Feedback Summary: T~vo formulae regarding reduc- tion of output variation of an elementaql feedback system due to variations in system parameters are derived and discussed. Cer- tain confusing and often erroneously treated aspects are detailed and illustrated. Both authors, one from The Peoples Re- public of China and the other from the United States, have often encountered student diffi- culty in understanding how feedback reduces the incremental output variations caused by incremental variations in system parameters. There exist two different approximation solu- tions in texthooks [ 1 ~ 2,3] in use today. This difference was eliminated in later editions of [2,3]. Rather than start out with the concept of system sensitivity, the ultimate tool for such investigations, this presentation follows the lead of the textbooks used in introductory feedback control-system courses. Given a unity feedback system with forward-path gain G(s), input R(s) and out- put C(s), let us suppose that G(s) is subject to an incremental change AG(s), this increment being of small amplitude compared to that of G(s). For the closed-loop, feedback system sub- ject to variations AG(s) in G(s), the differ- ence between the output of the system with the perturbed parameter G(s) +AG(s) and that from the system with nominal parameter G(s) is AC,(S) = G(s) +AG(s) 1 +G(s) +AG(s) - a] 1 +G(s) *R(s) ( 1) Received May 31, 1983; revised September 23, 1983, and November 23, 1983. Accepted in re- vised form by Associate Editor G. H. Hostetter. Hongcai Zhang Richard S. Marleau Senior Member, IEEE If AG(s) is assumed small compared to 1 +G(s) in the denominator, then the dis- turbance in theoutput is which is sometimes accepted as the measure of the system response error. Starting again with equation ( 2) , but sub- stituting 6G(s) for AG(s) in the denominator, we have Obviously, the result of equation (4) is a better approxi mati on than that of equation (2). Equation (2) is essentially of no value and should be considered an in- correct approximation. Conclusion Realizing how readily this error was made in quality textbooks suggests that special care should be taken in presenting this topic to persons learning feedback concepts for the fi st time. This reaffirms the continuing care that must be taken in dealing with small [l +G(s)] G(s) AG(s) - G(s) [l +G(s) +~G(s)] 1 +G(s) +6G(s) - - [I +G(s)]AG(s) - G(s) 6G(s) [l +G(s) f 6G(s)] [ 1 f G(s)] Assuming 6G(s) =AG(s), and that both these terms are small compared to G(S), we have AC = 1 [I +G(S)] ~(~) which does not agree with equation (2). The result of equation (2) can be obtained from equation (4) by assuming G( s ) =0 and G(s) 4 G(s). The di f f erence i n equations (2) and (4) results from erroneous subtraction to obtain the small difference in two large numbers. (3) differences in large values, especially when approximations are involved. References [I ] K. Ogata, Moderrz Control Engineering, Englewood Cliffs, NJ : Prentice-Hall, Inc., 1970. [2] R. C. Dorf, Modern Control Systems, Addison-Wesley Publishing Company, Inc., 1967; Second Edition, 1974; Third mi- tion, 1980. [3] B.C. Kuo, Automatic Control Systems, Englewood Cliffs, NJ : Prentice-Hall, Inc., 1962; Second Edition, 1967; Third Edi- tion, 1975. [4] J. B. Cruz, Jr., Feedback Systems, New York: McGraw-Hill Book Company, 1972. 0272-1708/84/0joo-0317$01.03 0 1984 IEEE may I984 17