IEEE International Workshop on Intelligent Signal Processing and CommunicationSystems (ISPACS94) Held at Seoul (Korea) on 5-7 October 1994. Abstracts

摘要

The purpose of this paper is to design robust equalizers which have an ability tosuppress dependent impulse noise and nongaussian noise based on training information. The algorithm proposed here is based on piecewise approximations to a regression function involving only quantiles and partition moments which can be estimated by Robins-Monro Stochastic Approximation (RMSA) algorithm. For the purpose of testing the robustness, we adopted mixture distribution models, e.g. e-contaminated Gaussian distribution introduced by Tukey and Huber as noise models. The performance of the robust MMSE equalizers in satellite channels is considered for a wide range of SNR and various forms of non. gaussian noise. The results are compared to a single sample detector and conventional equalizers, in terms of error rate and MMSE. The superiority of the proposed robust equalizers is clearly demonstrated. Due to extreme difficulty in calculating PE and MMSE, Monte-Carlo simulations were selected as means of comparison.