Abstract-In a decision feedback equalizer (DFE), the structural parameters, including the decision delay, the feedforward filter (FFF), and feedback filter (FBF) lengths, must be carefully chosen, as they greatly influence the performance. Although the FBF length can be set as the channel memory, there is no closed-form expression for the FFF length and decision delay. In this letter, first we analytically show that the two-dimensional search for the optimum FFF length and decision delay can be simplified to a one-dimensional search and then describe a new adaptive DFE where the optimum structural parameters can be self-adapted.Index Terms-Adaptive algorithm, decision delay, decision feedback equalizer (DFE), tap-length.
I. PROBLEM STATEMENTST HE typical structure of a decision feedback equalizer (DFE) consists of a feedforward filter (FFF), a feedback filter (FBF), and a decision device. Although an ideal DFE generally has infinite length [1], most designs use finite length filters because of simplicity and robustness. This brings up the problem of how to choose the structural parameters, including the FFF tap-length , the decision delay , and the FBF tap-length , as they have great influence on the performance. Note that the decision delay determines to which transmitted symbol the detected symbol corresponds. To be specific, on the one hand, when and are fixed, there exists an optimum that minimizes the minimum mean-square error (MMSE); on the other hand, for a given , the MMSE is always a nonincreasing function of or , but too long a or not only unnecessarily increases the complexity with little MMSE improvement but also increases the adaptation noise when the adaptive algorithm is applied. Therefore, for all possible choices of , , and , there must exist a group of optimum values of them that correspond to the best MMSE performance. The purpose of this letter is, therefore, to search for the optimum structural parameters that normally vary with different channels.Some designs, by comparison, intend to reach a preset target MMSE with the simplest complexity, such as the smallest taplength. However, how to choose the target MMSE, thereafter, becomes another issue that is different from that of this letter, where we want the MMSE as small as possible. In some other approaches, the algorithm fixes the tap-length based on a tolerable complexity and finds the corresponding optimum . For example, in [2], fixing the total number of taps " ," Al-Dhahir et al. described a method to search for the optimum values of and that minimize the MMSE. Such an approach, however, has two disadvantages: First, it did not point out how to choose an appropriate value of , which, similar to that for the linear equalizer [3], should balance the complexity and performance and thus generally varies with the channel. Thus, the resulting structural parameters are only "locally" optimum for one particular choice of . Second, although the assumption of fixing attempts to place a constraint on the complexity, two DFEs with the same may vary...