- 學位論文
以主動噪音消除方式建立較佳睡眠環境
Building a better sleeping environment using active noise cancellation
摘要
良好的睡眠品質能提升工作及學習效率，且能預防心臟病及高血壓的發生，而影響睡眠品質的最大原因可歸咎於枕邊人的鼾聲及家用電器所發出的噪音，因此本研究期望發展一套自適性噪音消除系統，針對睡眠者耳朵附近做一降噪的動作以提升睡眠品質。 而一般自適性演算法系統架構的優點是單純且演算速度快，因此常被使用在即時系統上，但缺點在於多頻帶訊號的效能有限。本研究利用小波轉換可拆解及重建的特性，將噪音訊號先拆解再經由基本自適性演算法得到降噪訊號資訊，最後再使用小波重建得到降噪訊號。以此發展一套主動式即時噪音消除系統，提升系統對多頻帶訊號的效能。 本研究使用TMS320VC5510單晶片為運算核心，以組合語言及Q28定點數架構建立整個演算流程，並搭配泰菱公司的DSL-330噪音計及羅技公司的R-5主動式揚聲器，針對人耳最敏感的200Hz~1.2kHz訊號作量測。 本研究測試可分為Matlab模擬、驗證及實際聲場測試三部分，測試訊號可分為單頻、多頻及鼾聲訊號。在模擬結果中，本研究對基本LMS演算法及具有小波轉換的系統作一效能比較，確定本研究系統架構的可行性，並於驗證過程中，確定Q28定點數與浮點數運算的誤差可被忽略，在實際聲場測試中，系統受到環境影響，於特定單一頻率可降低15~20dB左右的噪音，特定多頻帶訊號則可降低6~12dB，本研究系統對實際鼾聲訊號表現並不如預期，但仍可衰減1~2dB左右。 本研究透過穩態小波及基本LMS達到與其它研究相似的效果，但仍無法完全忽略聲場迴授及硬體演算的精確度對系統的影響，因此往後的研究可朝向簡化系統演算、提升硬體精確度以提升系統收斂速度。
並列摘要
A Good quality sleep can increase the efficiency of work and study. It also can prevent cardiovascular disease and hypertension. Two main reasons that affect sleep quality are snore from the person sleeping nearby and noise from the household appliances. This study purposes to make an adaptive noise cancellation system that can reduce the noise around the sleeper so that it can increase the quality of sleep. The advantages of the common adaptive noise cancellation system are simple and fast calculation. Thus, it is often used in real-time system. However, its drawback is the limited efficiency of broadband signal. This study uses the characteristic of the wavelet transform that can split and reconstruct signals to separate the noise signal into different scales and then uses basic adaptive algorithm to calculate the de-noise signals of each scales. Finally, the wavelet is used to reconstruct the de-noise signal. The proposed method increases the efficiency of the broadband signal. In this study, TMS320VC5510 DSP chip is used as the calculation center, the software is realized using the assembler and the Q28 fix-point. The system is accomplished using the DSL-330 Sound Level Meter from the Tecpel Company and the loudspeaker from Loitec Company. The measurements are done for signals from 200Hz to 1.2Hz, which is the most sensitive signal to human ear. The study can be divided into three parts: Matlab simulation, hardware testing and testing with real signals. The testing signals can be divided as narrowband, broadband and snore signal. The results of simulation demonstrate the practicability of the proposed system after comparison between the efficiency between basic LMS algorithm and the proposed system using wavelet transform. It is found that the error using the Q28 fix-point format and the float point can be ignored. In the testing of real sound, the system performance is affected by the environment. It can attenuate noise by 15~20dB for a specific narrowband noise and attenuate 6~12dB for a specific broadband signal. The effect of proposed system on a real snore is less than what is expected. But the system can still obtain 1~2dB attenuation. In this study, comparable result is obtained using stationary wavelet transform and basic LMS. However, the sound field effect in real environment and the hardware accuracy still cannot be ignored. Thus, system simplification and hardware accuracy improvement to increase the convergence speed will be the research in the future.