Convergence properties of analytic signal-based gradient adaptive lattice filter

This paper studies the convergence properties of the gradient adaptive lattice (GAL) predictor for an input time series consisting of real-valued sinusoids in white noise. An analytic signal representation of the input signal provides more reduced prediction errors. Increasing the stepsize, a faster convergence is obtained without degradation on the performance. Simulation results show that a complex-valued GAL with the analytic signal is superior to the conventional GAL with regard to the power spectral estimation and the convergence rate