Performance of Generalized Regression Neural Network-based channel estimation in Vectored DSL systems

Abstract

It is well-known that Vectored Digital Subscriber Line (DSL) transmission promises significant theoretical data-rate increases for DSL technology; however, Vectored DSL requires full knowledge of the channel. The effectiveness of Vectored DSL transmission in a practical setting, where channel knowledge is subject to error, has yet to be determined. This paper proposes a Generalized Regression Neural Network (GRNN)-based approach to DSL channel estimation by interpolating between a subset of measured or estimated data-points. Furthermore, closed-form expressions for the effect of channel estimation error on he achievable Vectored DSL data-rate are derived, using a Zero-Forcing (ZF) interference canceller for upstream transmission and a Diagonalizing Pre-coder (DP) for downstream transmission. Finally, simulation results are provided to demonstrate the performance loss associated with channel estimation error for Vectored DSL transmission, based on the ANN approach and a linear regression approach.