Research Of Adjustment Theory Based On Extended Correlation Least Mean Square Algorithm

With the development of technology, the signal processing has become one of the hot research fields today. Many communication methods appeared, the perceptual quality has been gained more and more attention. And the echo and noise still remains as a big problem which seriously effects communication quality especially when high-delay. To solve this problem, many adaptive filtering were used such as Least Mean Square algorithm (LMS). The LMS algorithm, due to its simple structure, low computational complexity and easy to realize in the real-time processing, has got very fast development. But under double-talk condition, LMS algorithm can not work normally.The adaptive filter based on the correlation function is designed for double-talk condition specially. However, the existed algorithms are also problematic. In this article describes two related algorithms:the Correlation Least Mean Square algorithm (CLMS) and the Extended Correlation Least Mean Square algorithm (ECLMS). With the experiment data can be seen under the double-talk condition, these algorithm work correctly. But algorithm still has defects, such as the error signal to the convergence rate is not high, can not processing long echo impulse response and so on. To solve these problems, the main achievements of this work consist of following aspects:(1) Base on the ECLMS, the tap variable algorithm is presented, which is called Tap-Length variable ECLMS(TECLMS).According to the characteristics of the taps, compare the new and old error signal to select the appropriate value, makes the number of taps from big to small gradually changes, thereby reducing the computational complexity of the algorithm.(2)The step size variable algorithm is presented, which is called Step-Size variable ECLMS (SECLMS). Because the exponential function is used as the step size adjustment formula so that step size can be controled by the error signal. Form the results can be seen, SECLMS better than ECLMS. it can and can better apply to the double-talk condition.