A comparison of adaptive gradient and adaptive least-squares algorithms

This paper presents a comparison of three autoregressive adaptive predictor algorithms. These algorithms are: the Adaptive Line Enhancer (ALE), the Gradient Adaptive Lattice Structure, and the Adaptive Least-Squares Predictor. Least squares algorithms have been developed by Morf et al [la]. More recently, Satorius incorporated the lattice form of the algorithm in a decision directed equalizer. This investigation is concerned with an analysis of the performance of the above adaptive algorithms when the random process being observed is composed of sinusoids in additive noise. In particular, the problem of estimating and resolving the sinusoidal frequencies is considered. In addition, such performance properties as, signal-to-noise ratio (SNR) and convergence constants are also discussed. This comparison is done through a computer simulation and results indicate the relative advantage of adaptive least-squares algorithm.