Document Type


Date of Award


Degree Name

Master of Science in Electrical Engineering - (M.S.)


Electrical Engineering

First Advisor

Timothy Nam Chang

Second Advisor

Bernard Friedland

Third Advisor

Yun Q. Shi


In this thesis, the decentralized feedback structure for large scale, linear time invariant systems is studied. The internal differences and the relationship between decentralized feedback structure and centralized feedback structure are discussed. The conventional diagonal feedback structure. corresponding to the classical single loop design strategy, is first analyzed. This is followed by an arbitrary decentralized information flow constraint which is dependent upon the actual plant characteris¬tics. Although signal flow graphs have limited use in describing decentralized control systems. the concept of a control cycle unit based on signal flow is introduced as a supplementary tool to characterize some fixed modes and decentralized feedback structures. For the decentralized feedback structure, the Jordan normal form method and essential control tuple space method are presented. The later method can be readily applied in a computer-aided design environment.

From the theory a set of relationships of eigenvalues and eigenvectors between the plant system and the synthesis system are deduced. Based upon such eigenstructures, conditions have been found to determine the optimal decentralized feedback structure, that is, one with the least number of non-zero gain elements. The notion of a feedback gain lattice is introduced for both the diagonal and Jordan form representation of the plant state matrices. This lattice structure is then utilized algorithmically to generate the optimal decentralized feedback structure. These algorithms can be used to reduce hardware implementation and system complexity for the control of large scale systems.