INSTANTANEOUS COMPANDORS 707 



exposure. Stronger signals are amplified less highly. Loss, therefore, is re- 

 moved from the expandor as the signal increases and the noise increases 

 correspondingly. 



When the signals are conventional speech signals, loss is removed from 

 the expandor as the speech volume increases and consequently the noise 

 volume increases correspondingly. An instantaneous compandor has the 

 important advantage that level adjustments are frequent, for example, in 

 a pulse-modulation system at the rate of about 8000 times per second for a 

 message channel whose bandwidth approaches 4000 cycles. Consequently, 

 the increased noise will be continuously masked by increased speech sound. 

 During all silent periods, unwanted noise and interference receive maximum 

 noise suppression in the expandor. For an ordinary message channel these 

 advantages are substantial. 



Viewed broadly, an instantaneous compandor provides a read)' means for 

 making the noise susceptibility a function of the magnitude of the signal. 

 If the noise susceptibility is made less than that of a linear system in one 

 portion of the range, then it must be greater than that of a linear system in 

 some other portion of the range. Whether an over-all improvement results 

 depends entirely upon the nature of the signal. For example, in certain tjrpes 

 of picture transmission systems a given value of noise produces about as 

 much harm whether the signal be weak or strong. In this instance no benefit 

 would accrue from making the noise susceptibility a function of signal 

 Strength. 



An important consideration, therefore, is the evaluation of the noise ad- 

 vantage due to instantaneous companding. The theoretical treatment will 

 give relationships for signal-to-noise ratio and noise susceptibility. Applica- 

 tion of the theory to a particular example including a numerical evaluation 

 of the noise advantage will be deferred to the last section. 



Method of Analysis 



The analysis is based upon deductions'^ related to the sampling principle 

 and is illustrated by Fig. 1 which shows a schematic of one channel of a multi- 

 channel time-division system. 



The incoming signal (Fig. 1) is filtered by a low-pass filter designated Fy. 

 At the output of Fi the signal should be regarded as an arbitrary signal oc- 

 cupying the band of all frequencies slightly less than B. Brief samples of the 

 signal are taken uniformly at the rate of 2B samples per second. In this 

 manner the signal is converted into a series of PAM (pulse amplitude modu- 

 lated) pulses as indicated in Fig. 2. There is a unique relationship'^ between 

 signals and samples (PAM pulses) ; if we are given the signal wave we can 



