A radial basis function (RBF) neural network, combined with a channel
estimator, is used for fast time-varying channel equalisation. A new m
ethod for calculating RBF centres and weights is proposed to reduce th
e number of hidden units in the high-order RBF equaliser. A Rayleigh f
ading channel model and 10 million random symbols are used in the simu
lation studies. Very good equalisation performance is achieved by usin
g a high-order RBF network with only eight centres.