thermally compensated Vb bias circuit?

[davygrvy]

Hi Folks,

I was wondering about techniques to thermally stabilize a standard "adjustable battery" bias circuit. I was working on a three stage (diff,vas,output follower) poweramp circuit and found the VAS mosfet device was very susceptible to small temp variations. I double checked in SPICE and it confirmed DC offset would swing over 9 volts from 20C to 75C

I rearranged the circuit so that it used a dual diff front-end so that both top and bottom of the VAS is now driven, and that solved most of the problems. SPICE confirmed the swing is now just a volt, but now the adjustable battery in the middle between the two driven mosfets is the remaining problem. The layout is similar to this:

http://home.alphalink.com.au/~cambie/ETI466cct.zip

If we pretend that circuit is mine, what techniques are there to stabilize Q8?