A Frequency Domain Analysis of Learning Control

Abstract

The convergence of learning control is traditionally analyzed in the time domain. This is because a finite planning horizon is often assumed and the analysis in time domain can be extended to time-varying and nonlinear systems. For linear time-invariant (LTI) systems with infinite planning horizon, however, we show that simple frequency domain techniques can be used to quickly derive several interesting results not amenable to time-domain analysis, such as predicting the rate of convergence or the design of optimum learning control law. We explain a paradox arising from applying the finite time convergence criterion to the infinite time learning control problem, and propose the use of current error feedback for controlling possibly unstable systems.