Bob,
To have an apples and apples comparison of "TMC" versus TPC you MUST have the same unity loop gain frequency and loop gain at all frequencies of interest about the second stage and the output stage.
This means the minor loop gain with "TMC" and the major loop gain with TPC must have the same locations for the coincident poles and the zero restoring a single pole roll-off.
This is common sense; these criteria are, self evidently, not arbitrary at all and can ONLY be met by using the same component values when comparing "TMC" and TPC.
Now, can you DEMONSTRATE that these criteria can be met using component values that are different with "TMC" viz a viz TPC?
Last edited:
I use the lead network with TPC when I wish to constrain the closed loop frequency response and/or increase stability margins. I make no apologies for it.
I don't think any evidence has been adduced to support your averment that the lead network introduces RFI into the amplifier; this is unlikely to be true if a correctly designed Thiel LCR network is incorporated.
Hi Mike,
are we speaking about something like this? On the right-hand sides (TPC), I have set the input transconductance a bit too high, so that magnitude curves remain slightly different. The second picture is effectively without lead network.
Edit:
Just became aware that I didn't check whether the poles coincide. Very probably, they don't.
BR,
Matze
Last edited:
Hi Mike,
Apples apples only requires the same amount of stability around the minor loop, in terms of phase and gain margin, in a real amplifier. This can be achieved in numerous ways. Moreover, it can be a function of the real amplifier. If you simulate some real amplifiers, and at the same time look at the gain and phase margins of the total loop gain enclosing the output stage, you will get a better understanding of this.
It is also important that you recognize that and apples-apples comparison between TMC and TPC must be done without your lead network.
BTW, Ed Cherry originally raised the concern about feedback lead networks providing a path for EMI back to the input stage.
Cheers,
Bob
Folks,
In going through all of this feedback compensation discussion, I was reminded of what I called "Bridged T" compensation on pages 178-179 of my book. This is where a small bridging capacitor is added across the T of TPC. It tames the very large loop gain peak that can occur in the audio band (often in the vicinity of 1kHz - see Figure 9.5) when TPC is used.
I did not give any references for that refinement of TPC because I was unaware at the time of any previous mention of it. Has anyone here seen this form of compensation discussed prior to publication of my book? If so, I'd appreciate a reference so that I can give due credit.
Cheers,
Bob
In going through all of this feedback compensation discussion, I was reminded of what I called "Bridged T" compensation on pages 178-179 of my book. This is where a small bridging capacitor is added across the T of TPC. It tames the very large loop gain peak that can occur in the audio band (often in the vicinity of 1kHz - see Figure 9.5) when TPC is used.
I did not give any references for that refinement of TPC because I was unaware at the time of any previous mention of it. Has anyone here seen this form of compensation discussed prior to publication of my book? If so, I'd appreciate a reference so that I can give due credit.
Cheers,
Bob
SPICE .ASC uses as least 3 extremely important extra bits of info.
If you use 'good' models like Mr. Cordell's, .ASC models the changes of ft & hfe with current and Ccb with voltage.
IMHO, it does an excellent job of simulating the parasitic oscilations that sometimes appear on part of the cycle with different loads.
I would like to see someone model the instabilities in Stability-analysis-EF-output-stages using a purely 'linear' method like .AC, Bode, Nyquist or Tian. Especially the effect of Cordell's evil[*] 10u across the 'bias spreader'
But if you really want supa dupa performance, just use da default 'npn' and 'pnp' transistors in LTspice for everything. Pity Mouser doesn't stock them.
[*] .. just in case anyone is unsure, dis be a joke
____________
I've used it circa 1990 when I first started playing with TPC & pure Cherry but abandoned it when it seemed to make absolutely no difference to the closed loop performance ... both in simulations and more importantly ... in real life.Bob Cordell said:
You only see it in simulations cos the difficulty of measuring open loop gain.
This might be worth some investigation for your 2nd ed. ie Does it actually do anything for real life closed loop performance?
Last edited:
Hi, Bob,
It will be better, if you put that bridge involving output stage. A little bit less THD for free.
Hi Bob,
I used something similar in my first TMC amp before I bought your book, look here:
http://www.diyaudio.com/forums/solid-state/186981-bootstrapsccs-t-tmc.html
and for TPC here:http://www.diyaudio.com/forums/solid-state/186981-bootstrapsccs-t-tmc-3.html#post2538319
BR Damir
Last edited:
An apples and apples comparison requires that you have the same loop gain about the minor loop with "TMC" as you would have about the whole amplifier with TPC.
