INSTANTANEOUS COMPANDORS 717 



Figure 7 illustrates the relationship between noise susceptibility and signal 

 voltage. Noise susceptibility (Fig. 7) is expressed in db relative to that of a 



linear system. 



Value of S/Nr 



To evaluate the noise advantage which results from the use of an instan- 

 taneous compandor, it is necessary to know what requirement to place on 

 S/Nr. This ratio refers to noise at the output of the system during intervals 

 when the signal magnitude is within the exponential range of the expandor. 

 During these intervals the noise susceptibility of the system is propor- 

 tional to the signal magnitude, so that the character of the noise is entirely 

 different from that encountered in a linear system. People listening to speech 

 transmitted through a system equipped with an instantaneous compandor 

 have mistaken this type of noise for the distortion produced by an over- 

 loaded amplifier. Accordingly, experiments were made to determine how 

 small S/Nr could be in a telephone channel whose frequency range was 

 200 to 3500 cycles. 



A test circuit was devised which simulated the noise performance of a 

 system equipped with a logarithmic compandor, and arrangements were 

 provided so that the signal-to-noise ratio S/Nr could be adjusted over a 

 wide range of values. The test procedure was to allow an observer to listen, 

 during two consecutive intervals of time, to speech from the output of a 

 Unear system and from the compandor system. Conditions were arranged so 

 that the noise at the outputs of the two systems was the same when the 

 signal voltages were within the linear range of the compandor. The sequence 

 in which the two conditions were presented to the observers was changed in 

 a random manner, so that there was no way of identifying the compandor 

 system except for the effect of the enhanced noise susceptibiUty during 

 intervals when the signal magnitude was within the exponential range of the 

 expandor characteristic. Twenty- two observers participated in these tests 

 and different speech volumes were used covering a range of 26 db. 



Experimental results showed that the compandor system could be readily 

 identified when S/Nr was 16 db or smaller, whereas the difference between 

 the two systems was difficult to detect when S/Nr was 24 db or greater. 

 An acceptable value of S/Nr for a typical telephone system is therefore some- 

 where between these two limits. A value of 22 db* was selected as a con- 

 servative estimate. To confirm this, several people experienced in rating the 

 quality of telephone systems were asked to listen to the output of the test 

 circuit with S/Nr adjusted to 22 db. The consensus was that the quaUty was 

 satisfactory. 



* This value is in agreement with the one used by C. B. Feldman and W. R. Bennett in 

 studies of bandwidth and transmission performance, reported in reference 10. 



