468 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1957 



The noise figure of the audio amplifier following the detector is given in 

 general by: 



^AMP = n i42 



Kg 



where Eg is the generator input resistance in this case the barretter 

 resistance and /?equ is an eciuivalent noise resistor in series with the 

 genei'ators. The best input tube that we found was the General Electric 

 GL 6072 triode* for which we measured an /^ equ at 100 c.p.s. of '^10" il. 

 (This is due to flicker noise of the tube and has a \'J dependence.) If 

 we were to connect the barretter, having a resistance of ^200 12 straight 

 to the grid of the input tube we would get a noise figure of '^oOO. How- 

 ever, b\' using a step-up transformer with a turns ratio n > -^^ the 



Kb 



noise figure of the amphfier can be reduced to nearly unity. We^\-ill as- 

 sume in the following analysis that this has been done. 



The noise temperature of the barretter Ib was thought to be approxi- 

 mately 2 since it is merely a platinum wire operating at an ele\'ated 

 temperature. To our surprise the measured value turned out to vary 

 for different units between -i and 40. f The noise figure was measured 

 on about 20 different units obtained from 4 different manufacturers 

 (P.R.D.; F.X.R. Xarda, Sperry). The reason for this noise is not en- 

 tirely clear at present. A possible explanation is the non -uniform heat- 

 ing of the wire which could set up air currents. They in turn can cool 

 the wire in a random fashion giving rise to an additional noise com- 

 ponent. An improvement of the noise figure was noted upon evacuating 

 the barretter. The noise figure of a unit which was initially 10, dropped 

 to the expected value of 2 after evacuation. However, it should be 

 pointed out that this cannot be taken as a definite proof for the "air 

 current theorj^" since the characteristics of the barretter changed 

 markedly after evacuation. The sensitivit\' of the evacuated barretter 

 went up from oV.'mW to 200n 'mW which necessitated a reduction of the 

 dc current from 8 to 1.5 niA. Also the response time went up by a factor 

 of 20, so that the effectiveness of any noise mechanism with a 1// spec- 

 trum would be greatly reduced. This approach however looks definiteh- 

 promising in trying to design more sensitive and less noisy barretters. 

 In the present work commercial unevacuated barretters were used, their 

 noise temperature being taken as 4 in the following anah^sis. Under 



* We are indebted to R. G. Shidman for bringing this tube to our attention. 

 t One unit which exhibited an extreme!}' large noise figure of 1,000 was elim- 

 inated entirely. The solder point of the platinum wire was apparentl}' defective. 



