Update. This thread now covers other advanced stability analysis techniques.
The Middlebrook GFT is supposed to be the "Final Solution" to feedback measurement and I have started to experiment with it in LTSpice.
So far I have created a simpler version of Frank Wiedmann's implementation that seems to work - it returns the same results.
But I find the interpretation of these results is not always clear.
Anyone else have any examples of GFT use?
Best wishes
David
The Middlebrook GFT is supposed to be the "Final Solution" to feedback measurement and I have started to experiment with it in LTSpice.
So far I have created a simpler version of Frank Wiedmann's implementation that seems to work - it returns the same results.
But I find the interpretation of these results is not always clear.
Anyone else have any examples of GFT use?
Best wishes
David
Last edited:
I have Middlebrook's ICAP manual but the examples are rather limited.
I will recheck their site in case there's more.
There's probably some on the Yahoo LTSpice forum but I dislike that interface so much, hoped there was someone here.
I don't have the Vorperian book. Worth the effort to obtain it?
I have a tidy diff to common mode probe I can send if you haven't already done one.
Best wishes
David
its been a long time since I opened Vorperian's book
my memory is that you needed to jump in, work the problems and I found the writing the null injection equations "from inspection" opaque
probably good in end the end if you do the work, not so good for skimming, getting an overview then picking just one later subject
my memory is that you needed to jump in, work the problems and I found the writing the null injection equations "from inspection" opaque
probably good in end the end if you do the work, not so good for skimming, getting an overview then picking just one later subject
I went to the library anyway, while I waited for your response, and my impression is similar to yours.
But I think this is just the natural consequence of a deep topic.
One always needs to work some problems to really understand any subject with some depth and subtlety.
So, for the record, if anyone else is ever sufficiently interested in the future to find this thread in a search, I think it's a pretty worthwhile book.
Hi Arthur
Not yet! I want to understand the theory completely first.
Harry Dymond says the Bode 100 from Omicron Labs works well for an inexpensive instrument.
Its about $6000 so I assume that it's "inexpensive" for someone else's money, University probably.
Waly posted a link to a nice thesis project to build a GFT probe but they didn't quite finish, so it looks to be a several man-year project.
Waly also wrote that he had done a probe himself but suddenly went quiet when I asked for details.
Best wishes
David
I noticed you sent me a PM but my mailbox was full. Fixed now if you want to resend.
Last edited:
The simulated THD@20kHz is about 2 PPM at 100W into 8 ohms with 1 pair, one 4302 and one 4281 output transistor.
This is the baseline, 100W per pair, more or less in accord with Cordell's thermal limit. He calculates just a bit lower than this but assumes mediocre thermal pads.
I haven't added extra transistors just to overwhelm the distortion with sheer number of devices
Best wishes
David
Low resistance in the LTP. 33 ohms.
Then effectively halved because of the complementary pair.
Slew rate maintained because not Miller compensated.
Low resistance in the feedback network too. 22 ohms.
This pulls down the resistance of that side of the circuit, similar to a so-called "CFA", and pushes any LTP pole out.
Costs dissipation in the feedback network of course, acceptable for a tweeter driver that is not heavily driven.
Best wishes
David
Amp questions
Hello Dave,
With regards to the compensation of the main amplifier I am assuming higher than 1st order, is this a correct assumption. With the CMCL is this compensated by a a higher than 1st order . What is the loop gain at 1KHz and 20KHz af the amp.
What are you doing on the input stage to eliminate the input bias current. Are you matching the Hfe of all the input transistors so the NPN and PNP input bias currents cancel themselves out to a very small insignificant level or are you doing something else
Regards
Arthur
Hello Dave,
With regards to the compensation of the main amplifier I am assuming higher than 1st order, is this a correct assumption. With the CMCL is this compensated by a a higher than 1st order . What is the loop gain at 1KHz and 20KHz af the amp.
What are you doing on the input stage to eliminate the input bias current. Are you matching the Hfe of all the input transistors so the NPN and PNP input bias currents cancel themselves out to a very small insignificant level or are you doing something else
Regards
Arthur
The main loop is based on Bode's ideas. Approximately TPC but with some extra zeros to cancel out some transistor poles and a zero in the feedback network to increase stability.
Thanks for this question, it made my half realized ideas more explicit.
The CMCL loop does not need or benefit from ambitious compensation, as far as I can see, and backed up by my simulations and what I have seen from Edmond.
