An adaptive blind channel shortening algorithm for MCM systems

In this paper, we propose an adaptive blind channel shortening algorithm for MCM systems such as ADSL. The algorithm is composed of two phases. In the first phase, a stochastic gradient descent algorithm is utilized to search the minimum of the proposed cost function. It is demonstrated that the cost surface is multimodal and not all minima attain good performance. In the second phase of the procedure, genetic algorithms are employed to find the best solution according to a pilot deviation criterion among all solutions to the cost function. The major advantage of the proposed algorithm is, it inherently provides shortened channel information in contrast to similar algorithms in the literature in which channel estimation has to be performed separately after shortening, employing either training sequences or blind channel estimation.