적응보상기를 채용한 적응잡음제거기 및 그 적응알고리즘

손경식 慶北大學校 大學院 1991 국내박사

During a signal is being transmitted, and interference signal can be introduced through and unknown channel. In this case, and adaptive system, so called adaptive noise canceller, can restore the original signal from the corrupted signal by first indentifying the unknown interference channel on the minimum mean square error criterion, and then by cancelling the interference signal using the identified interference channel. While this method is quite effective when a priori knowledges about the characteristics of the interference signal and of the interference channel are unknown or time varying, it has a drawback that the presence of the estimation error which converges to the original signal makes the parameter value of the identified interference channel fluctuate around the optimum value. To alleviate the above-mentioned drawback and restore the original signal more accurately, a new adaptive noise canceller is presented in this thesis, which employs an adaptive compensator at the output of the usual adaptive noise canceller to filter out the original signal component from the estimation error. The new adaptive noise canceller with an adaptive compensator is given both in time-domain and frequency-domain. Compensated least mean square(CLMS) algorithm and compensated frequency-domain least mean square(CFLMS) algorithm are developed for the proposed adaptive noise canceller. The CFLMS algorithm is based on the "overlap-save" technique used in frequency-domain filtering. Frequency-domain noise canceller has less computational load and faster convergence rate than the time-domain counterpart. For the number of taps which is more than 32, the proposed CFLMS algorithm offers a significant reduction in computation compared to the conventional LMS algorithm and the CLMS algorithm. Also, for a large number of taps(≥512), the CFLMS algorithm and the FLMS algorithm are almost same in computational effort. The performance of the proposed algorithm is illustrate by computer simulations. The CLMS and CFLMS algorithm drive the parameter value of the identified interference channel much closer to the optimum value than the conventional LMS and FLMS algorithm for both of the uncorrelated and correlated inputs. Although all of the algorithm have almost the same convergence rate for the uncorrelated input, the CFLMS algorithm converges slightly faster than the CLMS algorithm for the correlated input.

적응잡음제거기를 위한 적응알고리즘의 성능비교 및 성능향상에 관한 연구

배종갑 부산대학교 대학원 1998 국내박사

The steady-state excess mean-squared error(mse), a useful measure of steady-state performance, in the adaptive noise canceller reflects the error due to the ongoing adaptive process, that is, the fluctuation of the weights about their optimal values according to the target signal after they have converged in the mean. When the step-size parameter. However, it is not possible to make it arbitrarily small by reducing the step-size parameter, because the convergence time of the LMS-type algorithm is inversely proportional to the step-size parameter. In this paper, the excess mse and the convergence time of NLMS, Sum-LMS and Sign algorithms when applied to the adaptive noise canceller are theoretically analysed and compared respectively. The results of theoretical analysis are verified by computer simulation. In this paper, also NLMS-SA algorithm is proposed which uses the combination of NLMS and Sign algorithm applys in of ANC and NLMS algorithm applys in silence region and adaptive transient region. The speech region is detected by output speech of ANC and NLMS algorithm applys in silence region and adaptive transient region. The speech region is deteched by voice detector. The voice detector uses property of adaptive prediction error filter which can filter the highly correlated speech. To detect speech region, estimation error which is the output of the adaptive filter is applied to the voice detector. So the region of speech is detected by the ration of input and output energy of voice detector. The system mismatch, transient convergence and output SNR of the above-mentioned three algorithm, proposed NLMS -SA combination algorithm and CLMS algorithm are compared by computer simulation. As result of several computer simulations, clearly, in steady state there is very good agreement between the simulations and analytical results. From these comparisons, the Sum-LMS algorithm shows the most excellent performance when the input power level is comparatively low, and the NLMS-SA combination algorithm shows the most excellent performance when the input power level is comparatively high. As results of computer simulation, when the input power is 500, the NLMS-SA algorithm when compared with the Sum-LMS algorithm has improved output SNR and segmental SNR up to about 10 dB, 5 dB, respectively.

적응잡음제거기에서 SUM-LMS 알고리즘과 LMS-SA결합 알고리즘의 성능비교

김창기 부산대학교 대학원 1997 국내석사

A conventional adaptive noise canceller(ANC) using LMS algorithm suffers from the misadjustment of adaptive filter weights due to the gradient-estimate noise by input speech signal at steady state. In this paper, an ANC is proposed which uses the combination of VSLMS(variable step size LMS) and SA(sign algorithm) to improve steady state performance and convergence rate. SA algorithms is applied to in speech resign to prevent the weights from perturbing by output speech of ANC and VSLMS algorithm is applied to improve convergence rate and channel tracking ability in silence region and adaptive transient region. In computer simulation, the performance of the proposed VSLMS-SA combination algorithm is much better than NLMS algorithm and the algorithm., recently proposed by Greenberg[5], with adaptation step-size parameter determined by sum method in convergence rate, channel tracking and steady state performance.

