K
Kevin Aylward
John said:My NMR and MRI gradient amps use a similar topology... the driver
opamp has its supply rails cascoded to make signal currents, and the
upper and lower power stages are active current mirrors. I'd argue
that adding an opamp per power fet makes things faster, stability
better, and compensation simpler, since the gate/Miller capacitance
disappears...
And your argument would be incorrect I am afraid to say. I have already
outlined these issues recently in another post.
There is no such thing as a free lunch. You cannot achieve better stability
by using extra feedback loops in this manner. The transfer function of the
op-amp and fet combined will cause considerable grief. Only if you roll off
the system much earlier will it be stable.
each fet now looks like a very fast, pF input
capacitance, DC-perfect device, and essentially disappears from the
overall loop dynamics.
Unfortunatly not. Consider a two stage amp circuit, one first stage, being
the conventional gain stage, followed by a second stage connected with a
local loop to make a CLG of unity. Additionally, the second amp connecting
back to the 1st (via a beta network if used) to form the composite. Now
calculate the system loop gain by breaking both feedback connections *at
once* an compare it with the 2 stage system both running open loop. The
transfer functions are identical. So, having the two stages means at least 2
roll offs. That is including an *additional* op-amp (to get better lf
performance) around the mosfet, will generate an additional pole that would
not have been there. Therefore the system is inherently more unstable.
I suggest you actually perform the *detailed* simulations of these type of
circuits.
I point out two circuits
http://www.kevinaylward.co.uk/ee/circuits/VeryLowDistortionAmp1.jpg
http://www.kevinaylward.co.uk/ee/circuits/VeryLowDistortionAmp2.jpg
They are similar, but one has a local feedback loop around the output
devices, one doesn't. One has better LF accuracy, one can be stabilised to a
higher UGF. Which is which?
The fundamental trade off here is basic physics, there is an inherent
constraint of H(Power, Accuracy, Speed) = 0, e.g.
http://www.kevinaylward.co.uk/ee/cmospafl/cmospafl.html
You just can't just get better LF performance, and expect to get it all. It
don't work that way. Like,....
When I was young, I prayed and prayed to the Lord to get me a bike. After
many years, and no success, I realised that HE, doesn't work that way...So I
stole the bike, and preyed for his forgiveness...
Now just pile on as many opamp+fet pairs as you
need. DC balance and current sharing become as good as the opamp
offset voltages, microvolts if you like, so fet gate threshold
variations and transfer curves don't matter any more. So use very
small source resistors and cut losses.
For LF, the op-amp approach is very nice, for HF, its not so nice.
My amps often work in pulse mode, when doing chemical NMR. They settle
to PPMs of the target value (which is current, since we're driving
gradient coils) in 10's of microseconds.
This is very slow, by about 2 orders of magnitude at least. At this slow
speed, op-amps are probably a good choice. You just clobber with a big cap.
By fast, I meant at the < 100ns level.
Kevin Aylward
www.kevinaylward.co.uk