensation 圓擬合校正及 Gabor Transform 時頻分析的完整演算法流程。此流程能有效從 CSI 中分離動態成分，估測瞬時頻率，再轉換為實際位移與速度資訊，並具備在低速、小振幅運動下仍保持高靈敏度的能力。
為驗證系統效能，本研究設計三類實驗：線性滑軌週期運動模擬呼吸、極低速位移測試以評估演算法極限，以及人體實驗以觀察六類典型呼吸波型（Eupnea、Biot、Bradypnea、Sighing、Tachypnea、Kussmaul）。所有實驗均以高精度雷射測距模組作為對照組。結果顯示，本研究提出的方法在毫米至公分尺度範圍內皆能準確重建呼吸波型，並於頻率軌跡上呈現良好的一致性與抗雜訊能力。
;This study aims to develop a non-contact dynamic physiological monitoring system based on Wi-Fi Channel State Information (CSI), capable of estimating subtle human motions—such as respiration and heartbeat—in real time through rigorous theoretical modeling and clinical-grade validation. As sleep disorders and chronic respiratory diseases continue to rise globally, existing wearable or contact-based monitoring devices often suffer from intrusiveness and poor suitability for long-term use. Therefore, this work proposes a Wi-Fi–based physiological sensing method that leverages ambient wireless signals, offering a high-privacy, low-cost, and long-term deployable solution for continuous health monitoring.
The core of this research is built upon Wi-Fi OFDM and MIMO architectures, extracting CSI from the physical layer and modeling multipath propagation and dynamic reflections using a Ray Tracing framework. To enable CSI phase information to resolve millimeter-scale thoracic and abdominal motions induced by respiration, this study introduces a complete signal-processing pipeline that includes a Ratio Model for phase error elimination, Principal Component Analysis (PCA) for dynamic component enhancement, DC compensation via circular fitting, and Gabor Transform–based time-frequency analysis. This pipeline effectively isolates motion-induced variations within CSI, estimates instantaneous frequency, and converts it into displacement and velocity with high sensitivity, even under low-speed and small-amplitude movements.
To validate system performance, three categories of experiments were conducted: linear-guide periodic motion to simulate breathing, ultra-low-speed displacement tests to evaluate algorithmic limits, and human-subject experiments featuring six representative respiratory patterns (Eupnea, Biot, Bradypnea, Sighing, Tachypnea, and Kussmaul). All measurements were benchmarked against a high-precision laser distance sensor. The results demonstrate that the proposed method can accurately reconstruct respiratory waveforms across millimeter-to-centimeter scales, maintaining strong consistency in frequency tracking while exhibiting robust noise resilience. |
