Bob Cordell's Power amplifier book

Administrator
Joined 2004
Paid Member
Hi Bob,
I also finally received a copy of your book. I'm reading very slowly in order to catch anything that may slip by. It's excellent in that it fills in the gaps in my education pretty well so far. A good read by all accounts. I also have Doug's Small Signal book that I'll read afterwards. Both books look to be very complete, but different for sure.

Having these books included in the electronics course curriculum would be a positive step. Applied knowledge helps to strip away the idea of "ideal devices" that don't exist. Students need that dose of gray area that isn't covered well in my opinion.

I hope you succeed there Bob.

-Chris
 
spice magic

Here's for example what happens with the earlier posted TMC sim if the load is changed to 100nF in series with 4 ohms (looks like a typical Zobel network!) and the bias of the output stage reduced by changing the 495 ohm resistor to 2000 ohms. The input signal is 5 kHz 10 mV pk sinewave and the amplifier oscillates at about 200 kHz.

It doesn't look dangerous neither for the amplifier nor the speaker though with such a low amplitude.

Hi Joakim.

1. How large is Iq of the OPS trannies?
2. A load of 100nF + 4R looks NOT like a decent Zobel network.
3. TPC with the same compensation components (as TMC) oscillates just as much, at least with the test circuit I'm using (dual EF OPS). With different component values TPC oscillates even more.

So, what do you want to prove with your simulation?

Cheers,
E,
 
But the TMC network is a feedback path, just that it bypasses the LTP.

Why this worry about increasing the gain crossover frequency from 1 MHz to 1.7 MHz in the loop around the input stage when you had it up much higher (wasn't it tens of MHz?) in the "input inclusive compensation" scheme? At the same time why are there no worries about the increase of gain crossover frequency of the loop around the output stage from 1 MHz in the original single pole compensated amplifier to 1.7 MHz in the TMC version?

Isn't it the output stage with its load that has the big unknowns in its transfer function at single MHz frequencies and needs the phase margin and is the reason we need to be careful with our crossover frequency?

Both the TMC and TPC+lead versions have the phase dip at ~20 kHz in the loop around the output stage. Why is this okay while the LTP must not see the same phase dip? The output stage loop crossover frequency will be in that range when the stage is starting up, when bias and thus transconductance is low. If the output impedance of the output stage is high due to low bias and the load is significantly capacitive at this frequency there may be trouble both for the TPC and TMC versions.

Wouldn't a "fair" comparison be one where the forward gain and feedback gain of the front end (including LTP, VAS and compensation network) are the same, giving the same loop gain and crossover frequency around the output stage? This stage is the one that is slow, has high distortion, large parasitics and moreover large unknowns in initial parameters, significant output current depedency and load dependency in its transfer function.

Hi megajocke,

These are good "what's good for the goose must be good for the gander" questions, and I'll scratch my head on it some.

However, what I meant by the lead network being needed for the TPC to do as well as the TMC was the extra components needed in the global feedback path. TPC needs more components to work as well as TMC - that's all I was saying.

The local loop around the input stage in my Miller Input Compensation (MIC) that I used in my MOSFET power amplifier did indeed have a high gain crossover frequency above 10 MHz. This is no problem because it does not enclose the output stage. The input LTP is quite fast and not a big speed problem.

Your question about gain crossover frequencies around the output stage in the TMC circuit is the one that gives pause. The conventional global feedback loop in the TMC design has a gain crossover frequency of about 1 MHz, just like the conventional Miller compensation. However, the higher local loop gain crossover frequency around the output stage can be up at 1.7 MHz. I think that is the apparent contradiction you are pointing out. That's the one I want to think about some more. Indeed, it was earlier suggested that the 1 MHz global gain crossover frequency I chose was a bit conservative, and I agreed that I just set it at that.

Here is a speculative clue or observation. Had we merely taken the conventional Miller compensation to a gain crossover of 1.7 MHz, at a 6 dB/octave slope, it seems we would have reduced THD by maybe a factor of 1.7. However, the TMC reduction factor was much larger than 1.7. Some of this may have to do with the advantages of nested feedback loops. I don't know the answer to this apparent contradiction, but I'll give it some thought.

