Digital Communications Channel Equalisation Using the Kernel Adaline

For transmission of digital signals over a linear channel with additive white gaussian noise, it has been shown that the optimal symbol decision equaliser is non-linear. The Kernel Adaline algorithm, a non-linear generalisation of Widrow's and Hoff's Adaline, has been shown to be capable of learning arbitrary non-linear decision boundaries, whilst retaining the desirable convergence properties of the linear Adaline. This work investigates the use of the Kernel Adaline as equaliser for such channels. It is shown that the Kernel Adaline performs comparably to the Bayesian optimal equaliser for these channels, and further has something to offer even if the channel noise is non-white.