Simulation Analysis of several unique Allison-based output stages.

[keantoken]

Hi all.

Simulation file is in the .zip file I have uploaded. Models for the transistors are included in the netlist and you must use LTSpice.

I'm no expert, mostly because I don't have very much experience. However, I still like to contribute things like this every once in a while simply to socialize and also in the hope that others will find it educational. Knowledge is power. Maybe it's some teaching spirit showing through.

Anyways, you may be inclined towards skepticism upon seeing the figures, so to coax your curiosity here is Sander Sassen's ExtremeA, which uses the Allison output to great effect.

http://www.hardwareanalysis.com/cont...diy-amplifier/

The distortion figures of this output stage neatly line up with the figures obtained from this ExtremA. So it is likely that the simulator is not spouting nonsense.

Also I believe some (many?) people do not know how to use LTSpice, the simulator I am using, correctly. When used correctly, the simulator can be a powerful friend, but only if used correctly. If someone is interested, I can compile a PDF detailing the proper use of LTSpice. Usually you can get questions answered quickly from the yahoo group anyways though.

Here is the schematic:



All simulations are done at 34V pk-pk output, into an 8-ohm load. 36V rails.

Starting from the right in the schematic, going towards the left:

Circuit 1: Standard Allison

This is the normal Allison output stage. Despite all modifications, this version still has the lowest distortion.

THD @ 100Hz is .000120%
THD @ 20KHz is .000182%

Circuit 2: Allison with my modification, thus: Allison-Keane (not trying to unneedingly claim credit, but we might as well name it something. If it already has gone by some other name then I will happily release my title)

This modification allows one to use smaller emitter resistors, increasing maximum output power if you're using low bias current. It's useful for minimalist amplifiers for squeezing the most out of low supply voltages. Currently, it doesn't make much sense with such high bias current.

THD @ 100Hz is .000351%
THD @ 20KHz is .000528%

Circuit 3: Allison-Keane+Diamond

It may help your understanding of this circuit if you note that the additions have created two Rush Cascodes. As far as I know, I am the first to do it this way.

It cannot be innocently compared to circuit 2, since this version is not cascoded. However, it provides an indicator of the effect of cascoding if we compare it to circuit 4. We find that cascoding reduces distortion.

The main advantage of this circuit is it's rather high input impedance. This lowers distortion by loading the VAS less. This means that the distortion will increase less when the power level increases, and that you can set open-loop gain lower which will make the circuit more stable.

THD @ 100Hz is .000256%
THD @ 20KHz is .000292%

Circuit 4: Cascoded Allison-Keane+Diamond

This version of Circuit 2 serves to point out the effect of cascoding.

THD @ 100Hz is .000169%
THD @ 20KHz is .000222%

Afterword:

This is great and all, but if you actually want to give it justice in a proper amplifier, you will need to hook it up with a really good VAS. Good luck.

I hope this has been helpful and educational to viewers.

Bonzai,
- keantoken

[h_a]

Hi,

very kind of you to share your work here!

Let me however say 2 things that come immediately to mind: first, posting THD-100 and THD-20k does not make much sense to me. The general THD-figures one sees everywhere are for 1k , so your numbers are not directly comparable. THD at 20kHz has also only academic interest since humans can't even hear the first higher order frequency (40 kHz). 100 Hz is not reproduced by small bookshelf speakers. At 1 k the ear is most sensitive, so maybe you could post THD-figures for this frequency next time

2nd, at this low level wiring and pcb-artwork become crucial; also it would be IMHO essential to check simulated results against a real world prototype. This is not directly addressed at you, but more to the others also reading this thread. One does not get 0.00x % at full power easily, no no.

Nice work!

Have fun, Hannes

[bear]

THD @20k is a good idea.

The 1kHz. is a standard.

We are interested in performance, not just what people can or can not hear. The 20kHz. performance can be indictative of how it will perform in the areas that we normally can hear.

_-_-bear


PS. do we have a citation for this "Allison" person/circuit?

[KSTR]

Originally posted by bear
PS. do we have a citation for this "Allison" person/circuit?

http://www.diyaudio.com/forums/showt...33#post1356833

-------:-------

I've never seen this with CFP's, @keantoken, but then again, what do I know. With plain EF's (like in the ExtremA) it might be more tolerant to adverse operating conditions, like clipping and especially transition into class-B, which is likely to happen during transients. See the ExtremA thread here for a little tweak (also only in virtual reality) I found that improves the transition glitch of the original OS a bit.

