Neural Sequence Detector for Digital Equalization

In this paper a new approach to the equalization of digital transmission channels is introduced and described. The proposed solution makes use of a fast neural architecture, coupled with a innovative error functional, and is able to perform the equalization task in a Viterbi-like fashion applied to a Decision Feedback architecture for the purpose of improving the resitance to imperfect knowledge of the channel and interference. Performance comparisons with standard techniques for different channels demonstrate the validity of the proposed approach, expecially when the data model departs from assumptions and the computational cost is a critical issue.