BLINDSFTF stable fast transversal filters algorithm applied to blind equalization

SFTF (stable fast transversal filters), an exact least squares adaptation algorithm with complexity of order 8 times the number of filter coefficients, has been applied to the solution of blind equalization cost functions with dramatic improvements in convergence speed compared to stochastic gradient (lms style) adaptation. This method requires only four times more computation than stochastic gradient methods and has the advantage of being eigenvalue spread independent. The SFTF exact least squares adaptation removes the eigenvalue spread of the channel problem and leaves us now with the speed of convergence depending only on the type of signal used and the initial ISI caused by the channel. Gray´s optimum cost function for the generalized Gaussian pdf is discussed and used in the simulations. In summary, an optimal blind cost function is combined with an optimal filter adaptation method. The topic of entropy is discussed and it´s relation to kurtosis and speed of convergence properties