Hi Mike,
I have verified the same thing with my own simulations, where there is a substantial peak in the total loop gain around the output stage (as seen by the output stage). This is quite interesting, but is certainly not any kind of hard evidence that TPC is somehow superior to TMC, which is what you have been arguing.
Moreover, bear in mind that this peaking behavior when seen in the total output stage loop gain with TMC, is arrived at via two feedback loops surrounding the output stage. This in my mind is a significant architectural distinction from TPC (which only has one loop around the output stage).
The important question remains, which approach, in practice is better, where better's definition remains somewhat of an issue of contention.
If we adjust each of the three techniques individually so that the output stage sees the same phase margin and at least, say 6dB gain margin, and then see which one delivers the lowest distortion, maybe that is a means of fair comparison. In such an approach, not all of the designs would have the same gain crossover frequency, either around the output stage or in the amplifier's global loop. We would essentially be saying which one gives lowest distortion for the same amount of stability as seen by the output stage.
Of course, in my view, we would first do this without any use of a lead network in the global feedback path. If we subsequently wanted to allow comparisons using a lead network, we would then allow its use in any of the three compensation approaches.
Cheers,
Bob
Hi Edmond,
Initially tau1 and tau2 were just the time constants of the lead type transfer function in the equation for I3 (see my third handwritten page). From the beginning I hadn't considered what the transfer functions from the two inputs of the front end to the output actually were, but I continued by deriving the transimpedance of the TPC VAS in figure (8), a front end equivalent to the TMC version, which is
Z(s) = - D (s tau3 + 1) / (s^2), where tau3 = R (C1+C2) and D = 1 / (R C1 C2)
the output voltage of the front end then becomes
Vd = Z * I3 = D (s tau3 + 1)/(s^2) gm [Vin - Vfb/A (s tau2 + 1)/(s tau1 + 1)]
but tau1 from the I3 expression is the same as tau3 because
tau1 = k R C1 = R (C1 + C2) = tau3
which makes the front end output voltage
Vd = D gm [ (s tau1 + 1) Vin - (s tau2 + 1) Vfb/A ] / (s^2)
for the TMC and the calculated equivalent TPC+lead version.
I realized later that there is another practical way to implement a TMC equivalent TPC circuit which does not require a lead network in the feedback path, and that is to select another TPC VAS transimpedance Z' with tau3' = tau2 instead of tau1.
Then a lag network with transfer function (s tau1 + 1)/(s tau2 + 1) is inserted in the input signal path to the LTP instead of the lead network in the feedback path to give the same Vd.
Vd = Z' * I3' = D' (s tau3' + 1)/(s^2) gm [(s tau1 + 1)/(s tau2 + 1) Vin - Vfb/A]
which also becomes, because tau3' = tau2,
Vd = Z' * I3' = D' gm [ (s tau1 + 1) Vin - (s tau2 + 1) Vfb/A ]/(s^2).
But after that, you may want to consider removing the zero at tau1, which shows up in the closed loop response, by inserting a standard low-pass filter instead with transfer function 1/(s tau2 + 1). If the impedance of the source feeding the power amplifier is low or known this could just be the standard input slew rate/RF ingress limiting filter.
Cheers and happy new year everyone! (Skål och gott nytt år till er alla!)
Last edited:
No. What i have been arguing (and provided proof in simulation) is the fact that the total loop gain about the output stage (that due to the major loop and minor loop) alone with TMC is the same as the major loop gain about the whole amplifier with TPC.
i.e: TMC does not improve the total loop gain about the output stage over that provided by TPC. This is why TPC is superior to TMC because this same loop gain is available for the whole amplifier not just the output stage.
As relentlessely said , TPC load the VAS and has between
6 to 15 db less OLG to start with compared to TMC...
It seems that you didn t check this point, otherwise you wouldn t
insist in this claim, which hold more of some sort of dogmatism
that from some basical scientific approach...
Last edited:
yes, it s two posts above....
For the OLG difference to be noticable when closing the loops,
that is a different story.
As pointed many times, THD reduction seems comparable for both
compensation scheme, although TPC seems to have an advantage
in some cases that wasn t pointed yet in this thread, but i just wait
for some more elements before coming to conclusions..
The remaining debate is wether TMC ot TPC has "better" step
response/stability along with a misunderstanding about the initial
conditions, i.e, the values of network that optimize the said compensation
schemes, as using the same values for both TPC and TMC seems to me
more of an exercise of style than a clever way to find what exactly
can be extracted from the two contenders...
Lets not “do” the naïve, Audiophoolish, Sophomoric time/frequency thing again, please
to stretch/extend the analogy the "coin" is infinitely thin so the reverse follows exactly the contour of the obverse - ie fourier and time domain convey exactly the same information being "duals" and it is only a question of ease of human interpretation that that one would choose between them depending on what you want to "see"
I hope everyone here has a some what more sophisticated knowledge of Signals and System but I get really really tired of people thinking they're saying something meaningful when thay try to discredit the "meter readers" for "only looking at sine waves" - or fft, or THD...
