Hi Mike,
If one uses identical values for TPC and TMC networks, it is a true statement that the total loop gain in the minor loop is the same as the major loop gain with TPC, IF you use an ideal output stage with unity gain and no phase lag. But this is not a very interesting case.
Things become more interesting when you allow C2 to be larger than C1 and have an output stage with gain less than unity and some phase lag. Throw in some excess phase from the input stage and VAS, and it can get more interesting.
TPC and TMC are different, and we end up with the best performance when we allow each technique to exploit its own advantages, as long as we are diligent in designing each approach with the same margin against instability.
Cheers,
Bob
Hi Bob
Let's not forget that the same effect can be achieved with an extra resistor instead of an extra capacitor, as I explained here. As a practical example, the compensation network I showed in this post uses that technique.
I feel that's a worthwhile modification. Perhaps you'd care to play with the idea to confirm it works, and maybe mention it as an alternative in the next addition?
Regards Godfrey
p.s. Sorry, don't know who to credit for the original idea.
I think this peak is primarily an artefact of simplistic models.
The model on p 178 uses perfect transconductors for the IPS and VAS and these can produce unlimited gain.
In reality the peak is minimal.
I pointed this out a while back but never had a response.
Dennis Feucht explains pretty well the way the circuit behaves as the components become more "ideal".
Harry Dymond's simplified analytic model also predicted an unrealistic peak.
This contradicted the results of his Spice simulation.
He included a comment about this in his errata after I raised it with him.
Best wishes
David
Last edited:
I think everyone has missed the OP's point.
More a question about practical experience rather than theory.
For instance the inner loop is probably less vulnerable to speaker impedance variations.
Does it need the same safety factor?
Also an inner loop can be conditionally (Nyquist) stable, as in Edmond's SuperTIS. What is appropriate for that?
Edmond himself expressed some uncertainty about that.
So it seems a sensible question and I would be interested in the answer too.
Best wishes
David
Last edited:
Hi Matthias
The transconductance of G2 seems unrealistically low.
I would expect at least 1m, to 10m or even more for ultra-low noise.
Do I misunderstand?
Best wishes
David
ihan said:
Bob has in fact highlighted an example of this with his cap. across the TPC network. Since this is a 2nd order loop, it cannot oscillate (provided there are no other evil phase lags). But it can come very near to it, ie the big peak at mid frequencies.
But in this example, the main loop (and the final amp) is not bothered at all and there is no effect on closed loop performance in the practical experience of several people. A Nyquist view of this is that stuff happening far from (-1,0) is of little or no importance provided it doesn't break into local oscillation.
However, other inner loops, ARE more than 2nd order. eg pure Cherry around VAS, driver & output, MIC, evil TMC etc. .. so CAN oscillate.
Here, the 'local' peaks are also very close to the main loop's (-1,0) so WILL kink Nyquist stability for either good or evil.
So the answer, is "it depends" and its probably wise to make sure these local HF 'peaks' are nice & damped.
Again, a nasty 'closed loop' response for these inner loops will tell you if things are wonky (big peaks at HF) but the converse may not hold.
Last edited:
Hi Bob
Presumably Cherry considered it a concern but not an insuperable one.
The lead capacitor does provide an extra free parameter to optimise the feedback network.
That could be useful if the EMI concern can be overcome.
A back-of-envelope estimate makes me think it is not a real problem.
Do you have any data on this?
Best wishes
David
Perfectly reasonable, if you can derive equations for the locations of the coincident poles and zero within the minor loop of TMC.
But these are likely to be intractable, which is why I recommend you start by designing a stable TPC network and then merely connect the resistor to the output.
lead compensation?
Hi Bob & David,
If you are talking about Cherry's NDFL amp as published in J. Audio Eng. Soc., Vol.30, No.5, 1982, then there ain't no phase lead cap in the FB network. The cap you see is just part of a sort of capacitive voltage divider (rather weird, I would say). His original configuration can be transformed into another equivalent one, which contains just a resistive voltage divider. See: PGP Amplifier fig.1 (C1 & C5) vs fig.2 (R1 & R2).
