':i. -o.i 



4 J. 6 



is 









.V • 33 

 >i .. 0.CO 



< r c m 0.01 



\ 









\ l" c 











1 





1 1 





■\ 







V • 100 



>l - O.Ol 

 S. = O.OO^ 



■ a 



1 













V- 







I 







1 



0.1 0.2 03 O.k 



Figure '(. Frequency response. 5 (r) versus the normalised fre- 

 queue,)' r for tvo sets of parameters. Curves A and B illus- 

 trate that., with r„ fixed, a small h gives a sharper actual 

 cutoff frequency r ac > but only at the expense of increasing 

 .V if the laaxiraun absolute gain error in the filter is not 

 permitted to increase. 



discussion discloses the significance of increasing A' Lfi 

 achieving a si^ail absolute gt-in error E when the actual 

 cutoff frequency is made sharper by decreasing the parameter 

 h. The gain errors are not observed in the curves A and B 

 because of their small magnitudes relative to the scale 

 used in the plots. 



To indicate further the relative influence of the parameters, 

 r c , h, and .V on Lhe gain error E, attention is d-rected to 

 figure 6. The curves A, B, and C in rigure S indicate (for 

 a given r c and A") how the gain error S decreases as h Is 

 increased. This decrease in gam error E by increasing h 

 is obtained at a sacrifice of increasing the actual or 

 realisable cutoff frequency r^, which increases with h. 



21 