...they're all tools with strengths and weaknesses
and the formal "time domain" system theory tool is "Modern Control Theory" State Space representation - are you seriously proposing to introduce that into a diy forum? - I don't think I've ever seen a practicing engineer use state space tools for amplifier circuit design
to stretch/extend the analogy the "coin" is infinitely thin so the reverse follows exactly the contour of the obverse - ie fourier and time domain convey exactly the same information being "duals" and it is only a question of ease of human interpretation that that one would choose between them depending on what you want to "see"
I hope everyone here has a some what more sophisticated knowledge of Signals and System but I get really really tired of people thinking they're saying something meaningful when thay try to discredit the "meter readers" for "only looking at sine waves" - or fft, or THD...
...they're all tools with strengths and weaknesses
and the formal "time domain" system theory tool is "Modern Control Theory" State Space representation - are you seriously proposing to introduce that into a diy forum? - I don't think I've ever seen a practicing engineer use state space tools for amplifier circuit design
Last edited:
We had specific courses on control theory when I was in school and IIRC it was not taught out of the EE department since it had application to many fields. Ogata was the text book and I was impressed with both the text and the course:
Amazon.com: Modern Control Engineering (5th Edition) (9780136156734): Katsuhiko Ogata: Books
Do you have a preferred text?
"there is nothing so practical as a good theory"
the course I took was taught by the Mech E dept and used Ogata (1st ed)
I've since bought Dorsey's, Goodwin's control theory books
you need the "cultural" knowledge of State Space to read any recent control literature and many fundamental properties of control theory are expressed as proofs in advanced Linear Algebra using State Space representation - many theorems have extensions to nonlinear systems that have only been proven in the last decade or so - and at those levels my knowledge is strictly "cultural"
but I think that for problems that can be adequately represented as SISO LTI then "Classical Control Theory" growing from Bode's work is far more accessible/useful - and some advanced results including nonlinear techniques are usable in "Classical" form (Popov, Small Gain Theorem's Circle criteria)
I try not to miss any chances to recommend BJ Lurie's work - although his books are hard to understand and "buggy" - needing 2nd editions but he really shows how to use Classical Control techniques - you can still view his old site with archive.org Dr. Boris J. Lurie's Homepage: Classical Feedback Control
one thing Lurie does really well is show that the “conservation” relation for the total amount of feedback - the “Bode Integral” is exactly such a practical "good theory" - and has been the underpinning fundamental argument behind my posts in this thread
http://trs-new.jpl.nasa.gov/dspace/bitstream/2014/19495/1/98-0905.pdf
Happy New Year, and may all of your nonlinearities be Locally Lipschitz
the course I took was taught by the Mech E dept and used Ogata (1st ed)
I've since bought Dorsey's, Goodwin's control theory books
you need the "cultural" knowledge of State Space to read any recent control literature and many fundamental properties of control theory are expressed as proofs in advanced Linear Algebra using State Space representation - many theorems have extensions to nonlinear systems that have only been proven in the last decade or so - and at those levels my knowledge is strictly "cultural"
but I think that for problems that can be adequately represented as SISO LTI then "Classical Control Theory" growing from Bode's work is far more accessible/useful - and some advanced results including nonlinear techniques are usable in "Classical" form (Popov, Small Gain Theorem's Circle criteria)
I try not to miss any chances to recommend BJ Lurie's work - although his books are hard to understand and "buggy" - needing 2nd editions but he really shows how to use Classical Control techniques - you can still view his old site with archive.org Dr. Boris J. Lurie's Homepage: Classical Feedback Control
one thing Lurie does really well is show that the “conservation” relation for the total amount of feedback - the “Bode Integral” is exactly such a practical "good theory" - and has been the underpinning fundamental argument behind my posts in this thread
http://trs-new.jpl.nasa.gov/dspace/bitstream/2014/19495/1/98-0905.pdf
Happy New Year, and may all of your nonlinearities be Locally Lipschitz
Last edited:
Hi Mike,
You did not prove your point with a whole-amplifier simulation, which I did. A loop gain analysis alone is only a part of the whole amplifier performance contributor.
Let's see some amplifier simulations from you here. You are welcome to use my simple amplifier as the guinea pig, and then it will be easy for others to assess and interpret your results.
Cheers,
Bob
Tilting at windmills
This is true only when the phase information from the FFT is included. After all, when I wrote my first FFT routine in assembler the output came in complex form. However I've never seen the complex output shown in plots on this forum. Have you? I was referring only to the common display of magnitude.
Get tired all you like - creating imaginary enemies must be rather wearing.
When writing what I wrote, I was thinking of one specific case - those who purport to read 'noise floor' from the 'baseline' of an FFT without reference to the measurement bandwidth. There does seem to be a common (mis)perception that the output of an FFT is a graph (rather than a histogram) which must contribute to this nonsense about 'noise floor'.
This is true only when the phase information from the FFT is included. After all, when I wrote my first FFT routine in assembler the output came in complex form. However I've never seen the complex output shown in plots on this forum. Have you? I was referring only to the common display of magnitude.
Get tired all you like - creating imaginary enemies must be rather wearing.
