I promised a bit of low-noise JFET comparison data. I looked at 4 BF862s from different reels, three ON Semi parts, the CPH3910, NSVM2394, and NSVJ3557, and the Fairchild MMBFJ310.
I swapped them in the same QL01 photoreceiver prototype, and found that all flavours of BF862 (from 4 reels) and two ON Semi parts, the CPH3910 and NSVJ2394 were indistinguishable (within a fraction of a dB out to 1 MHz), while the NSVJ3557 was about a decibel worse and the 310 a good 3 dB off the pace. The circuit is a fancy bootstrap (A_V = 0.9997 out to
1 MHz by actual measurement) with about 22 pF total capacitance on the gate--11 pF FET C_iss and 11 pF from the photodiode. This makes the FETs a bit musclebound for the application, because ideally its capacitance and g_M both go like the area, and the noise voltage goes like 1/sqrt(g_M). That pushes the optimum in the direction of a bit more noise and a lot less capacitance.
I have some smaller JFETs on order: BF556A, BF545A, B, and C, so we'll see when they get here.
However, for present purposes, it looks like we have a good second source for BF862-compatible JFETs.
I'm not posting the pictures because they wouldn't mean much without the schematic--there's a LP filter in the box and an external HP filter to protect the spectrum analyzer input, so on 2 dB/division, the curves look like inverted 'U's, with the e_N C noise fighting the filter rolloff.
Cheers
Phil Hobbs