Recently much attention has been given to the design of timedomain equalizers or TEQs for discrete multitone modulation (DMT) systems. In this paper, we present a new method for the design of such equalizers which minimizes both the intersymbol interference (ISI) and noise power observed in a DMT system. Furthermore, we show how this method can be used for the design of fractionally spaced equalizers or FSEs. Experimental results are presented showing that our design method performs better than other known techniques in terms of achievable bit rate.
ChannelEqualizerFig. 1. SIMO-MIS0 channel and equalizer model.
THE TEQ DESIGIN PROBLEM 1. INTRODUCTIONIn recent years, one problem which has been of great interest has been the design of time-domain equalizers or TEQs for discrete multitone modulation or DMT systems [ 1, 4, 61. Due to the long impulse response of typical channels encountered in DMT systems such as twisted pair telephone lines [7], TEQs are needed to shorten the overall channel response to one sample more than the length of the cyclic prefix used.Several methods previously proposed for the design of such TEQs involve the design of the effective channel and not the equalizer coefficients directly [l, 41. With these methods, the equalizer coefficients must then be chosen to best fit the desired optimal effective channel. A new method, however, was recently introduced [6] which deals directly with the equalizer coefficients. In this method, the objective was to minimize the delay spread of the overall channel. The optimal equalizer coefficients were found to be related to an eigenvector of a particular matrix. In [8], we generlized this method to also account for the noise present in the system. The equalizer coefficients were chosen to minimize an objective function consisting of a convex combination of a channel shortening objective and a noise-to-signal ratio objective.In this paper, we consider minimizing a different objective function consisting of a weighted sum of the intersymbol interference (ISI) power and the output noise power. Much like the methods of [6, 81, the optimal equalizer coefficients will also be found to be related to an eigenvector of a particular matrix. Furthermore, we will show that our results can be extended for the design of fractionally spaced equalizers or FSEs. Though FSEs have not been traditionally used as TEQs for DMT systems, the results shown here give merit to their possible future use. Denote the impulse responses of C ( z ) and H(z) by c(n) and h(n), respectively. The effective channel is ~r ( n )and has length L, + Le -1. Here, the output y(n) is of the form,where z f ( n ) and q(n) are, respectively, the filtered input signal and output noise sequences given by the following.We wish to choose H(z) to shorten the effective channel to a length Ld < L,. In other words, we wish to choose H(z) such that most of the substance of cer(n:) resides in a window W A 5[A, A + Ld -11, where A represents the delay of the desired shortened channel. Here we must have 0 5...