When writing what I wrote, I was thinking of one specific case - those who purport to read 'noise floor' from the 'baseline' of an FFT without reference to the measurement bandwidth. There does seem to be a common (mis)perception that the output of an FFT is a graph (rather than a histogram) which must contribute to this nonsense about 'noise floor'.jcx,
The Fourier transform can be useful for analyzing radar signals, microwave signals and stuff like that.
Heisenberg's uncertainty principle states that the momentum and the position of a moving particle cannot be known simultaneously, likewise, spectral components cannot be known at any given instant in time. Some of us may find that harmonic distortion measurement is only meaningful with a (truthful) spectral representation. On the other hand, not knowing the spectrum of a low THD amp could be an advantage from a psychological standpoint as it´s not a very beautiful spectacle...
The Fourier transform can be useful for analyzing radar signals, microwave signals and stuff like that.
Hi WuYit,
When discussing time domain and frequency domain departures from perfection, it is important to bear in mind their relationships when the system under discussion and the imperfections under discussion are quite linear. This is the well-known relationship for minimum-phase systems, which most audio amplifiers approximate extremely well, especially in the ranges of frequencies of greatest concern below 100 kHz. One does not need to discuss Fourier transforms and uncertainty principles to understand this.
The bottom line is that for a minimum phase system operating in its linear region, the time and frequency response characteristics are virtually inseparable. If one is degraded, the other will be degraded. If one is very good, the other will be very good.
This does not mean that we should abandon both forms of measurement or analysis. Often, one form will lend some insight that was lacking in the other domain. This is why we routinely look at squarewave response as well as frequency response, for example. Nevertheless, if one is very good, the other is likely to be very good, as long as the system is linear.
It is extremely important that we do not confuse linear distortions with nonlinear distortions. This can be a semantical trap leading to needless misunderstandings and arguments. Linear systems can produce linear distortions, like group delay distortion; this kind of distortion does not create new frequency spectra. These kinds of linear distortions, albeit callled distortions, still obey the time-frequency releationship rules if the system is of minimum phase. Nonlinear distortions, like THD, are a different beast, and do create new frequency spectra.
If the system is non-minimum phase, as with an all-pass network included, then the relationship between time domain performance and frequency domain performance is broken. Also, if the system is rather nonlinear, this relationship between the time domain and frequency domain may be broken. All bets are off when an amplifier is experiencing slew rate limiting, for example.
Cheers,
Bob
Hi Mike,
Please post an up-to-date, ready-to-run, simulation here so that everyone can easily reproduce and evaluate your results. The easiest path for you to follow to accomplish that is to just put your own compensation components into the amplifier that I posted, but that is entirely up to you.
At the same time, in that post it would be helpful for you to quote your comparative distortion results.
Cheers,
Bob
Hi Bob, Happy New Year!
the domain properties conflict with each other.
It´s a very bad sign when a simple control function requires such a comprehensive calculation. (Especially so for a poorly observable dynamic system with number of unknown or uncertain parameters).
Excuse my question, but what is the second branch of your amplifier doing besides introducing distortion and phase errors?
the domain properties conflict with each other.
It´s a very bad sign when a simple control function requires such a comprehensive calculation. (Especially so for a poorly observable dynamic system with number of unknown or uncertain parameters).
Distortions don`t come alone.
Excuse my question, but what is the second branch of your amplifier doing besides introducing distortion and phase errors?
Hi WuYit,
Could you clarify what part of my amplifier circuit you are referring to as the second branch? For example, is it the second feedback path formed by the TMC circuit where the output of the amplifier is coupled back to the input of the VAS? Or is it some other branch you are referring to?
When you say "besides introducing distortion..." can you be specific about the kind of distortion you are referring to, i.e., linear distortions or nonlinear distortions?
Cheers,
Bob
evidence?
That's no sim, that are schematics without any useful comment. No graphs, no distortion figures, no phase margins, step response, etc.
The only thing you have proved with that post is that you don't understand a iota of TMC. Maybe C2=C4=200pF is okay for TPC, but it is totally wrong for TMC.
Besides, that wasn't the only time you gave evidence of your 'lack of knowledge' (your language) on TMC. Let's have a look at this comment:
http://www.diyaudio.com/forums/soli...erview-negative-feedback-302.html#post2354445
In particular the phrase: select the capacitor values as you would for double pole compensation. So, selecting equal caps (for TMC) wasn't just an incidental mistake, rather the result of your misconception about TMC.
That's no sim, that are schematics without any useful comment. No graphs, no distortion figures, no phase margins, step response, etc.
The only thing you have proved with that post is that you don't understand a iota of TMC. Maybe C2=C4=200pF is okay for TPC, but it is totally wrong for TMC.
Besides, that wasn't the only time you gave evidence of your 'lack of knowledge' (your language) on TMC. Let's have a look at this comment:
http://www.diyaudio.com/forums/soli...erview-negative-feedback-302.html#post2354445
In particular the phrase: select the capacitor values as you would for double pole compensation. So, selecting equal caps (for TMC) wasn't just an incidental mistake, rather the result of your misconception about TMC.