In addition you need the same stability margins, which requires the same pole location and the same zero location; there are no "numerous ways" of doing this: you need to have the same component values in both cases.
The lead network isn't necessary to demonstrate this fact. So I think your invoking it is a red herring.
Indeed cherry did not express any concerns about the lead network; on the contrary, he actually recommends them in his NDFL amplifiers.
Last edited:
The forward frequency response peak with TPC is not very large and is barely perceptible in the closed loop frequency response of the amplifier.
Note also that the peak is a function of the current gain of the TIS, which is why the peak is more pronounced with a beta enhanced TIS than with a cascode TIS where it is all but nonexistent if a simple Miller loop is used and not MIC.
At any rate the peak is completely innocuous; it does not in any way affect the performance of the closed loop amplifier.
Last edited:
No, you don't need phase lead to prove that TMC and TPC are "related", but you need phase lead to prove that TMC and TPC are "equivalent".
You can't exactly reproduce the TMC "performance" without the phase lead (and lag, in fact). Where "performance" stands for the four criteria you listed.
As usual, this is a semantics issue.
No, without a lead-lag network TMC and TPC are not exactly equivalent. For the same stability margins, TMC is (probably in most cases, but not necessary always) marginally better. But certainly the TPC overshoot (as much as that can be a problem, which I don't believe) can't be avoided.
Which means that comparing TMC with TPC without lead-lag network is not apple to apple.
Waly you're completely wrong. You don't need phase lead compensation to demonstrate that "TMC" is simply TPC localised to the second stage and the output stage.
Use the same component values for "TMC" and TPC and you'll find that total loop gain in the minor loop is virtually the same as the major loop gain with TPC.
This is all very elementary: I simply cannot parse Bob's arguments on the issue.
Use the same component values for "TMC" and TPC and you'll find that total loop gain in the minor loop is virtually the same as the major loop gain with TPC.
This is all very elementary: I simply cannot parse Bob's arguments on the issue.
The lead network is part of the Cherry's NDFL, so "recommending" is not an option.
Michael, I'm sorry, but you are completely wrong. TMC is not equivalent to TPC without the lead-lag network.
I'm suspecting you have a distorted and fuzzy definition of "TMC is simply TPC localised to the second stage and the output stage". Conceptually, you could be right (after all, TMC and TPC are both methods of 2nd order compensation) but the devil is in the details, so please explain what exactly "TMC is simply TPC localised to the second stage and the output stage" means. Using numbers is fine.
I think that what he s saying is that without the additional lead compensation
a TPC compensated amp using the same network will not have the same
impulse response as with TMC , hence the global transfer functions wont be identical.
Mike,
Can these two points be taken together, so that it would be equally valid to design an optimal TMC network, then move the resistor from output to ground to have an optimal TPC network?
SPICE .ASC uses as least 3 extremely important extra bits of info.
If you use 'good' models like Mr. Cordell's, .ASC models the changes of ft & hfe with current and Ccb with voltage.
IMHO, it does an excellent job of simulating the parasitic oscilations that sometimes appear on part of the cycle with different loads.
I would like to see someone model the instabilities in Stability-analysis-EF-output-stages using a purely 'linear' method like .AC, Bode, Nyquist or Tian. Especially the effect of Cordell's evil[*] 10u across the 'bias spreader'
But if you really want supa dupa performance, just use da default 'npn' and 'pnp' transistors in LTspice for everything. Pity Mouser doesn't stock them.
[*] .. just in case anyone is unsure, dis be a joke
____________
I've used it circa 1990 when I first started playing with TPC & pure Cherry but abandoned it when it seemed to make absolutely no difference to the closed loop performance ... both in simulations and more importantly ... in real life.
You only see it in simulations cos the difficulty of measuring open loop gain.
This might be worth some investigation for your 2nd ed. ie Does it actually do anything for real life closed loop performance?
Thanks for bringing this to my attention. Sometimes there is nothing new under the sun
. I'll mention that it has been used by others before. I'm surprized that Doug Self didn't pick up on this (at least as far as I know).My simulations also showed that there was no discernable change in closed loop response. When I first saw that very large peak in the loop gain with TPC, right in the middle of the audio band, I was alarmed, thinking that it might somehow show up as audible. On the other hand, it would seem odd that a small 3pF capacitance could make a difference.
One caveat is that these evaluations are small-signal. It might be interesting to see if this peak in loop gain could cause a difference in behavior under mild or strong nonlinear conditions.
Cheers,
Bob