So this quantum limited photoreceiver again.
After spending yesterday finding several fractional-decibel bits of buried treasure, it's shot-noise limited above 30 nA in a full 1 MHz bandwidth, which is pretty good. CPH3910s rock, and the optimal drain current is a tradeoff: as you increase it, e_N goes down slowly but C_iss goes up rapidly.
It has an LM2594 switcher right in the box, which produces spurs too small to see easily on the spectrum analyzer, even at narrow bandwidths. Soft edges, a toroid, a cap multiplier, an electrostatic shield, and2-1/2 inches of distance get rid of the junk pretty well.
However, some wall warts make the box produce larger spurs--for instance, a Thinkpad supply produces a spur about 10 dB above the noise floor in 5 kHz bandwidth. I can see that this is going to be a generic problem.
I have some things to try, e.g. a CM choke and so on, but I thought I'd ask what you folks do about it.