136 SAMPLED-DATA CONTROL SYSTEMS 
(6.23) other terms which would not add to the pulse transfer function. 
For example, one could add terms of the form 
a; 
(s + a;)? + («/T)? 
with arbitrary a’s and a’s without changing the pulse transfer function 
since the z transform of all these addition terms is zero. This method is 
treacherous, however, because these terms add to the intersample ripple 
and can cause enormous “‘hidden oscillation” in the output of the system, 
as shown in Sec. 8.5. There is a certain arbitrariness to the selection of 
the K(z) given by (6.19), but no other method seems to lead to a specifica- 
tion which can be realized by a practical tandem network. In particular, 
the methods of Chap. 7 for the design of digital controllers are not satis- 
factory for the problem of tandem network compensation. However, by 
a slight modification of the structure and the possible introduction of an 
additional sample and hold circuit, one can avoid the problem completely, 
as shown in the next section. 
6.6 Design of Pulsed-network Compensators. Method 5 
The original work on pulsed networks for sampled-data control systems 
was done by Sklansky,®* and the treatment here follows that work. A 
block diagram of the basic system being considered is shown in Fig. 6.16. 


Hold 

| 
| 
Plant 
| 
| 
| 
| 

Pulsed network 


Fic. 6.16. Pulsed network compensation of sampled-data system. 
In this figure, the transforms P(s) and Q(s) represent the transfer func- 
tions of RC networks and supply the shaping or compensating action in 
the control loop. The design of the pulsed network for a control applica- 
tion may follow either of at least two directions. In the first place, the 
designer may select a specific form for the pulsed network and obtain the 
best design of the system in terms of the best parameter values of his 
chosen network. This is referred to as conventional design. A second 
alternative is to select at the outset the over-all transfer function which 
satisfies his needs and to calculate the transfer function of the network 
which will do the job. This alternative is the time-domain synthesis 