The natural response of my CMCL is very close to first order just from transistor roll-off in a fairly simple circuit..
This is a very tame loop and I have not added any extra components to modify this, saves a capacitor
I will post the loop plots.
Yes. The NPN and PNP are pretty closely matched in the batches I have.
Maybe luck but is probably due to that famous Japanese quality control.
I will add a fine trim in the final version.
Best wishes
David
Hello Ian
Not yet. There's a neat idea from a friend in it and I am not sure what to do, publish in Linear Audio or else show it here first and lose the exclusivity but maybe help clarify some ideas.
I feel I am pretty close but still don't quite see how to be sure it is optimal.
In particular I don't see yet how to apply the Middlebrook GFT to the OPS.
Best wishes
David
By the way, a while back I recall that we discussed the weird phase response that LTSpice sometimes shows for a feedback loop.
I now believe this not just an LTSpice failure. At least that it reflects a real anomaly, just presented rather unclearly.
Did you ever find out more about this?
The ICAP manual and Dr Middlebrook's site have been my main references too.
Have you worked out how to apply the GFT in LTSpice to an OPS?
Best wishes
David
Hi David,
No, sadly I didn't find out anything more about what caused that phase issue. I'd be interested to hear your thoughts about the anomaly that causes it.
By the way, I did have practical success with the output stage pole cancelling technique (whereby a series R-C is placed across the VAS emitter resistor). Distortion was reduced to around the level of the spec of the measurement analyser. It was interesting, that I could make the amplifier pass every stability test I normally use, however, if the power supply voltage was halved, then oscillation at a particular output amplitude on the sinewave occurred. I suspect that when the supply voltage was halved, the output stage pole 'moved', since, its location is dependent on Cgd, which in turn is dependent on Vds. It was good fun, trying to cancel a moving target, but at least I can confirm that the technique works in practice.
I didn't go on to complete the build of the full stereo amplifier chassis, so I cannot comment on the sound quality of an output inclusive design. I am generally of the opinion that all well designed amps sound pretty similar, and since I already have a decent quality amp, it seemed pointless building another. For me, it was more of a learning exercise.
Best wishes,
Ian
Phase is a bit arbitrary, do you call it -180 or + 180.
It seems that if there is a loop with 2 contributions then LTspice somehow picks a reference phase that may not be "sensible".
It always seems to indicate some kind of null or cancellation, so I take it as an alert that there is some issue that should be studied.
Yes I find the VAS emitter resistor bypass useful too.
I must admit I didn't try it with drastically reduced power supply rails.
Your explanation makes sense but is it a problem?
Feucht has some stuff on the web about EF output stability, have you read it?
There's a few bit's there that I find a bit obscure too.
Best wishes
David
Last edited:
Hi David,
Thanks for your thoughts on the phase issue. I'll look closer next time I see it in spice.
I've read some Feucht articles, but I don't own his book - 'Analog Circuit Design'. It seems well regarded, so I may buy a copy.
(My recent theoretical hobby work has been focusing on trying to get blameless levels of distortion (or better) using the same or less transistors than a blameless. I have found that input inclusive compensation combined with TMC is a very effective technique. With this compensation scheme, it is possible to use a front end LTP with very little degeneration.)
I know you are keen to develop amps with very low noise, so minimal degeneration is highly desirable. The only downside of less LTP degeneration that I can think of, is less immunity to RF (more degeneration allows a larger RF signal before nasty rectification occurs). Are you concerned about anything else related to a high gm LTP?
BTW, sorry that I'm way off-topic.
Best wishes,
Ian
Thanks for your thoughts on the phase issue. I'll look closer next time I see it in spice.
I've read some Feucht articles, but I don't own his book - 'Analog Circuit Design'. It seems well regarded, so I may buy a copy.
(My recent theoretical hobby work has been focusing on trying to get blameless levels of distortion (or better) using the same or less transistors than a blameless. I have found that input inclusive compensation combined with TMC is a very effective technique. With this compensation scheme, it is possible to use a front end LTP with very little degeneration.)
I know you are keen to develop amps with very low noise, so minimal degeneration is highly desirable. The only downside of less LTP degeneration that I can think of, is less immunity to RF (more degeneration allows a larger RF signal before nasty rectification occurs). Are you concerned about anything else related to a high gm LTP?
BTW, sorry that I'm way off-topic.
Best wishes,
Ian
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.