114 BELL SYSTEM TECHNICAL JOURNAL 



TABLE III 



Errors REstrLTiNG from Use of Peak-Type Volume Indicator 

 ON A Long Program Circuit 



Upper Frequency Limit of Program 

 5000 Cycles 8000 Cycles 



Male Speech -3.5 db -4.5 db 



Female Speech —1.5 —3.0 



Dance Orchestra —2.0 —1.5 



Brass Band -3.0 -2.0 



Piano -0.5 -1.5 



The large errors indicated in the table are, of course, intolerable. 

 The effect of the line on the reading of the peak instrument is partly 

 due to the cumulative effects of the slight non-linearity in the many 

 vacuum tube amplifiers and loading coils in the circuit, and partly to 

 phase changes which alter the wave front and amplitude of the peaks. 

 It might be thought that phase changes which destroy some peaks 

 would tend to create others. However, a Fourier analysis of a sharp 

 peak will show that an exact phase relationship must exist between all 

 of the frequency components. The probability that phase shift in a 

 line will chance to cause all of the many frequency components of a 

 complex wave to align themselves in the relationship necessary to 

 create a peak where none existed before, is very slight, — indeed in- 

 finitesimal compared to the probability of the occurrence of a peak in 

 the original wave. 



Loudness 



Another important consideration is the correlation between volume 

 levels and the comparative loudness of different types of programs 

 [item (c) in the list given earlier]. This was tested by a method similar 

 to the "group method," described above in connection with the tests 

 on aural overload distortion. A group of observers was permitted to 

 listen to alternate repetitions of a test program and a reference pro- 

 gram, and was asked to vote upon which appeared the louder. A 

 particular selection of male speech was used as the reference program 

 for all of the tests and its level was kept constant. The test programs 

 included several different types and several samples of each type of 

 program. The samples of program were about 30 seconds in length. 

 Each test program was presented at a number of levels covering a 

 range from a low level where all the observers judged the reference 

 program to be the louder to a higher value where all of them judged 

 the test program to be the louder. 



Thus, a curve was established for each type of program between the 

 per cent of observers judging the test program to be the louder, and 

 the level of the test program. A sample of such a curve is shown in 



