Periodic Filters and Their Applications in Blind Channel Identification and Equalization

• Main Authors: Jwo-Yuh Wu, 吳卓諭
• Other Authors: Ching-An Lin
• Format: Others
• Language: en_US
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/48443054462586381640

博士 === 國立交通大學 === 電機與控制工程系 === 91 === We study periodic filters and its inverse design problem, with applications to blind channel identification and equalization. The adopted multi-input multi-output (MIMO) time-invariant representation of periodic filters allows considerable simplification in problem formulation as well as in analysis. We propose a method for constructing an approximate inverse for a given periodic filter in the presence of noise. The criterion is to minimize the asymptotic block mean squared error and this allows a natural formulation of the problem in terms of transfer matrix of the associated time-invariant representation as an optimal model-matching problem. Based on inner-outer factorization we derive a closed-form optimal solution and show that there is a lower bound due to noise on the objective function. When the filters are FIR, we show that the solution can simply be obtained as the least squares solutions of a set of over-determined linear equations. After we establish the inverse design method, we investigate the applications of periodic filters in blind channel identification. We propose a method for blind identification of FIR time-invariant channels based on periodic modulation. The adopted time-domain approach in terms of block signals and MIMO time-invariant channel model is simple compared with existing frequency-domain approach. The method exploits the linear relation between the product channel coefficients and autocorrelation matrix of the block received signal as well as the decoupled nature of the resultant linear systems of equations. The identification equations thus obtained are relatively simple. The time-domain framework also leads to a very natural formulation of the optimal modulating sequence selection problem; the proposed optimal solution minimizes the effects of noise and finite-sample estimation errors on the identified channel. Simulation results show that the proposed identification method yields improved performance as compared with existing frequency-domain subspace methods, and the proposed approximate inverse, when applies to channel equalization, achieves satisfactory equalization performance.