The Effect of Computational Delay on Performance of Adaptive Control Systems

In many adaptive control applications, especially where the recursive-least-squares (RLS) algorithms are used, the real-time implementation of high order adaptive filters for estimating the disturbance dynamics is computationally intensive. The delay associated with the computational burden is usually either underestimated as no delay or overestimated as one sample delay in the control system design and analysis. For a stochastic disturbance dynamics, the H2 optimal control performance for the case of one-step delay is worse than that of no delay due to the nonminimum phase plant zero introduced by the delay. The optimal performance for a fractional delay is bounded between these two extremes. The paper investigates the effect of the fractional computational delay on a variable order adaptive controller based on a recursive least-squares adaptive lattice filter. The trade-off between the adaptive filter order and the computational delay is analyzed and evaluated by an example.