Page 264 - Modern Control Systems
P. 264

238              Chapter 4  Feedback Control System  Characteristics

                           Then, in terms  of F(s), we define  the sensitivity function  as





                           Similarly, in terms of the loop gain, we define the complementary sensitivity function as
                                                                L{s)
                                                       cw                                      (46)
                                                            = TTjfe-

                           In terms  of the functions  S(s) and C(s), we can write the tracking error as
                                          E(s)  = S(s)R{s)  -  S{s)G{s)T d(s)  + C(s)N(s).    (4.7)

                           Examining Equation  (4.7), we see that (for a given G(s)), if we want to minimize the
                           tracking error, we want both S(s) and C(s) to be small. Remember that S(s) and C(s)
                           are both  functions  of  the controller, G c(s), which  the control  design  engineer  must
                           select. However, the following special relationship between S(s) and C(s) holds

                                                       S(s)  + C(s)  =  1.                    (4.8)
                          We cannot simultaneously make S(s) and C(s) small. Obviously, design compromises
                           must be made.
                              To analyze the tracking error equation, we need to understand what it means for
                           a  transfer  function  to be "large" or to be "small." The  discussion  of magnitude  of a
                           transfer  function  is the  subject  of Chapters 8 and 9 on frequency  response methods.
                           However, for our purposes here, we describe the magnitude  of the loop gain L(s) by
                           considering the magnitude  |L(/o>)| over the range of frequencies,  a>, of interest.
                              Considering  the  tracking error  in Equation  (4.4), it  is evident  that, for  a given
                           G(s), to reduce  the  influence  of  the  disturbance, T d(s), on the  tracking error,  E(s),
                           we desire L(s)  to be large over the range of frequencies  that characterize the distur-
                          bances. That  way, the  transfer  function  G(s)/(1  + L(s))  will be  small, thereby  re-
                           ducing the influence  of T d(s). Since L(s)  = G c(s)G(s),  this implies that  we need  to
                          design  the controller G c(s) to have a large magnitude. Conversely, to attenuate  the
                          measurement  noise, N(s), and reduce  the influence  on the tracking error, we desire
                          L(s)  to be  small  over  the  range  of frequencies  that  characterize  the  measurement
                          noise. The transfer  function  L(s)/(1  +  L(s))  will be small, thereby  reducing the in-
                          fluence  of A^(^). Again, since L(s)  = G c(s) G(s), that implies that  we need to design
                          the  controller  G c(s)  to  have  a  small  magnitude. Fortunately, the  apparent  conflict
                          between  wanting  to  make  G c(s)  large  to  reject  disturbances  and  the  wanting  to
                          make  G c(s)  small  to  attenuate  measurement  noise  can be  addressed  in  the  design
                          phase by making the loop gain, L(s), large at  low frequencies  (generally  associated
                          with the  frequency  range  of disturbances), and making L(s)  small at  high  frequen-
                          cies (generally associated  with measurement noise).
                              More  discussion  on  disturbance  rejection  and  measurement  noise  attenuation
                          follows in the subsequent sections. Next, we discuss how we can use feedback  to re-
                          duce the sensitivity of the system to variations and uncertainty in parameters in the
                          process, G(s).This is accomplished by analyzing the tracking error in Equation  (4.2)
                          when T d(s)  =  N(s)  =  0.
   259   260   261   262   263   264   265   266   267   268   269