^. 1 A /; 11 7 /> 77/ .1 \D TRA NSMISSION PERFORM A NCR 563 



mitting and receiving Ultcrs arc to be tlaussian, a problem arises as to how 

 to divide the total selectivity (in column C) between them. If most of the 

 pulse shaping is done at the transmitter, the Gaussian receiving filter must 

 be extremely broad, with the result that discrimination between pulses in 

 adjacent bands is poor and the bands must be spaced widely in order to 

 keep cross-tire down. If, on the other hand, all of the shaping is done at 

 the receiver the wide spectrum of the unshaped transmitted pulse spills 

 over into neighboring bands unless the bands are widely spaced. Clearly, 

 an optimum proportioning of selectivity exists and it is interesting and 

 enlightening to analyze this problem. Such an analysis was made for j^ulses 

 4, 5 and 6. This analysis pertains only to crossfire and not to signal-to-noise 

 ratio as influenced by curbing (shortening of the rectangular pulses) and 

 by the division of selectivity between transmitting and receiving filters. 

 Wide receiving filters accept more noise and narrow ones may prevent the 

 transmitted pulse from attaining full height in the receiving filter output if 

 curbing is used. If the curbing is pronounced, as in pulses 6 to 1 1 , amplifica- 

 tion may have to follow the transmitting filter to establish the desired 

 transmitted power level. For divisions of selectivity close to the optimum, 

 the receiving filter selectivity appreciably reduces the transmitted pulse 

 height in the case of pulse 5 and seriously reduces it in the case of pulse 6. 

 Crossfire from a pulse in an unwanted band appears as a transient in the 

 wanted band. In some circumstances, this transient has peaks which occur 

 while the crossfiring pulse is rising and falling and has a minimum between 

 which sometimes dips below the level fixed with the steady-state discrimina- 

 tion to the crossfiring carrier. If the pulses in the crossfiring band are 

 synchronized with those in the wanted band as they might be in PAM and 

 PCM only the minimum, central, crossfire might be significant. If, as in 

 PPM, the pulses cannot be synchronized, the peak crossfire is significant. 

 Curves for two values of band separation are shown in Fig. 27, one appro- 

 priate to yield minimum crossfire in the 25 to 35 clb range and the other to 

 yield peak crossfire in that range. This is the range that is sufiicient for 

 binary PCM. The steady state discrimination is also shown. \\c conclude 

 from this study that pulses 4 or 5 are about equally good in respect to mini- 

 mum central crossfire and that pulse 4 is slightly preferable in that the 

 trough and the crest are more symmetrical. For PCM in which the pulse 

 spacing is made equal to T,, this symmetry means that there is the same 

 margin for misalignment of the gating pulse, as regards correctly inter- 

 preting a space or a mark. Pulses 5 and 6 appear to be about equally good 

 in respect to peak crossfire but both (and particularly pulse 6) incur a signal- 

 to-noise penalty because the receiving filter does not permit the transmitted 

 pulse to attain full height. 



In practice, the approximations to Gaussian filters have shown worse 



