The main objectives of this thesis are to develop a computationally efficient realization method of adaptive filtering and to study the use of the developed ADF in a practical application.
First, we discuss various realization methods of a class of FIR ADF's including the fast transversal filter (FTF). Then, a computationally efficient frequency bin fast transversal filter (FBFTF) is introduced. By analysis of the computational complexity, we show that a dramatic reduction of multiplications can be achieved by the FBFTF as compared with the FTF. Also, we discuss the instability problem which can arise for an ADF with large order and some particular inputs.
As an application of the FBFTF, we study a noise canceller system. Specifically, the performance of the FBFTF used in a noise canceller is investigated. It is shown that its performance is almost the same as that of the FTF, yet its computational complexity is reduced to one tenth to one thirtieth as compared to the FTF without degrading the performance.
본 논문에서는 최근에 각광을 받고 있는 adaptive digital filter (ADF)들 중 몇가지 형태에 대해서 검토하고, 효과적인 구현 방법 및 실제 응용에 관한 새로운 연구 결과들을 다루고 있다.
먼저, least-mean-square (LMS), frequency-domain LMS (FLMS), 그리고 이것들의 block type algorithm들인 time-domain block LMS (TBLMS)와 frequency-domain block LMS (FBLMS) ADF들을 살펴보고, 또한 block least-squares (BLS) 및 recursive least-squares (RLS) algorithm들을 검토하여 새로운 frequency-bin fast transversal filter (FBFTF) 구조를 제시하였다. 즉 fast Fourier transform (FFT)과 데이터의 적절한 sectioning 방법에 의해서 효율성을 상당히 개선시킬 수 있음을 보였다.
앞의 효율성을 뒷받침하기 위하여 computer simulation이 행해졌는데, 이때 사용된 system은 adaptive noise canceller로써, 일반 고정 filter들을 불가능한 경우에도 ADF는 뛰어난 성능을 발휘하게 된다. 여러가지 신호와 환경에 대해 simulation 해 본 결과, FBFTF는 FTF 방법과 동일한 성능을 얻는 반면 계산의 복잡도는 $\frac{1}{10}-\frac{1}{30}$ 로 줄일 수 있음을 보였다.
