Simulation Analysis of several unique Allison-based output stages.

[darkfenriz]

Series diodes will spoil everything, furthermore, exact comlementary pairs are not needed here.

[keantoken]

Ken has a good idea about using matched Fets.

I can't vouch for the THD (though it's below .01%), but this one has virtually infinite input impedance!

EDIT: output impedance is also .05 ohms!

- keantoken

[kenpeter]

Perfect. We got folks who deliver and pickup stuff to TI every day.
I also work within spitting distance of the Plano Fry's.

[darkfenriz]

Originally posted by keantoken
Ken has a good idea about using matched Fets.

I can't vouch for the THD (though it's below .01%), but this one has virtually infinite input impedance!

EDIT: output impedance is also .05 ohms!

- keantoken

this will work properly only for JFETs with Idss<6mA




Edit: after some analysis it comes to me, that this (the whole scheme) will not work very properly in class AB. Two separate feedback loops will fight against each other resulting in either crossover distortion or hard cross conduction.
Note, that for class AB, the total drop on emitter resistors for high load curent is much higher than for idle conditions.

[kenpeter]

For AB, you need Schottky diodes, or other nonlinear resistances.
The fixed voltage span between two corrected errors means that
fixed resistor currents will always sum to a constant. AB requires
quiescent current sum less than peak, and thats in direct conflict...

Anyways: So what if diode non-linearity results in "error"? This
does not screw anything up. As long as it is confined narrowly
between the offsets, it can't go anywhere. It just makes the AB
action real smooth, smoothest crossings you have ever seen.

This is a hard crossing?

[darkfenriz]

OK, I see the point now.

[keantoken]

Originally posted by kenpeter
For AB, you need Schottky diodes, or other nonlinear resistances.
The fixed voltage span between two corrected errors means that
fixed resistor currents will always sum to a constant. AB requires
quiescent current sum less than peak, and thats in direct conflict...

Anyways: So what if diode non-linearity results in "error"? This
does not screw anything up. As long as it is confined narrowly
between the offsets, it can't go anywhere. It just makes the AB
action real smooth, smoothest crossings you have ever seen.

This is a hard crossing?

I wonder if we could use some other device, such as a thermistor, for the "nonlinear" resistances...

I think you should compare your AB Allison side-by side with the usual darlington EF + Vbe multiplier to see how they compare.

- keantoken

[kenpeter]

In designs with 2 emitter drop span, you can add a front panel
switch to short out the AB Schottky's with .22ohms on demand.
The wire and switch together might even be that many ohms?
Of course, its gonna run a bit hotter in full A.

Not sure which way temperature coefficient of a power Schottky
tries to bend? Need to grab a spec sheet or two and find out.

Just checked, and it goes the wrong way. Hotter raises quiescent.
0.1 ohms pure resistance I left for insurance might be insufficient?

Or just make sure the Schottky's stay cool. Its not like they are
gonna burn more than a half Watt most. And they have big tabs.

[keantoken]

I went off on a tangent, looking into superfast output stage topologies.

I then wanted to know how the Allison might compare bandwidth-wise if I used all 5769/5771 transistors.

The result, is a 1.5Mohm input impedance at 1MHz, 5V pk-pk, and .0027% THD at that same output level!

Into 10Kohms, of course.

Fest your eyes on the beauty. I can now be a zero THD1000 exponent.

I think the usage of bootstrapping here is neat. Obviously, the bootstrap caps are not large because I was only testing it at 100MHz+. 100nF caps can go here if you're working around 1MHz. The bootstrapping is far better an option than a discrete CCS because it has virtually no parasitic capacitance (I consider it necessary).

The 100 ohm resistors help stability a lot. Otherwise, it could oscillate around 500MHz.

- keantoken

[keantoken]

This design still has nonlinear input impedance, though.

THD at 10MHz into 10k load, 10V pk-pk is .0027% (I misquoted myself earlier).

THD at 10MHz into 10k load, 10V pk-pk with 1k source impedance is .049%

BUT.

THD at 1MHz into 10k load, 10V pk-pk with 1k source impedance, is only .0037%

With 0 ohm source impedance, and the same output levels, this circuit gives .001% THD at 1MHz.

Maximum input impedance is about 2.48 megaohms, which begins to roll off at 200KHz. LF impedance will depend on bootstrap caps but will likely not be able to exceed the 100KHz impedance with modifications, without negatively affecting distortion or bandwidth.

Performance can be improved much by decreasing R8, so that Q7 and Q8 operate farther from 0mA. I consider the current value a safe one.

- keantoken