400 BELL SYSTEM TECHNICAL JOURNAL 



curves show the effect of delay upon noise in a 250- to 2750-cycle audio 

 band, as measured with the Weston db meter, for various differences 

 in steering. The curves are labeled in terms of the difference in the 

 phase settings of the two branches. The 40-degree or 80-degree curves 

 correspond in practice to steering on each side of a single bundle of 

 waves. The 160-degree curve typifies steering at two separate wave 

 bundles. The use of 100 microseconds (or more) delay is generally 

 advantageous for the audio addition when steering at one bundle. 

 Since this amount of delay makes the audio noise addition nearly 

 random and for widely different steering the addition is also random 

 the assumption that the noise from the two branches adds on a power 

 basis, made in reference to Fig. 34 (x = V2), is justified. 



The effect of delay is to produce an interference pattern in the audio 

 noise spectrum. This accounts for the dip in the noise curve for a 

 delay of 300 microseconds which locates the first interference minimum 

 at about the center of the audio band. The asymptotic approach to 

 3.5-decibel reduction corresponds more nearly to a reduction ratio of 

 2 fir than to 1/V2 due to the fact that the Weston db meter is nearer 

 linear than square law in response. 



In obtaining these curves it was desired to simulate the reception of 

 two waves for which the corresponding branches were phased to add. 

 It was not convenient to set up locally such a two-wave case but the 

 single wave input should give identically the same results provided 

 phases were avoided which resulted in a signal at the second detector 

 too low to demodulate the noise. A signal level so high that further 

 increase did not affect the noise output was used for all points. The 

 real purpose of the signal was to insure that the demodulated noise 

 was not dependent upon the intermediate-frequency bands and that 

 the results would be unaffected by possible differences in intermediate- 

 frequency bands. ^* 



As mentioned, noise has been measured with an unweighted 250- to 

 2750-cycle frequency band. Had a weighting network ^^ which em- 

 phasizes frequencies in the vicinity of 1000 cycles been used the dips 

 in the curves marked 0° and 40° would have been deeper and would 

 have occurred in the vicinity of 500 microseconds delay. 



Returning now to Tables III and IV the measured improvements 

 were corrected to correspond to the effect of the delay which would 

 probably have been used to obtain the best audio addition for the 

 signal. The 1.2-decibel correction for the one-bundle case represents 



-^This precaution was subsequently found to be unnecessary; i.e., similar results 

 were obtained with no input signal. 



2' Barstow, Blye, and Kent, "Measurement of Telephone Noise and Power Wave 

 Shape," Elec. Engg., vol. 54, pp. 1307-1315, December, 1935. Technical Digest 

 published in Bell Sys. Tech. Jour., January, 1936. 



