On Thursday 29 September 2005 08:45 pm, Greg Cunningham wrote: > I'm not an engineer, but I recollect years ago Elekto mag > published a very lo-noise mic amp by parallelling a > dozen-or-so gp bipolars. Their argument was that the > randomness of the noise sources produced cancellation, while > summing the in-phase signals added.
The correct argument is that by paralleling devices, the effective noise source resistance is lowered, which lowers the noise voltage. Making the transistors bigger has the same effect. This will only work if you have a low source impedance. It will make the noise worse if you have a high source impedance. An almost correct argument, that is closer to what you said is that when you sum coherent (in-phase) signals, you get a 6 dB increase. When you sum incoherent signals, you get a 3 dB increase, hence a 3 dB improvement in signal to noise ratio. But actually, it can only go so low. If the microphone itself has an impedance of 150 ohms (which is typical) that determines the theoretical best you can do. The practical best is 1 or 2 dB worse. How much worse than theoretical is called the "noise figure".
