適用於數位助聽器之低功耗聲學回授消除電路設計與實現

Abstract

一個世紀多以來，助聽器已幫助眾多聽力損失的患者，提升他們日常生活的品質。然而由助聽器喇叭與麥克風間隙造成的聲學回授路徑，形成惱人的「嘶吼」聲，也成為助聽器配戴者最常抱怨的問題之一。嚴重的回授音損害了聲音品質，也使患者對語音的理解程度下降。降低助聽器的放大增益雖可免除回授音(或稱震盪)的發生，但同時卻往往使助聽器無法達到臨床處方所額定的補償增益。因此，助聽器勢必需有回授音消除的功能。拜現今數位信號處理科技之賜，許多研究與演算法相繼提出以壓抑或減低回授來改善問題。目前最先進的回授消除演算法提供良好的消除效果與可用之穩定增益值，且在語音環境中仍然耐用，不過往往因著過高的運算代價以致不適於真實的應用。 出於對要求消除效能以及低功率消耗之二者的考量與權衡，本論文提出一回授消除演算法，以求兼顧效能及低複雜度。本演算法包含以下部分：(1) 近似-正規化最小均方差之適應性濾波器，其運作與傳統之正規化最小均方差濾波器接近，但同時降低大量的運算、(2) 可調整式步階時序規劃器、(3) 語音偵測器整合式之系統控制，可動態控制適應性濾波器的更新與否、(4) 高效率震盪偵測器，用以偵測回授成分的累積。模擬結果顯示本論文提出的演算法可提供20至28分貝的可增穩定增益，並且聲音品質相較其他提出的設計優。最後，本論文將此演算法設計並實現至積體電路，共使用59492閘級數，使用TSMC 65nm GP HVT製程在10MHz與0.5V操作下，得到40.9 uW的低功率消耗表現。

For the past century, hearing aids have helped hearing impaired patients to improve their quality in daily life. However, the annoying howling sound introduced by acoustic feedback path between speaker and microphone of hearing aids becomes one of the most complained problems by hearing aids wearers. Sound quality is degraded and speech intelligibility becomes poor as well. Lowering the hearing aids gain can avoid the occurrence of feedback and oscillation. However, it fails to achieve the target compensation gain which is designated by clinical prescription. As a result, functionality of acoustic feedback cancellation (AFC) is required in hearing aids. Thanks to the technology of digital signal processing, various researches and AFC algorithms are reported to reduce and suppress the feedback problem. State-of-the-art AFC algorithm performs good cancellation results for added stable gain increase and robustness on speech environments. However, the cost of computation is often too high to put into practical application. This thesis presents an AFC algorithm, targeting both good performance and low computation complexity. The proposed algorithm consists of following parts: (1) an NLMS approximation (NLMS-A) adaptive filter structure acting as NLMS adaptive filter with plenty of computation is saved, (2) a variable step-size scheduler, (3) a VAD-combined state controller for dynamically enable the adaptation of adaptive filter, guaranteeing the adaptation is free from disturbing of speech signal, (4) an efficient oscillation detector, detecting the occurrence of accumulation of feedback components. Simulation results indicate that proposed AFC system provides 20 dB to 28 dB ASG, and the sound quality performance is superior to reported design. Finally, proposed AFC is designed and implemented to integrated circuits, with 59492 gate counts, consuming 40.9uW at 10MHz, 0.5V, with TSMC 65nm GP HVT technology.