To my mind, the ultimate fair comparison is how well you can do in overall distortion with each compensation technique with the same degree of stability margin. This stability margin must include margin against local loops becoming unstable as well.

Cheers,
Bob
 
Thank you for making this observation, one that Bob and Edmond pretend not to see.

Ounce again you're probably the only other person in this discussion who knows what he's talking about. THANK YOU!

Mike,

Can't you take the hint that insulting other people on the thread is not appreciated and does not help make your point? You are a very smart guy, and you stir the pot well, making all of us think harder, but your smug attitude and not-so-subtle personal attacks on the other participants are really getting old.

Bob
 
Administrator
Joined 2004
Paid Member
:cop:
Hi Mike,
Your posting style has been noticed, and it isn't helpful. Your point of view is valid as your point of view. Continually repeating it isn't helpful, and attacking other members who don't accept your view will probably bring you grief.

Please allow normal discussions to continue and make your points in the professional manner I know you can present.
:cop:

-Chris
 
Hi Joakim.

1. How large is Iq of the OPS trannies?
2. A load of 100nF + 4R looks NOT like a decent Zobel network.
3. TPC with the same compensation components (as TMC) oscillates just as much, at least with the test circuit I'm using (dual EF OPS). With different component values TPC oscillates even more.

So, what do you want to prove with your simulation?

Cheers,
E,

Hi!

1. There is none, which is the kind of startup scenario you mentioned earlier - but for the OPS instead of the LTP.
2. Maybe not a decent one but still not uncommon... A 4 ohm resistive load gave me oscillation too.

Number 3 is exactly my point: the TMC circuit will oscillate if the output stage is debiased, just at the TPC does. I thought I would be able to get the TPC version to do the same by decreasing the LTP tail current but it caused the whole amplifier to debias and full positive rail output before the crossover frequency around the OPS reached dangerously low.

I don't see this instability as a big problem though. I'd still use TMC (or TPC+lead) compensation.
 
Hi Bob,

Probably, my post 1229 have lead to some confusion, sorry my fault. I misinterpreted/overlooked the punctuation (and erroneously read: "it does not require a lead circuit in the feedback path to make it so, this may be a subjective matter.")
Nevertheless, I would ask you to have a look at it (again) and ask your opinion about the increase in the unity loop gain frequency seen at the input stage due to lead compensation in the FB path and the consequences of the additional compensation with regard to a fair apples to apples comparison between TMC and TPC+lead compensation.

Cheers,
E.
 
Hi megajocke,

These are good "what's good for the goose must be good for the gander" questions, and I'll scratch my head on it some.

However, what I meant by the lead network being needed for the TPC to do as well as the TMC was the extra components needed in the global feedback path. TPC needs more components to work as well as TMC - that's all I was saying.

The local loop around the input stage in my Miller Input Compensation (MIC) that I used in my MOSFET power amplifier did indeed have a high gain crossover frequency above 10 MHz. This is no problem because it does not enclose the output stage. The input LTP is quite fast and not a big speed problem.

Your question about gain crossover frequencies around the output stage in the TMC circuit is the one that gives pause. The conventional global feedback loop in the TMC design has a gain crossover frequency of about 1 MHz, just like the conventional Miller compensation. However, the higher local loop gain crossover frequency around the output stage can be up at 1.7 MHz. I think that is the apparent contradiction you are pointing out. That's the one I want to think about some more. Indeed, it was earlier suggested that the 1 MHz global gain crossover frequency I chose was a bit conservative, and I agreed that I just set it at that.

Hi,

Well, TMC sure is a neat trick with a lower component count than TPC for the same amount of distortion reduction, frequency response and stability margin in the output stage loop.

That the input stage is so fast is why I got the idea to replace the feedback going directly into the VAS with an equivalent on the input side of the LTP. These two circuits then give the same loop gain around the output stage and hence the same distortion reduction of the output stage distortion.

