Research On The Dynamic Length Algorithm Of The Decision Feedback Equalizer

In adaptive filtering, the filter parameters obtained in preceding time are used toadjust the filter parameters at present adaptively. The adjustment can adapt theunknown or time-varying statistical property of the signal and the noise, and realizethe optimal filtering. This technology is used in adaptive prediction, the adaptiveinterference canceling, the adaptive system identification and so on. It also widelyused in domains of telecommunication, radar and biomedicine.Linear adaptive filter has low computation complexity and strong stability, andits performance depend largely on the length of the filter. Currently, the research onthe dynamic length algorithm of linear adaptive filter is quite mature, the main severalalgorithms are SG- DLA (Segment Dynamic Length Algorithm), GD-DLA(GradientDescent Dynamic Length Algorithm), FL-DLA(Fractional Length Dynamic LengthAlgorithm), VT-DLA(Varying Threshold Dynamic Length Algorithm). VT-DLA givesa general explanation to all the dynamic length algorithms, by setting the parametersin VT-DLA, it can be translated into other algorithms.The decision feedback equalizer is one kind of universal application non-linearfilter, its performance must surpass the linear filter, its structure is also more complex.The decision feedback equalizer's performance relies on three structural parameters:feedforward filter length, the feedback filter length and the decision delay. As theinfluences of the three parameters on the filter performance are not independent, threedimensional adjust can not be achieved at present. One reasonable solution is tochoose a suitable adjustment order of the three parameters. First, according to thesimilarity between linear filter and feedback filter, a range of the decision delay is set,and the three-dimensional problem is transformed to the two-dimensional problem.Then according to the consistency between feed forward filter length and feedbackfilter length on the adjust direction and amplitude, adjust the two parameterssimultaneously. The two-dimensional problem is further reduced to one-dimensionaltreatment. In this method, the feed-forward length and the length of feedback canconverge to the ideal length and has a good error performance.