Has anyone seen this front-end before?

[Dave Zan]

Originally Posted by Edmond Stuart

Hi Bob,

As you said:

>"TPC encloses all three stages (IPS, VAS, OPS) in a second-order loop, while TMC encloses only the VAS and output stage in the more complex loop. Some will say that this is a disadvantage of TMC, not including the IPS." and:

>"TMC does not load the VAS as much as TPC, since the shunt resistor is bootstrapped over the majority of the frequency range by the output stage."

This is the reason (according to me) why in real amplifiers TMC and TPC yield similar distortion figures.

I think Bob has nicely separated some points that actually relate to two separate problems..
1. A feedback optimisation problem. How can feedback be allocated across stages to minimize distortion, subject to stability constraints. This is essentially a mathematics problem, covered by Bode's theorems, Lurie and Horowitz touch upon this too.
2. An implementation problem. How can the feedback be picked up so as to minimize the effect of the load. This is of course an electronics problem. It is not discussed much in control theory because the load of the sensor is usually trivial. Horowitz mentions that it makes amplifiers a more difficult problem that a similar control system.

I think the discussion has been a bit inconclusive because of a failure to separate these issues.

TMC appears to lead to a superior implementation in the way it minimises the load on sensitive parts of the circuit.
TPC appears to have some theoretical possibility to increase the feedback.
In particular I notice that TPC feedback can look more like Bode's plot of maximized feedback.
So I want to see if it is possible to retain the excellent load characteristics of TMC and actually improve it even further.
Need to understand Edmond's earlier posts on TTMC first. Also reconcile the Harry Dymond TPC paper and Dennis Feucht's TPC notes to produce the correct "Bode fillet" in the feedback.

Don't want to beat a dead "TPC versus TMC" horse but the observation on TPC and Bode maximization is new AFAIK.
"Has anyone seen this feedback observation before?"

Best wishes
David

[Edmond Stuart]

Hi David,

You nailed down precisely why in the past the comparison between TMC vs TPC was subject to much controversy. This was, because we failed to make a clear distinction between a feedback optimization problem and a implementation problem.

Thx.

Cheers,
E.

[R Dijk]

Originally Posted by Edmond Stuart

But my question was: Is it new; has anybody seen it before?

Cheers,
E.

Front End

Originally Posted by Edmond Stuart

Hi Ronny,

Please, show us how to do that without degrading the gain of the LTP-VAS combo (and without using a CMCL, of course).

Cheers,
E.

Mirror symmetry or camera playback ?
Do you mean somthing simple like this, far from optimal. Twisted mirror;

[jcx]

Don't want to beat a dead "TPC versus TMC" horse but the observation on TPC and Bode maximization is new AFAIK.
"Has anyone seen this feedback observation before?"

Dr Richard Mitchell's Selected Research Papers and Presentations

one of the few other places I've seen Bode's maximum Feedback discussed - and at least one paper goes a little sideways - probably was unaware of BJ Lurie's work

[Edmond Stuart]

Originally Posted by Bob Cordell

[..]
It is not difficult to design an IPS-VAS that is far better than the output stage, leaving the output stage to strongly dominate and ultimately limit the distortion performance of the amplifier.
[..]
Cheers,
Bob

Hi Bob,

I fully agree with that. Yet I can't resist to nitpick on one point, if you don't mind.
Strictly speaking, it's the IPS that ultimately limits the performance*, as the non-inverting half of this stage isn't (and can't be) enclosed within the global NFB loop.

edit: * Unless someone invents a zero error subtracter.

@Pete, thanks for the links.

Cheers,
E.

[AndrewT]

Originally Posted by Edmond Stuart

..........as the non-inverting half of this stage isn't (and can't be) enclosed within the global NFB loop...........

Is it wrong to go on from there and suggest that better matching of the input pair would maintain the LTP performance, even though half of it does not benefit from the distortion attenuation of the feedback?

It also seems that using the more sophisticated front end topologies, eg cascoded input and Baxandall pair input would help improve the overall LTP performance, due to that lack of feedback to the non-inverting half.