To me it feels kind of arbitrary to use the loop gain around the wide-bandwith input stage as a metric for stability when it's the output stage that tends to cause problems. The LTP is just one of two parallell feedback paths going into the VAS, the other being the TMC compensation network itself. It might not be so clear when the circuit is drawn the usual way but when drawn like in my hand-drawn figure (2) it is more apparent.

Here is a speculative clue or observation. Had we merely taken the conventional Miller compensation to a gain crossover of 1.7 MHz, at a 6 dB/octave slope, it seems we would have reduced THD by maybe a factor of 1.7. However, the TMC reduction factor was much larger than 1.7. Some of this may have to do with the advantages of nested feedback loops. I don't know the answer to this apparent contradiction, but I'll give it some thought.

There isn't really any contradiction. The loop gain around the output stage has, for both the TMC and TPC configuration, a double pole at a relatively low frequency where the VAS runs out of gain giving an initial loop gain peak and a subsequent 12 dB / octave rolloff and a high frequency zero which brings back the slope to 6 dB / octave before the crossover frequency is reached.

This zero is given by tau2 in my calculations and was at 414 kHz for the values in you TMC simulation. The TPC network in the simulation you posted on the other hand caused the zero to fall at tau1, 710 kHz, closer to the crossover frequency which is also lower at 1.2 MHz giving a poor phase margin.

The forward gain of the front end is too of double-pole single-zero type but with the zero at tau1 for both configurations. This means that for the TMC version the input-to-output and feedback-to-output responses have the HF zeros at different frequencies while for the TPC version without lead compensation both are at the same frequency.

What the lead compensation I designed for the TPC case does is that it has a pole at tau1 which cancels the zero in the transimpedance of the VAS which is at tau1 and a zero at tau2 which is then the sole zero in the feedback response.

In the TMC compensation network there are three component values you can choose. These together with the two other parameters, input stage transconductance (gm) and feedback divider ratio (A), determines:
1. the forward path zero at tau1 = (C1+C2) * R
2. the feedback path zero at tau2 = tau1 + A * C1/gm
3. the gain crossover frequency
4. the closed loop gain = A
5. a scaling factor for the impedances, determining the magnitude of the current coming out of the LTP.

In the TPC+lead version there are more degrees of freedom however. The loop gain can be made to have an extra pole-zero pair if you do not need to cancel the tau1 VAS transimpedance zero with the lead network pole. By distributing these poles and zeros over a large frequency range you could get an intermediate rollof slope for example and get rid of the phase dip around the output stage if you wanted. Distortion reduction would be lower though.

Cheers,
Joakim
 
Hi,

Well, TMC sure is a neat trick with a lower component count than TPC for the same amount of distortion reduction, frequency response and stability margin in the output stage loop.

That the input stage is so fast is why I got the idea to replace the feedback going directly into the VAS with an equivalent on the input side of the LTP. These two circuits then give the same loop gain around the output stage and hence the same distortion reduction of the output stage distortion.

To me it feels kind of arbitrary to use the loop gain around the wide-bandwith input stage as a metric for stability when it's the output stage that tends to cause problems. The LTP is just one of two parallell feedback paths going into the VAS, the other being the TMC compensation network itself. It might not be so clear when the circuit is drawn the usual way but when drawn like in my hand-drawn figure (2) it is more apparent.



There isn't really any contradiction. The loop gain around the output stage has, for both the TMC and TPC configuration, a double pole at a relatively low frequency where the VAS runs out of gain giving an initial loop gain peak and a subsequent 12 dB / octave rolloff and a high frequency zero which brings back the slope to 6 dB / octave before the crossover frequency is reached.

This zero is given by tau2 in my calculations and was at 414 kHz for the values in you TMC simulation. The TPC network in the simulation you posted on the other hand caused the zero to fall at tau1, 710 kHz, closer to the crossover frequency which is also lower at 1.2 MHz giving a poor phase margin.

The forward gain of the front end is too of double-pole single-zero type but with the zero at tau1 for both configurations. This means that for the TMC version the input-to-output and feedback-to-output responses have the HF zeros at different frequencies while for the TPC version without lead compensation both are at the same frequency.

