Equalization and the impulsive MOOSE: fast adaptive signal recovery in very heavy tailed noise

Eugene Dubossarsky, Thomas R. Osborn, Sam Reisenfeld

Research output: Chapter in Book/Report/Conference proceedingConference proceeding contributionpeer-review

Abstract

We combine Weighted Order Statistic (WOS) filters as efficient reconstructors of signals with extremely impulsive noise with the deconvolution capabilities of artificial neurons in a fast, memory efficient way. The Multi-rate Optimising Order Statistics Equalizer (MOOSE) neural architecture combines a FIR WOS Hybrid (FWH) pre-filter and a neural deconvolver with differing sampling rates. These cooperate synergistically after training. In the simulated equalization of a BAM channel with high ISI and additive noise. with Gaussian and frequent, high amplitude impulsive components. the linear MOOSE proved far superior to linear filtering. MOOSE computational complexity is far below that of the comparably performing L1 filter.

Original languageEnglish
Title of host publicationProceedings of the international workshop on applications of neural networks to telecommunications 3
EditorsJoshua Alspector, Rodney Goodman, Timothy X Brown
Place of PublicationMahway, NJ
PublisherLawrence Erlbaum Associates
Pages232-240
Number of pages9
ISBN (Print)0805829008
Publication statusPublished - 1997
Externally publishedYes
Event3rd International Workshop on the Application of Neural Networks to Telecommunications (IWANNT 97) - MELBOURNE, Australia
Duration: 9 Jun 199711 Jun 1997

Publication series

NameINTERNATIONAL NEURAL NETWORKS SOCIETY SERIES
PublisherLAWRENCE ERLBAUM ASSOC PUBL
Volume3

Conference

Conference3rd International Workshop on the Application of Neural Networks to Telecommunications (IWANNT 97)
CountryAustralia
CityMELBOURNE
Period9/06/9711/06/97

Cite this