Hi David.
You're a very astute observer. Indeed, the (optional) TF parameters are missing. Don't ask me how or why this happened.
They seem to be not that important, because simulation results from the Andy-Bob models were almost identical. Despite of that, I will use these models from now on.
BTW, the TR parameters are also missing.
Cheers,
E.
Bob's memory is correct, the default is TF=0 so the transistor Ft will be unrealistically high.
Since the issue was loop stability I expected that the Ft would be very important, that's why I made the smiley
joke. I am surprised that the simulation results are similar and would like to know more. Is the transistor Ft much different?
If so then some factor must restrict the high frequency response before the transistor Ft matters much. If it is the compensation then that implies there is still some freedom to optimize. As the circuit is optimized the transistor Ft (and thus TF) will matter more.
TR doesn't matter in your (or most audio) circuits, I think.
Best wishes
David
BTW A minor punctuation correction.
"You're a very astute observer. Indeed, ..."
should be
"You're a very astute observer indeed. ..."
Last edited:
Waly
I'm perplexed by your comment or more so the motivation behind it.
From what I see Edmond is provided his circuits and design evolution with insight and clarity. I'm really enjoying and learning from this thread, so I dont understand why you made your comment in such a negative manner?
Thanks
-Antonio
ft
Hi Bob,
edit: and David,
TF an TR parameters are optional in that you may omit them. In that case CJE, CJC, MJE and VJE are used for computation, though, admittedly, will yield different results (in particular when these parameters are not correct).
That's right, zero.
Perhaps that's true for your simulator, but regarding a 2SC3503 for example, I got no 'extreme high' values for ft, rather too high values, see below (ft in MHz):
As you see both methods differ from the data sheet values. Setting TF and TR to zero yields indeed optimistic results, though not that much. When a 2SC3503 is used as VAS (Ic = 5...10mA) the difference are almost meaningless. That's what I mean by 'optional'.
Nevertheless, I will use your models.
Cheers,
E.
Hi Bob,
edit: and David,
TF an TR parameters are optional in that you may omit them. In that case CJE, CJC, MJE and VJE are used for computation, though, admittedly, will yield different results (in particular when these parameters are not correct).
That's right, zero.
Perhaps that's true for your simulator, but regarding a 2SC3503 for example, I got no 'extreme high' values for ft, rather too high values, see below (ft in MHz):
Code:
Vce = 30V ic = 5mA Ic = 10mA Ic = 50mA
TF = TR = 0 ft = 75 ft = 103 ft = 249
TF= 1.6933e-10, TR= 1.0E-8 ft = 70 ft = 93 ft = 148
Sanyo data sheet ft = 105 ft = 150 ft = 200Nevertheless, I will use your models.
Cheers,
E.
Last edited:
annoying trolls
Hi Antonio,
That's why I've put this annoying troll (who can't even handle a simulator) on my ignore list.
The same applies to michaelkiwanuka, who, most likely, made a stupid remark again.
Thanks for your support.
Cheers,
E.
Hi Antonio,
That's why I've put this annoying troll (who can't even handle a simulator) on my ignore list.
The same applies to michaelkiwanuka, who, most likely, made a stupid remark again.
Thanks for your support.
Cheers,
E.
Have you built any of Edmond's ideas to confirm they are more than theoretical speculations, based purely on simulations using models of an unknown quality, and with ideally matched devices?
Speaking of theory, you may want to read the control theory bibliography recommended by jcx. You'll find out that all systems with the same number of degrees of freedom are (feedback wise) equivalent. Lurie concluded that given an open loop characteristic, and the same number of degrees of freedom, there is no silver bullet to increase the loop gain while keeping the same stability margins. If you can linearize the open loop system, that's fine, but otherwise embedding loops in loops in loops... won't help a iota. And systems with three degrees of freedom or more are of little practical importance, at least for discrete audio.
Not negative, just trying to keep things within the common knowledge and sense. Each time I see something like 12ppb (let's see, that's 0.0000012%) THD-20 performances I got itches.
Sorry for "trolling" again, I'm still not used to the fact that discussing design ideas is not Edmonds strongest trait, unless you start your message with "You are a Genius!".
Last edited:
ft & IMD
Hi David,
As a matter of fact, not all simulations were almost identical. The Miller loop, being the one with the highest ULGF (10..15MHz), is affected the most by the lower ft. As a result, the phase margin drops by about 20 degrees. Although the margin is still adequate, it's most likely that the frequency compensation needs some adjustments for optimal performance.
NB: I've updated the website with several more sims. The 19/20kHz IMD test shows a hard to believe low distortion of only 47ppb. So I ask everybody is there any amp (which I can simulate) with a known and accurate 19/20kHz IMD figure in order to validate my procedure for IMD simulation.
Cheers,
E.
Bob's memory is correct, the default is TF=0 so the transistor Ft will be unrealistically high.
Since the issue was loop stability I expected that the Ft would be very important, that's why I made the smiley
Is the transistor Ft much different?