What the lead compensation I designed for the TPC case does is that it has a pole at tau1 which cancels the zero in the transimpedance of the VAS which is at tau1 and a zero at tau2 which is then the sole zero in the feedback response.

In the TMC compensation network there are three component values you can choose. These together with the two other parameters, input stage transconductance (gm) and feedback divider ratio (A), determines:
1. the forward path zero at tau1 = (C1+C2) * R
2. the feedback path zero at tau2 = tau1 + A * C1/gm
3. the gain crossover frequency
4. the closed loop gain = A
5. a scaling factor for the impedances, determining the magnitude of the current coming out of the LTP.

In the TPC+lead version there are more degrees of freedom however. The loop gain can be made to have an extra pole-zero pair if you do not need to cancel the tau1 VAS transimpedance zero with the lead network pole. By distributing these poles and zeros over a large frequency range you could get an intermediate rollof slope for example and get rid of the phase dip around the output stage if you wanted. Distortion reduction would be lower though.

Cheers,
Joakim

Hi Joakim,

I'm still thinking about whether there is a free lunch (or cheap lunch) in TMC, but the points you are making here are very good ones. It also keeps alive the question of what constitutes a fair comparison of all of the techniques.

With respect to degrees of freedom (which when more are allowed make comparisons more difficult with more permutations and combinations), we should bear in mind that if a lead compensation R-C is allowed in the global feedback path for TPC, it is only logical to allow it in conventional Miller and TMC compensation schemes if they are to be fairly compared. This is one reason I like to compare all three techniques without resort to added components for lead compensation. This then begs the question of whether the addition of lead compensation benefits one of the compensation techniques more than another.

Thank you for your thoughtful and level-headed analysis.

Cheers,
Bob
 
the "problems" lie in the time domain rather than in the frequency domain and both aspects have to be observed simultaneously

That's not possible even in principle, even before engineering issues are taken into account. Rather like the uncertainty principle in QM, only applied to signal processing. As the time interval gets smaller it means frequency becomes increasingly indeterminate.
 
I'm still thinking about whether there is a free lunch (or cheap lunch) in TMC, but the points you are making here are very good ones.

Good question, Bob...
TMC has indeed a serious drawback if the amp s design
doesn t take care of its peculiarities.
As i already pointed it to Edmond, there s benefit to use
TMC in conjunction of TPC, that is, increasing the open loop
gain of the TMC compensated amp.
This will of course have an influence over the impulse response,
i.e, creating the usual peaking , although of lower amplitude
than the single TPC compensation..
More to come as soon as i ve checked the validity of the
relevant simulations..

cheers,

w
 
lead compensation

Hi Joakim,

Referring to your post #1252, I have a few (dumb?) questions:
How did you calculate tau1 and tau2 for the TPC version? The same way as for the TMC version? Also, how do you define tau1 and tau2 for the lead compensation. Maybe the answers are obvious (and I have missed something), but I want to make absolutely sure that during the further evaluation of TPC and TMC, we are not going to talk at cross-purposes.

Cheers,
Edmond
 
The loop gain around the output stage has, for both the TMC and TPC configuration, a double pole at a relatively low frequency where the VAS runs out of gain giving an initial loop gain peak and a subsequent 12 dB / octave rolloff and a high frequency zero which brings back the slope to 6 dB / octave before the crossover frequency is reached.

THANK YOU! Precisely what i have been saying all along and been told by some it didn't constitute "hard evidence".
 
it wasn't exactly what I had in mind, but that´s entirely valid for the Fourier transform, where Heisenberg's uncertainty principle applies - I agree with you 100%, but you would surly have a hard time convincing others.

I think on reflection I read you over-literally, perhaps your meaning was closer to 'let's get a balance between the two domains, not focus over-much on the frequency domain'. In that I'm with you - I've noticed a tendency (not so much in this thread, more so in the digital ones) for high-res FFTs to bedazzle people into thinking they're seeing the whole picture when they're merely appreciating one side of the coin...