A continuous time structure for filtering and prediction using Hopfield Neural Networks

Abstract

Transversal FIR adaptive filters have been widely used in echo and noise cancelers systems, in equalization of communication channels, in speech coders and predictive deconvolution systems in seismic exploration etc., almost all of these systems implemented in a digital way. This is because with the advance of digital technology it became possible to implement sophisticated and efficient adaptive filter algorithms. However, even with the great advance of the digital technology, the transversal FIR adaptive filters still present several limitations when required to handle in real time frequencies higher than those in the audio range, or when a relative large number of taps are required. To avoid the limitations of digital FIR adaptive filters, several different structures have been proposed. Among them, the analog filters appears to be e desirable alternative to transversal FIR digital adaptive filters because they have the ability to handle very high frequencies, and their size and power requirements are potentially much smaller than their digital counterparts. This paper propose a continuous time transversal adaptive filter structure whose coefficients are estimated in a continuous time way by using an artificial Hopfield Neural Network. Simulation results using the proposed structure in cancellation, prediction and equalization configurations are given to show the desirable features of the proposed structure.KeywordsAdaptive FilterRecursive Less SquareHopfield Neural NetworkConvergence PerformanceUnknown SystemThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.