독립성분 분석 적응필터를 이용한 헤드폰 유입잡음 제거 연구
Adaptive Noise Cancellation using independent component analysis for headphones application
- 주제(키워드) 도움말 noise cancellation , independent component analysis , 독립성분분석
- 발행기관 서강대학교 일반대학원
- 지도교수 박형민
- 발행년도 2013
- 학위수여년월 2013. 8
- 학위명 석사
- 학과 및 전공 도움말 일반대학원 전자공학과
- 실제URI http://www.dcollection.net/handler/sogang/000000052641
- 본문언어 한국어
- 저작권 서강대학교 논문은 저작권 보호를 받습니다.
초록/요약 도움말
본 논문에서는 헤드폰 내부로 유입되는 외부잡음에 대한 개선된 제거 방법을 소개한다. 독립성분분석을 이용한 적응잡음제거에 있어, 필터 출력 신호에 대해 일반감마분포 모델에 의한 유연한 점수함수(flexible score function)를 적용하고 일반감마분포 파라미터들에 대해 적응필터의 목적함수를 최대화하도록 최급강하법에 따라 갱신하여 성능이 개선될 수 있음을 보였다. 또한 외부잡음이 바람과 함께 마이크로폰으로 유입될 때 발생하는 문제를 다루었고 이를 해결할 수 있는 방법으로 잡음제거 적응필터에 독립성분분석에 의한 피드백 구조를 적용했다. 다양한 환경의 잡음에 대한 컴퓨터 모의 실험을 통해, 제안하는 방법이 문제를 해결하고 기존 방법들에 대해 더 좋은 잡음 제거 성능을 보임을 확인했다.
more초록/요약 도움말
This thesis presents noise cancel method for headphones application using adaptive filtering algorithm. Today’s the most popular algorithm for headphones is NLMS (Normalized Least Mean Square), but it only remove 1st and 2nd statistics between noise and source signal to play. Therefore, it cannot remove impulsive or intensive noise effectively. Thus ICA(Independent Component Analysis)-based adaptive filter is applied that effectively separates signals by maximizing their independence which can be obtained with information maximization between filter out signals, called ‘info-max’. However, ICA utilize fixed nonlinear score functions likes ‘tanh()’ or ‘sign()’ which are interpreted as the CDF of independent source signals. So, in this paper flexible score function, particularly modeled using ‘generalized gamma distribution’ which can reflects scale and shape characteristics of signals. Another practical problem of adaptive ANC(Active Noise Cancellation) is wind noise. Every microphone suffers from wind induced on its membrane which is very sensitive to air pressure or air shock. On the other hand, wind cannot pass through the headphones cup material which acts as the passive noise shield. Consequently, the reference input signal for FIR-based adaptive filter is corrupted by wind. The actual reference signal needed for filter coefficient updating is only external sound noise which is the target signal to cancel. Due to the contaminated reference signal filter updating does not work properly, This problem can be solved by feedback adaptive filter which generates its reference signal form the error signal that is the filter output. This paper suggests the proper way to solve aforementioned two practical issues by using feed-forward and feed-back adaptive filter based on ICA and with intensive computer simulations. Its results show the presented method can solve these problems.
more