力學脈搏感測器的運動偽影減少方法之研究

Abstract

在非侵入式生理監測裝置中，連續血壓（Continuous Non-invasive Blood Pressure, CNBP）與心率量測技術已廣泛應用於臨床與穿戴式照護領域。其中，力學式脈搏感測器相較於傳統光學感測（PPG）具備抗光源干擾與適用範圍更廣等優勢。然而，於實際應用過程中，裝置容易受到使用者移動、姿勢改變與外力觸碰等動作造成的運動偽影（Motion Artifact）影響，進而降低量測準確度。 本研究針對機械式脈搏波感測器所面臨之運動偽影問題，提出一套模擬資料建立、自適應濾波訓練與實際驗證之系統性處理流程。首先，依據臨床情境建構三種常見干擾類型：震動型高頻干擾、基線飄移與低頻滑移，並透過頻譜分析掌握其特性。接著，以乾淨的脈搏波信號與合成干擾相加，建立具備「真實答案」之訓練資料，供 Recursive Least Squares（RLS）自適應濾波器進行訓練與參數優化。 實驗結果顯示，於模擬情境下濾波後之 RMSE 平均改善 42%，SNR 則提升 6.1 dB，尤以震動干擾處理成效最佳。進一步於實際干擾情境下測試五位受測者之資料，在三種動作干擾（手掌晃動、手部輕拍、指夾施壓）下，濾波後之心率與血壓參數均明顯穩定，心率誤差由 ±10.2 bpm 降至 ±2.9 bpm，血壓平均誤差減少達 75%。 本研究證實 RLS 濾波器搭配模擬訓練資料可有效處理機械式脈搏感測器之運動偽影問題，並提升量測參數之準確性與穩定性，為連續血壓監測於手術或高動態情境下的應用提供實用可行之解法。

Continuous non-invasive blood pressure (CNBP) and heart rate monitoring technologies are essential in clinical and wearable healthcare systems. Compared to optical sensors like photoplethysmography (PPG), mechanical pulse sensors offer reduced sensitivity to ambient light and better robustness. However, they remain highly susceptible to motion artifacts caused by movement, posture shifts, or mechanical contact, which degrade measurement accuracy. This study presents a framework to suppress motion artifacts in mechanical pulse wave signals via synthetic data generation, adaptive filtering, and performance validation. Three typical artifact types—high-frequency vibration, baseline drift, and low-frequency shifts—were modeled from clinical scenarios. Their frequency characteristics were analyzed, and synthetic noisy signals were created by blending clean waveforms with artifacts. These were used to train a Recursive Least Squares (RLS) adaptive filter. Simulated results showed a 42% RMSE reduction and a 6.1 dB SNR gain, with optimal performance in high-frequency interference. Real data validation with five participants under three motion conditions (hand shaking, tapping, and sensor displacement) showed significant improvements: heart rate error dropped from ±10.2 bpm to ±2.9 bpm, and blood pressure estimation error was reduced by up to 75%. The findings confirm that the proposed RLS filter, trained with synthetic artifact-contaminated data, effectively mitigates motion artifacts in mechanical pulse sensors, enhancing CNBP and heart rate reliability in dynamic clinical environments.