[Dave Zan]

Originally Posted by jcx

one of the few other places I've seen Bode's maximum Feedback discussed

Some reading from Reading, how appropriate.
I don't think he actually says that TPC can approximate Bode maximum.
But nice work and thank you for the link.

Best wishes
David

[Waly]

Originally Posted by Dave Zan

TMC appears to lead to a superior implementation in the way it minimises the load on sensitive parts of the circuit.

TPC appears to have some theoretical possibility to increase the feedback.

On the dead forum Megajocke (using more of a "brute force" method, he also posted the details somewhere on diyaudio.com) and syn08 (using a canonical star-triangle circuit transformation of the TMC network) independently proved mathematically (and no, not by simulation!) that a TMC is, for all purposes, equivalent to a TPC plus a lead-lag compensation network. This equivalence should be understood in the sense that given a TMC compensated loop gain vs. frequency, then a TPC plus a lead lag compensation network can be calculated to provide mathematically exactly the same loop gain, at all frequencies. Anything else beyond this result is bedtime story.

Otherwise, TPC extra loads the VAS, TMC extra loads the input stage - pick your poison.

[Bob Cordell]

Originally Posted by Waly

On the dead forum Megajocke (using more of a "brute force" method, he also posted the details somewhere on diyaudio.com) and syn08 (using a canonical star-triangle circuit transformation of the TMC network) independently proved mathematically (and no, not by simulation!) that a TMC is, for all purposes, equivalent to a TPC plus a lead-lag compensation network. This equivalence should be understood in the sense that given a TMC compensated loop gain vs. frequency, then a TPC plus a lead lag compensation network can be calculated to provide mathematically exactly the same loop gain, at all frequencies. Anything else beyond this result is bedtime story.

Otherwise, TPC extra loads the VAS, TMC extra loads the input stage - pick your poison.

Hi Waly,

Theoretical proofs are indeed important, and provide insight, but they have their limitations. Whether or not this theory is correct, it must make assumptions about the implementation of the amplifier, and other assumptions to just make the problem tractable.

Simulation of a real amplifier design is closer to reality than theory, but it also has its limitations. Finally, building a real prototype is still a better indicator, but not necessarily the final word in the general case.

Your point about TMC "loading" the input stage probably refers to the fact that TPC provides more gain after the input stage than TMC, since for that aspect of analysis TMC behaves more like straight Miller compensation. Depending on the details of the amplifier design, some designers may be able to do better with TPC. My personal experience, based on many simulations of different amplifier designs, is that my outcome is better with TMC than with TPC when I do the best I can to optimize both, with stability of the two designs held reasonably constant in regard to overall fairness. Your mileage may differ.

If they are the same by adding a lead-lag network to TPC (presumably to flatten the closed loop response), why would anyone go for the extra complexity required to make TPC the same as TMC?

What counts for me is results that are obtainable in the real world with reasonable effort and reliability, not whether two approaches are theoretically the same.

I should also add that I am not a long-time detractor of TPC. I first used TPC in the late 1970's at Bell Labs in the design of operational amplifier ICs that were intended to allow the building of better active filters, such as for Biquad filters (BTW, Jimmy Tow worked for me). TPC was an important tool in making adequate operational amplifiers that had the limitation of slow lateral PNP transistors at the time.

Cheers,
Bob

[Waly]

Originally Posted by Bob Cordell

Your point about TMC "loading" the input stage probably refers to the fact that TPC provides more gain after the input stage than TMC, since for that aspect of analysis TMC behaves more like straight Miller compensation.

If they are the same by adding a lead-lag network to TPC (presumably to flatten the closed loop response), why would anyone go for the extra complexity required to make TPC the same as TMC?

Yes, same thing.

The discussion went over time well beyond some implementation details. It was stated that TMC is "vastly superior", that it is "a first order system" so that's why it offers flat phase, that "it shines" over TPC, etc... etc... etc... But if we agree the only difference (other than personal taste) resides in the cost of a cap and a resistor, then I may concede TMC is superior.

It's probably to late for me to think straight, but your statement "simulation of a real amplifier design is closer to reality than theory" doesn't make any sense. Last time I've checked, simulation is theory, so it can't go against theory .