40 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1951 



if all digits were produced simultaneously. In this type of coder a line beam 

 is used which covers the full width of the code plate and the code is registered 

 simultaneously on a plurality of digit collectors, one collector being used 

 for each digit of the code. This is one of the features that was included in 

 the coding tube used in these experiments. 



In the new tube the time required for quantization by feedback has been 

 avoided by using a different type of code which avoids the large errors which 

 are present in the conventional binary code for amplitude values one-half 

 way between the integer values permitted by the code. This new code, here 

 called the reflected binary code, is shown on the right side of Fig. 4. 



For present purposes it should be noted that at points one-half way be- 

 tween integer values the beam intersects only one edge in any code group. 

 Further, if the deflection axis is tilted so that an incorrect code group is in- 

 dicated, the resulting error is only one quantum level, instead of a much 

 larger value possible with the conventional binary code. 



In practice, even if the beam is properly lined up, there wdll be times when 

 the output for the digit in which the beam intersects an edge will be between 

 zero and full output. Since this digit must be unambiguously represented 

 either by a full pulse or no pulse, it is necessary to make a choice and quantize 

 the particular output under discussion. The output of the digit collectors is 

 amplified and the final quantization of any uncertain digit is done by a slicer 

 which is in the output of each digit amplifier. Use of this code thus localizes 

 final quantization to within a single digit, and an arbitrary choice results at 

 most in an error of one quantum level. 



The use of the reflected binary code for PCM applications was suggested 

 to the writer by Mr. F. Gray. As mentioned before, this code is translated 

 to the conventional binary code. The translator used in these tests was 

 designed by Mr. R. L. Carbrey w^ho developed it specifically for this ex- 

 periment. 



Results of Experiments 



We now pass on to some of the results obtained with this system. Figure 

 5 should help in understanding the results shown in the remaining figures. 

 It shows a triangular wave which has been analyzed into the three ''on" 

 and ''off" rectangular waves shown in the bottom part of the figure. 



In this paper we shall follow the convention that the digit of largest am- 

 plitude is the first digit, the next largest digit is the second digit and so on. 

 By this convention the first digit is J of the total amplitude range, the 

 second digit § of the first, or } of the total range. Thus, the amplitude of any 

 given digit would bejn of the total amplitude range. 



It is convenient to think of the first digit as a first-order approximation to 



