Application of Richardson's nonstationary iterative procedure to the design of adaptive equalizers

This paper considers the convergence properties and numerical behavior of an adaptive equalization algorithm. Particular attention is paid to the behavior of the algorithm when the computations are carried out with limited precision arithmetic as would be the case in a real implementation. It is shown that a previous conjecture regarding the best ordering of the step size parameters does not hold when numerical errors are significant. An alternative stable ordering is given in the paper.