High value current source design for thermal stability

[Fusion916]

Any recommendations for 1A+ current source designs that have very good thermal stability? Something with local feedback, not for accuracy, but for thermal stability?

 

[Fusion916]

bump

to clarify, im looking for something that maybe tracks vbe of the current source device so vbe, I, and heat don't forever increase.

 

[jcx]

headroom/dropout V, output compliance V range, output impednace vs frequency requirements are also useful for design in addition to actual Tempco/power stability requirement number

LM317 cascoded with IXYS depletion mode 6 A power MOSFET may meet some sets of requirements

 

[Fusion916]

headroom/dropout V, output compliance V range, output impednace vs frequency requirements are also useful for design in addition to actual Tempco/power stability requirement number

LM317 cascoded with IXYS depletion mode 6 A power MOSFET may meet some sets of requirements

I'm looking for a solution that only uses discrete devices.

Here is the scenario I'm trying to avoid for a circuit such what is attached (sorry for the crude drawing):

Constant bias being applied to the BJT, say 1V, and vbe at t=0 is 0.7, so the current is 0.3/R. And lets just say R is 0.3 for a 1A bias. I'm looking to avoid this:

BJT heats up > reduction of vbe > vbe/r current goes up > device heats up more > further reduction in vbe > etc

Then thermal runaway. Basically I'm looking for a circuit that has good tracking of the bias being applied to the BJT. I suppose I could mirror the current and thermal match, but that would add another power device. Looking to avoid that too.

 

[counter culture]

Why not read these...

http://waltjung.org/PDFs/Sources_101_P1.pdf
http://waltjung.org/PDFs/Sources_101_P2.pdf

It would seem a trifle dogmatic to exclude a solution just because it uses an IC.

 

[jcx]

or 2 Q in "loop" feedback?


"ring of two" bjt feedback ccs uses the Vbe of the feedback/control Q - the heating of the pass Q only has an effect through h_fe tempco (or MOSFET Vth TC divided by loop gain)

pic below was just something I had on hand - hope you can parse pnp source version of the topologies

left most uses a higer Vref - 10 V Zener, and "Baxandal" or "complementary Darlington" - the Vbe nominally cancel - but you would have to bolt both to the same heatsink if large power were involved - and the tracking would have long thermal time constant

the other 2 are the "ring of two" - middle one cascoded