적응잡음제거기에서 퍼지 논리를 이용한 수렴율에 관한 연구

유경종 금오공과대학교 1995 국내석사

백프로퍼게이션의 신경망을 이용한 적응잡음제거기 구현

추형석 蔚山大學校 大學院 1998 국내석사

Design of robust and efficient event-synchronous adaptive filter for real-time biomedical signal analysis

박호동 Graduate School, Yonsei University 2011 국내박사

자동차 환경에서 실시간 적응 잡음 제거기의 성능에 관한 연구

박재훈 대구대학교 대학원 2007 국내석사

적응신호처리 분야에서 steepest descent 방식인 LMS(Least Mean Square) 알고리즘은 식이 간단하고 계산 량이 비교적 적기 때문에 널리 사용되고 있다. 그러나 LMS 알고리즘은 수렴속도를 높이게 되면 주위 환경이 급격하게 변하는 경우 발산하게 되고, 안정성을 높이게 되면 수렴속도가 떨어지는 수렴속도와 안정성 사이에 trade-off가 발생한다. 이러한 LMS 알고리즘의 단점을 개선하기 위해 NLMS(Normalized LMS) 알고리즘, VSS LMS(Variable Step Size LMS) 알고리즘 등이 개발되었다. NLMS 알고리즘은 안정성을 크게 향상시켰으나 고정된 스텝 사이즈를 사용함으로써 변화하는 환경에 효율적으로 적응하지 못하는 단점이 있고, VSS LMS 알고리즘은 에러 신호를 이용하여 스텝사이즈를 조절하여 수렴속도와 성능을 동시에 향상 시켰으나 에러 신호만 이용하므로 자동차 환경과 같이 주위 환경이 변하는 경우 시뮬레이션을 통해 성능을 측정해 본 결과 완벽하게 적응하지 못하는 단점이 있었다. 본 논문에서는 자동차 잡음 환경에서 기존의 알고리즘 보다 개선된 성능을 얻기 위해 적응 필터의 출력단과 에러 신호 출력단의 신호대 잡음비를 이용하여 스텝 사이즈를 조절하여 시간에 따라 변하는 환경에 효율적으로 적응하는 adjusted step size NLMS 알고리즘을 제안하였다. 그리고 제안한 알고리즘을 적응잡음제거기(Adaptive Noise Canceler)에 적용하여 자동차 내에서 실시간으로 시뮬레이션을 하였으며, 각 알고리즘들의 성능을 비교·분석하여 제안한 알고리즘의 성능이 우수함을 입증하였다. LMS (Least Mean Square) algorithm using steepest descent way in adaptive signal processing requires simple equation and is used widely because of the less complexity. But LMS algorithm has some drawbacks. If a convergence speed increase, LMS algorithm has a divergence in case of sharp environment changes. And if a stability increase, the convergence speed becomes slow. This algorithm based on a trade off between fast convergence and system stability. To improve this problem, NLMS (Normalized LMS) algorithm and VSS LMS (Variable Step Size LMS) algorithm was developed. Normalized LMS algorithm improved system stability. But it has disadvantage that don't adapt efficiently in environment changes because of using fixed step size. And VSS LMS algorithm improved the convergence speed and performance as adjusting step size using error signal. But when we measured the performance through simulation, VSS LMS algorithm didn't adapt completely in case of environment changes because of using an only error signal. In this paper, an adjusted step size NLMS algorithm is proposed for improvements the performance of adaptive noise canceller in automobile environments. This algorithm adjusts the step size to adapt efficiently in environments changes continuously. The step size adjustment is controlled by signal to noise ratio (SNR), which is produced by the ratio of the average power of the error signal to the average power of the filter output. The proposed algorithm is applied to adaptive noise canceller in automobile environment and we did simulation by real time. Then compared performance of each algorithm and demonstrated that proposed algorithm has superior performance.

디지탈 신호처리기를 이용한 적응 잡음 제거기에 관한 연구

오창길 광운대학교 대학원 1990 국내석사

본 논문에서는 Analog Devices사의 ADSP-2100A 디지털 신호 처리기를 이용하여 실시간처 리 적응잡음 제거기를 구현한 것이다. 실시간처리 적응잡음 제거기는 8bit A/D 컨버터(ADC0804)와 12bit D/A 컨버터(AD667)로 구 성되는 입출력부와 ADSP-2100A를 사용하는 중앙 처리부로 이루어진다. 이 적응잡음 제거기의 신호 샘플링 주파수는 5kHz이내의 범위를 사용하였으며, 중앙 처 리부의 입력단에는 wait-state 발생회로를 포함하는 A/D 인터페이스 회로와 연결된다. 그리고 신호 처리기에 대한 실행프로그램은 ADSP ?2100/2100A 의 어셈블리어로 작성하였 으며 사용된 적응필터 구조는 TDL)Tapped ?Delay ?Line)형태로서 적응필터의 알고리즘은 LMS (Least Mean Square)을 사용하였다. 본 논문에서 구현한 실시간 적응잡음 제거기는 시뮬레이션 결과와 실시간 실험결과를 비 교 검증함으로써 그 성능을 평가하였다.

개선된 OBAI 기법을 이용한 적응 잡음 제거기에 관한 연구

이태표 成均館大學校 大學院 1993 국내석사

공장 소음 환경에서 음성 향상을 위한

김규동 경북대학교 일반대학원 2007 국내석사

General adaptive noise canceller(ANC) structure needs the reference signal which is separated from the input source signal for example, speech signal. However, in places like noisy environment of factory, the separated reference signal is hard to be acquired. So, we propose that considering that the factory noise is stationary and when people have a conversation, the factory noise does not change, the controller stops weights updating of adaptive filter in ANC when speech signal is inputted. Then the transversal filter which remains weights as coefficients filters out speech signal in speech signal mixed with noise and the loss of speech signal by stopped coefficients of transversal filter is restored by restoration filter which is inverse of stopped coefficients of transversal filter. We demonstrated that the proposed adaptive noise canceller by controlling weights effectively reduced factory noise and enhanced speech signal.