INSTANTANEOUS COMPANDING OF QUANTIZED SIGNALS 697 



D. Possibility of Using Automatic Volume Regulation 



The realization that the quantizing impairment experienced by weak 

 signals in the absence of compression stems from their inability to excite 

 a sufficient number of the quantizing steps which must be pro^aded to 

 accommodate loud signals, leads directly to the suggestion that auto- 

 matic volume regulation be used to permit all signals to be "loud," 

 i.e., to excite the entire aggregation of quantizing steps. In its simplest 

 form, this would be accomplished by automatic amplification of the 

 long time average speech power in each channel to provide a constant 

 volume input to the common channel ecjuipment. 



Study of the present results indicates that if all signals were of con- 

 stant volume, about 10 to 15 db below full sinusoidal modulation (to 

 provide an adequate peak-clipping margin), satisfactory operation, 

 corresponding to signal to c^uantizing error power ratios in excess of 20 

 db, might be achie\'ed without companding by using as few as five or six 

 digits per code group. In evaluating this alternative, the advantages of 

 reduction of bandwidth, decreased complexity of c^uantizing and coding 

 equipment, and elimination of the common channel compandor, must 

 be balanced against the disadvantage of providing separate volume 

 regulators in each channel. 



to"- 



E. Comparison with Previous Experimental Results 



The literature contains seemingly contradictory statements about 

 whether fi\'e, ' six, or seven ' ' digits per code group are required for 

 satisfactory performance in speech listening tests. Evaluation of these 

 conclusions is frecjuently hampered by the lack of specification of either 

 the degree of companding employed or the range of speech volumes 

 requiring transmission. Different conclusions may therefore be consist- 

 ent, inasmuch as the systems may differ significantly in the required 

 \olume range, degree of companding, size of the "effective dc component" 

 in the signal, and even in the subjective standards used to judge per- 

 formance. 



Fortunately, the description of an experimental toll quality system 

 by Meacham and Peterson is sufficiently detailed to permit some com- 

 parison. The range of volumes they considered suggests that direct 

 comparison with our hypothetical system is fairly reasonable. Their 

 empirical choice of seven digits, with a compression characteristic vir- 

 tuall}" indistinguishable from that corresponding to fj. = 100 (see Fig. 4) 

 is in excellent agreement with the present conclusions. 



Furthermore, the conclusion that five or six digits, without compand- 