If so then some factor must restrict the high frequency response before the transistor Ft matters much. If it is the compensation then that implies there is still some freedom to optimize. As the circuit is optimized the transistor Ft (and thus TF) will matter more.
TR doesn't matter in your (or most audio) circuits, I think.
Best wishes
David
[..]
Hi David,
As a matter of fact, not all simulations were almost identical. The Miller loop, being the one with the highest ULGF (10..15MHz), is affected the most by the lower ft. As a result, the phase margin drops by about 20 degrees. Although the margin is still adequate, it's most likely that the frequency compensation needs some adjustments for optimal performance.
NB: I've updated the website with several more sims. The 19/20kHz IMD test shows a hard to believe low distortion of only 47ppb. So I ask everybody is there any amp (which I can simulate) with a known and accurate 19/20kHz IMD figure in order to validate my procedure for IMD simulation.
Cheers,
E.
Omission of TF seems to have been partly cancelled by a model inaccuracy.
Not what I expected (not helped by the cross posts).
Ft lower than the data sheet is not what should happen if TF=0
The fact that it does happen at low current implies that Cbc is excessive.
I will try CJC lower and/or tweek some of MJC, VJC, XCJC.
Lucky the errors were mainly in the conservative direction so it didn't invalid the idea - Pity if such a nice circuit only worked with impossible transistors!
Best wishes
David
Hello Bob - while you are here, do you have any comments on post #253?
Best wishes
David
Hello Waly
>> Have you built any of Edmond's ideas to confirm they are more than theoretical speculations, based purely on simulations using models of an unknown quality, and with ideally matched devices?
I have built circuits that have been simulated by Edmond, and so has syn08 (=Ovidiu) as can be seen here.
PGP
Complex amplifier builds require a reasonable high level of skills to pull off successfully, but his circuits simulations are solid.
Regards
Arthur
>> Have you built any of Edmond's ideas to confirm they are more than theoretical speculations, based purely on simulations using models of an unknown quality, and with ideally matched devices?
I have built circuits that have been simulated by Edmond, and so has syn08 (=Ovidiu) as can be seen here.
PGP
Complex amplifier builds require a reasonable high level of skills to pull off successfully, but his circuits simulations are solid.
Regards
Arthur
FT parameter
Hi David,
Well, if a real ft, being say 50% lower than simulated, would jeopardize the project, I would have done a bad job.
In the final design phase I always check how the circuit behaves with different models.
A sound design should be tolerant to variations in transistor specs (within certain limits of course).
In the mean time, I'm looking at the impact from the modifications on models by Andy & Bob and also how I can compensate for the (small) loss of phase margin.
Cheers,
E.
Hi David,
Well, if a real ft, being say 50% lower than simulated, would jeopardize the project, I would have done a bad job.
In the final design phase I always check how the circuit behaves with different models.
A sound design should be tolerant to variations in transistor specs (within certain limits of course).
In the mean time, I'm looking at the impact from the modifications on models by Andy & Bob and also how I can compensate for the (small) loss of phase margin.
Cheers,
E.
I have read the books recommended by jcx. I found Lurie rather unclear, perhaps because he's a native speaker of Russian and it reads like a bad translation. Perhaps that has mislead you. Have you read Horowitz? Still idiosyncratic but clearer.
In particular he shows that extra loops connected to internal nodes of the amplifier allow more D.O.Freedom. So your point about "all systems with the same number of DOF" is true but not relevant. The extra DOF is precisely the theoretical foundation of TMC, I think. So systems with more DOF are of practical importance for discrete audio - at least if you want to do better than simple Miller compensation.
Best wishes
David
Last edited:
It was long shown (and even the reviver of TMC now seem to admit) that all being equal, TMC and other 2 DOF techniques (like two pole) allow the same amount of loop gain.
To put it simply, given a single pole open loop gain with a crossover frequency of F, the only way to squeeze out more loop gain at HF is to use N poles to roll of the gain before F and to add N-1 zeros to bring back the phase to the stability condition. Unfortunately, that barely works for N larger than 2, other than in theory, because of very practical reasons related to circuit sensitivity. Or even simpler put, a three gain stage amplifier is very difficult to compensate and stabilize. Cherry compensation used in PGP is the only notable exception that I am aware of.
Now, there are ways to further optimize multiple embedded loops, but they are so mind boggling I couldn't even think of a practical implementation in discrete circuits. Strictly speaking, a multiloop system doesn't have to have each and every feedback loop stable. All that it matters is the total number of circles around the critical point, in the phase space. That is, internal and external unstable loops can help each other to get an overall stable system. Here's an old article about this, now try and implement such a circuit with discretes
.But enough of this - have fun and don't forget to build your dreams!
Here's an old article about this, now try and implement such a circuit with discretes
But enough of this - have fun and don't forget to build your dreams!
That article is fun! Thank you. And "build my dreams" is exactly what this discussion is intended to help!
Best wishes
David
Hi David,
I forgot to answer this question. Sorry.
