There are also not many usual JFETs with a saturation current of at least 50mA, which was one of the requirements, and especially at low VDS voltages.
Actually, the title of this thread is "10-50 mA", so where you got that from is anybody's guess.
From the title. "10-50 mA" means from 10mA to 50mA to me, so potentially up to 50mA. Now if it meant anything in between as we see fit, nevermind.
(With this range, you'd need a JFET with at least 50mA saturation current to ensure a 50mA current sink. You can only adjust the current lower with a source-to-gate resistor, which is the common way of adjusting a JFET-based current source/sink. That was my point. You can also parallel several JFETs as you suggested, but that's more parts and more dispersion.)
Two 2SK170BL in parallel would do the job beautifully, and with much lower minimum voltage than the other solutions presented here.
Well, the 2SK170BL seems to have a ID vs VDS knee (@VGS = 0V) at a much lower VDS than with a lot of other common discrete JFETs, so that would be a decent fit regarding the operating voltage indeed.
And yes you can parallel several JFETs to sink more current. You'll also get more dispersion (and a lower output impedance if I'm not mistaken.) This transistor seems to be obsolete though and available only through old stocks. Maybe it has a more recent equivalent. Else the OP will have to resort to eBay.
The NPN version is more accurate as a current sink (this JFET's current will still vary up to at least 2-3mA over the whole voltage range, which may not be a concern here, I don't know), and as I said above, can be made to work at pretty low voltages given you use a low enough reference voltage. Yes, it will need more components. It will have some sensitivity to temperature, but the JFET as well. Not sure which of the two will be worse in that respect.
Yes the JFET solution is probably fine here if all the OP wants is to "bleed" some relatively constant current. That would not make a precision constant current sink, but I guess this is not what the OP is after.
In both cases, if they want to be able to set a defined current (even if not very accurately), they will need to make it trimmable. That's with an adjustable resistor between emitter and ground in the NPN case (and/or adjusting the reference voltage), or between the source and ground/gate for the JFET case. JFETs present a relatively large dispersion in their characteristics. If this is just a one-time project (or just a few), they can always hand-select them for a given current.