本論文基於常見的前饋式(Feedforward)、回饋式(Feedback)與混合式(Hybrid)三種基本的主動噪音控制器設計架構，分別提出改良設計與對應之數學分析，以增強現有的主動噪音控制(Active noise control, ANC)系統設計方法。對於前饋式多通道系統，本論文提出兩種簡化系統運算量的設計，以期提升系統性能。第一種是利用分時多工的概念，將多個參考麥克風輸入訊號重組為單一輸入訊號，第二種方法則是設計以類神經網路為架構的分類器，利用參考麥克風訊號大小來判斷噪音源方向，並自動選取適當的參考麥克風作為系統單一輸入訊號。兩種方法皆能簡化多個參考麥克風輸入訊號的運算量，且可自動選取符合最適當的參考麥克風以達噪音控制效果。針對回饋式主動噪音控制系統設計，本論文則同樣提出兩個提升系統效能之設計，消除窄頻(Narrowband)噪音。首先，利用適應性單頻訊號提升器(Adaptive line enhancer, ALE)結合回饋式主動噪音控制系統，改善傳統窄頻主動噪音控制方法與回饋式系統在實際應用中，非聲學感測器的量測誤差與窄頻頻率過近而減緩收斂速度的問題，本文之系統不需額外的非聲學感測器與訊號產生器，可減少硬體成本與增加系統強健性。本論文另提出ㄧ客製化濾波器設計，結合原回饋式主動噪音控制系統架構，應用於助聽器上，使該系統能根據助聽器使用者的聽力曲線，產生必要之反噪音訊號，以消除因為使用開放式(Open-fit)助聽器所導致的洩漏噪音(Noise leakage)，提升系統效能。本論文也針對基於方程誤差之適應性無限脈衝響應濾波器(Equation-error adaptive infinite impulse response filter)設計主動噪音控制系統，透過演算法之數學分析與推導，證明其等同於混合式主動噪音控制系統架構。所述改善之設計與方法皆透過電腦模擬或實驗，來驗證其理論與實際增強效果。;This dissertation proposes several designs to improve and optimize the existing active noise control (ANC) methods based on three basic ANC structures, the feedforward, feedback and hybrid control schemes, covering the whole spectrum of ANC systems. First, this dissertation proposes two designs to simplify the computational complexity of the multiple-reference/multiple-output (MRMO) ANC system. By applying the concept of time-division multiple-access (TDMA) technique, the multiple-reference signals are organized in one input signal vector but still covers sufficient noise information for the multiple-channel ANC system. The other design is using feedforward neural network (FNN) as a classifier to estimate the noise source direction based on the multiple-reference signals power and automatically select the proper reference microphone closed to the source as the single input. Both methods reduce computational complexity of the multiple-reference ANC system. The infant incubator ANC system is applied to verify the proposed designs. Second, this dissertation proposes two designs to improve the efficiency of conventional feedback ANC (FBANC) to cancel the narrowband noise. The FBANC system integrated with adaptive line enhancers (ALE) overcomes the frequency mismatch problem due to the inappropriate installation or aging of the non-acoustic sensors used in the conventional narrowband ANC (NANC) system. The proposed method avoids using non-acoustic sensor and signal generator, thus reduces the system cost and enhances the robustness. An axial fan is used for the case study. The other design adopts a customized filter for the conventional FBANC system to increase its efficiency for the open-fit hearing aids applications. According to the hearing threshold of the hearing impaired patient, the proposed filter-integrated FBANC system can attenuate the audible noise leakage without generating unnecessary anti-noise. In the last part, this dissertation develops the ANC algorithm based on the equation-error adaptive infinite impulse response (IIR) filter and performs basic theoretical analysis, also discovers this new algorithm is identical to the hybrid ANC system. These designs are verified through computer simulations and real-time experiments.
