An anti-windup design problem is posed and it is shown that all ''obse
rver-based'' anti-windup modifications solve this problem at least loc
ally. Sufficient conditions for an observer-based anti-windup modifica
tion to solve the global version of this problem are presented. A nove
l observer-based anti-windup design is then proposed such that the ant
i-windup problem can be interpreted in terms of a reduced-order system
. In particular, the proposed anti-windup design induces an invariant
subspace for the dynamic behavior of the mismatch between the constrai
ned and unconstrained closed-loops. The dynamics in this invariant sub
space are identical to the behavior of the plant with input saturation
starting at the origin, stabilized by linear state feedback and drive
n by the mismatch between the unconstrained input and this input passe
d through a saturation function. The second part of the paper shows ho
w the original dynamic compensator can be modified, while retaining th
ose dynamic features that produce a desirable closed-loop steady-state
response. to ensure that the requisite invariant subspace exists and
it is reasonably tuned for input saturation. Two case studies are carr
ied out on systems that have been investigated in the literature. (C)
1998 Elsevier Science Ltd. All rights reserved.