WuYit, indeed the problem is current. Noise spectrum vs. current is not shown on datasheets and we still must characterize each part, or at least each production run. We have fairly good correlation between fft spectra (looking at maximum peak values over a few minutes) with perceived noise character. Tonal balance is another question and we are still working on analytical methods for that. A typical microphone output stage (one emitter follower for each side of the line) has no more than 3.5 mA from phantom power so this is our limit. Some parts we have tried, like the THAT arrays, are very quiet at 20 mA but much noisier at 3. We have seen huge differences also from different foundries of the same maker -- Fairchild 2N5087 from one plant are entirely different from those from another.
Joachim, thanks, I will go look up your MPP thread, we are always interested in new topologies. We are concerned with the mic output stage and the mic preamp stage too, try to get as much transparency as possible so that the mic capsule's character is fully available to the a/d converter. It's that pesky path in between that we have to take care about...
On a related note, I learned that Renesas 2SA1084/1085 is now end-of-life, last buy from the factory is in December. They are still available for about 15 cents, we'll be buying a good stock of them. The NPN is on "no promotion" so it will soon be EOL as well, I think.