838 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 



equalization error at ''n" independent frequencies plus the knowledge of 

 the effect of each of the ''n" controls at each of the "n" frequencies. 



It should be noted that this statement of the rule assumes analysis by- 

 frequency rather than by transient behavior. This approach is used 

 because, at present, equalizers are usually designed on a frequency char- 

 acteristic basis. The validity of the rule is however more general. 



From these two rules we can derive another regarding the minimum 

 information principle that is similar to Rule I but is actually quite inde- 

 pendent of it. 



Rule IV 



No more information should he gathered from the system than is necessary 

 to provide sufficiently accurate control setting information for the equaliza- 

 tion operator. 



In this case the superfluous information may actually cause confusion 

 or harm. It will, at the very least, confuse a manual operator to know of 

 an error he is powerless to correct, whereas a mechanized system, on the 

 other hand, would probably go berserk if it obtained too much in- 

 formation. 



It is evident that the minimum amount of information that would be 

 obtained in accordance with Rule IV would be the same as that obtained 

 in accordance with Rule I only if the design of the equalization were 

 optimum, because then the equalizer shapes (and the number of shapes) 

 would just suffice to permit satisfactory transmission of the signals. 



Up to this point we have determined in general what information is 

 needed, when it is needed and the optimum form of its presentation. 

 Now let us proceed to examine what to do with the data; which involves 

 the nature of the equalization operator as well as his equalization tools. 



If we followed Rule II rigorously, we should have several different 

 type of controls; one for each of the effects having different time rates of 

 occurrence. For purposes of illustration, however, we need assume only 

 two rates — one quite rapid and the other very slow. It will be postulated 

 here that very rapid equalization operations are most economically per- 

 formed by machine, such as for example, the automatic regulation for 

 cable temperature variations. Likewise relatively infrequent adjustment 

 will be assumed to be best performed by a suitably informed human 

 operator. 



The first principle to note is that the only real distinction here is the 

 rate at which data should l)e refreshed and acted upon. In either case 