Generally I don't like current limiting resistors in series with the collectors. To be effective they have to be relative large, which creates a parasitic Miller loop. Admittedly, not a real issue, as the Miller effect can be neutralized by means of decoupling caps between the collectors (of the 'beta enhancement' trannies) and the supply rails.
There's one more reason I don't like it. In some applications, for example a CFB input stage, I'm not sure whether the maximum base current (seldom specified) will be exceeded or not.
Regarding incorporating the TF parameter, I have changed two resistor values of the amp of fig.12. Now the phase margins are as sound as before.
Cheers,
E.
Lurie's later book is more readable - but still buggy text, I still don't see a "method" to his nonlinear compensation, his website can still be found on Archive.org
Dr. Boris J. Lurie's Homepage: Classical Feedback Control
has some material from the new book
stepping up to "simple" nonlinear Popov/Circle "describing function" is just on the edge of my competence - I currently have little motivation to hurt my head with more advanced linear algebra
Atherton, Taylor seem to be good sources on Describing Function methods
http://www.eolss.net/Sample-Chapters/C18/E6-43-06-01.pdf
Google
while "proving" nonlinear system stability may be amusing I don't know how it really affects practical audio amplifier circuit design, relatively simple "nonlinear compensation" - which seems to end up looking like randomly sprinkled clamping diodes
I have managed to use Popov/Circle Criteria in my multiloop op amp analysis for a circuit with a region of 3rd order loop gain – to separately show stability with output saturation or input op amp slew rate limit – maybe these later techniques would let me show both together – but the hardware works - even without the clamps if you use fater slewing jfet front end op amp instead of the horrible OP07 I tried because they I had them lying on the bench
Dr. Boris J. Lurie's Homepage: Classical Feedback Control
has some material from the new book
stepping up to "simple" nonlinear Popov/Circle "describing function" is just on the edge of my competence - I currently have little motivation to hurt my head with more advanced linear algebra
Atherton, Taylor seem to be good sources on Describing Function methods
http://www.eolss.net/Sample-Chapters/C18/E6-43-06-01.pdf
while "proving" nonlinear system stability may be amusing I don't know how it really affects practical audio amplifier circuit design, relatively simple "nonlinear compensation" - which seems to end up looking like randomly sprinkled clamping diodes
I have managed to use Popov/Circle Criteria in my multiloop op amp analysis for a circuit with a region of 3rd order loop gain – to separately show stability with output saturation or input op amp slew rate limit – maybe these later techniques would let me show both together – but the hardware works - even without the clamps if you use fater slewing jfet front end op amp instead of the horrible OP07 I tried because they I had them lying on the bench
Last edited:
Dave totally agree regarding Bode.
I have the thirteenth printing fairly dog-eared, and indeed inspiring.
Thanks
-Antonio
Thank you very much. After the first book I wasn't inspired to seek out more, especially with the website apparently removed. But I shall now do so. The archived site is a find itself.
Actually it was a complaint I read about "randomly sprinkled capacitors" to fix stability problems that partly started this headbash. I am sufficient mathematician to realize that to optimize distortion/stability one should set up a nice matrix of feedback factors from node i to node j and then... but not the mathematician to work out what to do with it! But I retain the feel that simple one-loop Miller compensation can't be optimal. So I will study more Lurie.
Best wishes
David
This is drifting off topic, but at the risk of being again accused of trolling by our chief designer I should add that while the general Popov criteria indeed belongs to the non-linear stability analysis realm, for linear circuits there is a little known derived procedure called Sequential Return Difference (see also the other links on the page):
1. Open all feedback loops; the resulting system is now necessary stable.
2. Determine the loop gain of one feedback loop with all other feedback loops open. Count the number of clockwise turns around the Nyquist point.
3. Close the feedback loop in Step 2 and determine the loop gain of the next feedback loop. Count the net number of turns around the Nyquist point.
4. Continue the procedure until all loops are analyzed and their contribution to the net number of turns around the Nyquist point is determined.
5. If the total net number of turns (clockwise turns - counter clockwise turns) arount the Nyquist point equals zero, the system is stable. If the net is greater than zero, the system is unstable. If it's less than zero, there's a calculation error
.
Hi jcx,
I would appreciate if you comment on fig.12c on my website.
As you see, the 3rd order loop is conditionally stable, as the phase margin drops to -45 degrees at 100kHz. Till now, I didn't observe any instabilities, but maybe I'm overlooking something. The only deficiency is the driver, who can't cope with extreme fast and large transients. So I have to limit the rise and fall times to 1.2us by means of an input filter (no problem at all, as one needs a filter anyhow). As a result, the amp slews with a 'meager' 100V/us at full output swing.
Any hints to catch hidden bugs?
Cheers,
E.
edit: I've found one minor bug: the amp rings like hell at ~130kHz with supply voltages between 5 an 10V.
Ditto with TMC disabled. Gone when error FB in the OPS is disabled.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.