LeIand C. Scot wrote:
http://perso.orange.fr/f6itv/p2032001.htm (look at location of diodes in
photo at right side)

Yes, this method is similar to a few "reference" designs shown in the
Motorola RF Data Manual . This is a poor method for two resaons. The
first is what I explained before... trying to get two diodes in
parallel to turn on together is very difficult... if not impossible on
a repeatable basis. Second, the emitters you are feeding with the DC
bias are very low impedance. This creates its own problem when
attempting to control the base current. This biasing scheme with temp
compensation is a "brute force" method that dissipates large amounts of
power... and plain doesn't work well. It's not the first time that
circuits shown in a reference manual are not production ready. Motorola
also shows a much better method of temp comp bias in that same data
manual. It uses an op amp and sink mounted thermistor. I'm also using
an "active" approach to the temp comp bias in my amplifier... but a
totally different circuit using a bipolar transistor as the sense and
gain mechanism. I can hold the 500mA bias to 10% from -30 to +85C.

www.telstar-electronics.com