8 BELL SYSTEM TECHNICAL JOURNAL 



quantizing noise in this range. How large the effect may be is determined by 

 the degree of taper. To apportion the noise at different levels, the distribution 

 of steps over the amplitude range can be varied. 



Preliminary studies of quantization, involving listening tests and noise 

 measurements for various numbers of digits and various kinds of taper, led to 

 the choice of a seven-digit code (128 steps) for the present system. The taper 

 employed reduces the smallest steps 26 decibels below the average size and 

 increases the largest ones about 6 decibels. 



Coding. The coder is required to set up a pulse code combination for each 

 quantized signal value. A great many codes are conceivable, but in practice 

 a simple one in which the pulses correspond to digits of the binary number 

 system allows greatest simplicity at the receiver. 



While coders may take a wide variety of forms, they can be arranged in 

 three categories according to the way in which they evaluate speech amplitudes. 

 In the first category an amplitude is measured by counting out, with a binary 

 counter for example, the number of units contained in it one by one until the 

 residue amounts to less than a unit. In the second, the amplitude is measured 

 by comparison with one digit value after another, proceeding from the most 

 significant digit to the least, and subtracting the digit value in question each 

 time that value is found to be smaller than the amplitude (or its residue from 

 the previous subtraction). In the third, amplitude is measured in toto by 

 comparison with a set of scaled values. Modulators disclosed by Reeves^ 

 and by Black and Edson^ are of the first category. That described by GoodalP 

 belongs to the second, and the one described in the present paper is in the third 

 category. Generally speaking the number of operations and the time required 

 for coding decrease in going from the first to the third. Rapid coding is obvi- 

 ously desirable since it allows more channels to be handled in time division by 

 common equipment. 



Multiplex 



Channels may be multiplexed by arranging them in time sequence, or by 

 arranging them along the frequency scale. These methods are known as time 

 division and as frequency division, respectively. The first is accomplished 

 by gating, or switching, at precisely fixed times. One way of doing this im- 

 presses a more or less rectangular pulse on one of the grids of a gate tube, so 

 that a signal wave may be transmitted during the gating pulse. The second 

 method here refers to the use of amplitude modulators, each supplied with an 

 appropriate carrier, which translate the signals to their assigned positions on 

 the frequency scale. To avoid crosstalk in a time-division system, operations 

 in group equipment common to a number of channels must proceed without 

 memory of the amplitudes of preceding channels, requiring build up and decay 

 times to be held within limits. This implies a sufficiently wide pass band 



