Sort of loosely based on the idea of AoE3 Fig 8.42 (the bottom one), with some embellishments (cascode to deal with Ccb, CCS collector load, follower output, diff pair for obsessively stable bias servo).
Funny thing is, it measures about 2.4 nV/rtHz, which is better than most op-amps, but not as good as the 1nV/rtHz or less that should be typical of a single beefy transistor in CE.
And also, 2.4nV/rtHz, in a 20MHz bandwidth, on a solderless breadboard, with no particular precautions (there's no ground plane underneath, and the overhead light is LED with SMPS inside), which should adequately confuse some people here.
Just in case, I built it up on a proper ground plane, too.Same. Noise. Factor.
Simplified schematic:(Mainly missing some compensation components, that detract from the overall operation and shouldn't affect the gross performance.) (Incidentally, the breadboard got about 30MHz and the build-up got about 40MHz of bandwidth. SPICE said about 30MHz, and produced unusually realistic component values for once.)
I tried a bunch of other transistors, thing is, as the bandwidth is different for each one, the noise (Vrms into 20MHz) can be less, so it's a bad measurement. I haven't done a survey by noise density yet.
(On the other hand, I did notice my cleaned up wideband amp(not this exactly, hence, it's been cleaned up a bit..) is doing a rather satisfying 1.3nV/rtHz. Which doesn't make sense for the puny little junctions doing the work. What's up with /that/?)