At any rate, the results are nice, although this level of performance is to be somewhat expected with class-A and big amounts of tight feedback.

- Klaus

[keantoken]

I thank everyone for their kind comments.

I have been thinking that since I have plenty of time on my hands and have been working on LTSpice for 3 or 4 years, I could do more simulations like this and post the results. Sometimes you read lines on this forum saying "does anyone want to compare these in the simuator?" and it never gets done. So feel free to ask.

Originally posted by h_a
Hi,

very kind of you to share your work here!

Let me however say 2 things that come immediately to mind: first, posting THD-100 and THD-20k does not make much sense to me. The general THD-figures one sees everywhere are for 1k , so your numbers are not directly comparable. THD at 20kHz has also only academic interest since humans can't even hear the first higher order frequency (40 kHz). 100 Hz is not reproduced by small bookshelf speakers. At 1 k the ear is most sensitive, so maybe you could post THD-figures for this frequency next time

2nd, at this low level wiring and pcb-artwork become crucial; also it would be IMHO essential to check simulated results against a real world prototype. This is not directly addressed at you, but more to the others also reading this thread. One does not get 0.00x % at full power easily, no no.

Nice work!

Have fun, Hannes

Thank you for pointing this out. I didn't post 1KHz THD figures because they're exactly the same as the 100Hz figures. I take it for granted that the THD begins to rise somewhere around 5KHz, which is why I post 100Hz and 20KHz figures.

The Allison and its derivatives obviously are not only suitable for audio. Given some good transistors, it will do low THD into the hundreds of kilohertz.

But as you already pointed out, PCB art will definitely have a profound influence on the performance.

Originally posted by KSTR

At any rate, the results are nice, although this level of performance is to be somewhat expected with class-A and big amounts of tight feedback.

- Klaus

There is also the possibility to not include the output stage in the feeback loop. Just have a VAS connected to the input and nothing else. This would eliminate the need for so much feedback and would also make the necessary stability precautions minimal, although you wouldn't be able to cancel out the distortion of the output stage.

- keantoken

[keantoken]

Here is the simulated input impedance of all circuits:

Circuit 1: 117KOhms

Circuit 2: 120KOhms

Circuit 3: 2MOhms (impedance will be dependant on the beta of the diamond transistors)

Circuit 4: 17MOhms (cascoding?)

Important note:

These circuits can only be compared if the same devices are used, and even then, only relatively. The performance will vary depending on the performance of transistors used, but will usually stay within proportions.

- keantoken

[bear]

I'll ask again, what is the origin/source of the "allison" circuit??

Also, what is the relationship between what you posted and the ExtremA circuit?

Maybe I am dense, I don't see it. Also maybe I am ignorant (probably) but I am unaware of "Allison"... maybe except for Howard Stern's ex-wife?

_-_-bear

[keantoken]

Originally posted by bear
I'll ask again, what is the origin/source of the "allison" circuit??

Also, what is the relationship between what you posted and the ExtremA circuit?

Maybe I am dense, I don't see it. Also maybe I am ignorant (probably) but I am unaware of "Allison"... maybe except for Howard Stern's ex-wife?

_-_-bear

Originally posted by KSTR
http://www.diyaudio.com/forums/showt...33#post1356833

-------:-------
- Klaus

In my schematic, the bias control resistors on the far right circuit are the Allison multiplier. Sander's output stage uses the same Allison Vbe multiplier scheme.



The 3 other circuits (in my schematic from first post) are my own concoctions.

- keantoken

[KSTR]

Hi,

WRT input impedance, I like the way you tried to get a high Zin. OTOH, both in sims and in real life I found that one still needs low Z drive (some 10's of Ohms or less) to actually retain the low distortion. If you care to try, pls. re-run your sims with, say, 10kOhms drive impedance. IHMO a FET input is needed for a usable high-Z input.

Er, have you done stability analysis (loop gain probing, step response etc), with reactive loads etc? I mean, in your schematics I don't see any comp caps, base stoppers and the like. Are those the circuits you actually simmed or are they simplified ones, for ease of illustration?

- Klaus

[bear]

Doing my own finding out: http://www.diyaudio.com/forums/showt...33#post1356833

Shows a source of the Allison design

However, that one shows a cap across those vbe control transistors and the text talks about it working only at LF... So the Sassen design without the cap seems to perhaps work somewhat differently and respond to HF??

_-_-bear