A comparison of two fast linear predictors

Several adaptive linear prediction algorithms exist which require order N computations, where N is the number of prediction stages. Among these are some which derive from an error surface gradient approach and others result from cumulative squared error minimization. The gradient adaptive lattice and the least squares adaptive lattice are two algorithms analyzed here with the purpose of quantifying and comparing their performances in stationary and non-stationary signal cases. Through appropriate selection of algorithm parameters and some manipulation of the form of coefficient update equations, a close correspondence between the two algorithms is obtained. Simulations show like misadjustment noise and convergence rate properties for the two algorithms. Another result is a simple expression for the stationary misadjustment noise of the cumulative least squares algorithm.