Blind linear and decision feedback equalizers using fourth-order moments and their performance on twisted pair lines

High bit-rate data transmission over severely distorting channels with time varying transmission characteristics like copper lines in the local loop requires adaptive channel equalization. Usual equalization techniques are linear and decision feed back equalization. The adaption is in general based on the decisions of a slicer or tentative decisions of a Viterbi decoder (data-aided equalization). Evidently, data-aided equalization schemes need a fairly good initial setup of equalizer coefficients to ensure that most of the (tentatively) decoded symbols are correct. Blind equalization algorithms allow estimating the equalizer coefficients without any knowledge about the channel or the data sent. In the paper, a stochastic equalization criterion based on the kurtosis, i.e. a shape factor of the signal's probability distribution function, is introduced. Using this criterion and applying a well-known stochastic gradient algorithm, the blind deconvolution algorithm of Shalvi and Weinstein [1990] for linear equalizers can be deduced. The authors generalize this approach to decision feedback equalizers. To combat residual stationary noise due to statistical uncertainty they propose a mixed blind/data-aided equalizer structure. Simulation results are given for linear and decision feedback equalization using pure stochastic as well as mixed algorithms.