Adaptive kernel-based equalization for non-stationary digital communications channels

It is well known that the optimal symbol decision equalizer for noisy digital transmission channels is generally non-linear. A previous work demonstrated the performance of the so-called Kernel Adaline non-linear classifier in application to these channels, whilst pointing out that the Kernel Adaline is not an adaptive machine and thus would suffer in use on non-stationary channels. Here, we review Kernel-LMS, an adaptive extension of the Kernel Adaline, and introduce Fast Kernel-LMS, a low-complexity approximation to the same. The algorithm is augmented to implement decision feedback, and the performance of the fast algorithm in solving some synthetic non-stationary channel equalization problems is then investigated.