Cheers,
E.
Hi Bob & David,
If you are talking about Cherry's NDFL amp as published in J. Audio Eng. Soc., Vol.30, No.5, 1982, then there ain't no phase lead cap in the FB network. The cap you see is just part of a sort of capacitive voltage divider (rather weird, I would say). His original configuration can be transformed into another equivalent one, which contains just a resistive voltage divider. See: PGP Amplifier fig.1 (C1 & C5) vs fig.2 (R1 & R2).
Cheers,
E.
Your statement proves only that you are not comfortable analyzing the loop gain in terms of pole-zero distribution. Otherwise, you would of course note that one of the roles of the "capacitive divider" is to introduce a zero at HF - hence, given the topology, the "phase lead" effect.
Hi Mike,
in order to really hightlight relationships between TPC and TMC as you proposed, one has to do it with a conceptional circuit, i.e. with controlled current and voltage sources. Going to a "real" circuit just introduces more complications.
From the conceptional circuit follows: Connecting the TPC/TMC resistor to the OPS output instead to ground adds another transfer function to the overall loop gain around OPS, whereas the "old" contribution, the global loop gain from the TPC case, remains unchanged. The system is linear, thus superposition holds.
Best regards,
Matthias
Hi David,
on the left-hand sides, I better should have separated GNFB network and IPS as on the right-hand sides. There, it was necessary in order to introduce the lead network. On the left hand-sides, increasing IPS transconductance to 2mS and adding a feedback natwork with ratio 1/20 should not change the loop behaviour at all. The values correspond to the example in Bob's book, creating for TMC a ULGF of around 500kHz both in the inner TMC loop and the outer GNFB loop.
Best regards,
Matthias
My point was the following. In the transient simulation, we may see an OPS oscillation starting at some point in time. Often, it will start with small amplitude. If one makes a linear analysis with the operating point the OPS was seeing when breaking into oscillation, then one should be able to detect the reason of instability. Since the linear analysis calculates effective device parameters for the chosen operating point, all voltage- and current-dependent changes of ft and hfe are included into the evaluation.
If one cannot see the reason for instability with linear analysis, then I suspect that not all relevant loops have been taken into account, or that they are not probed in the right way. I assume that this is the case with my simple proposal in the EF analysis thread.
Matthias
It looks like a phase lead capacitor in the ETI November 1982 article that I have, but that's unimportant to my point.
I just meant that Cherry's use of a capacitor across the feedback resistor implied he did not consider it a major problem.
I think the extra free parameter could be useful to shape the feedback, and Mike's question reminded me that I planned to ask Bob about this same topic-
How real is the concern for EMI via this route if the Zobel and inductor are well done? Or specifically, what is the evidence?
Other people's input welcome too.
Best wishes
David
/OT, nevertheless quite important.
Please look here: NOS Nieuws - Livestream - Journaal 24
Please look here: NOS Nieuws - Livestream - Journaal 24
For those who don't understand Dutch:
Willem-Alexander to become Europe's youngest king as Prince Charles watches on - Telegraph
edit: At 14:00 hour (local time) our new king and queen will enter the church.
Willem-Alexander to become Europe's youngest king as Prince Charles watches on - Telegraph
edit: At 14:00 hour (local time) our new king and queen will enter the church.
Last edited:
Hi David,
only can comment from the point of listening experience, if this is allowed
. Additionally, I cannot swear that the Zobel and inductor were optimal.I once did a comparison for a structure where a zero at 400 kHz was necessary in the global loop. There was a possibility to realize the zero somewhere inside of the circuit or in the global feedback network. The result was that introducing the zero in the feedback network changed the amplifier sound towards the harsh side.
Apart from the EMI issue, one should not forget that high-frequency distortion residuals at the OPS output also might be a problem.
Best regards,
Matthias
Last edited:
Why would distortion be a problem?
The phase lead capacitor increases the feedback and if it has any detectable effect on distortion then it should reduce it.
Best wishes
David
While he did not state in any manner that he considered it a 'major' problem, in one of his multiple papers regarding TIM et al, he did not regard the lead cap in the feedback network as a good solution in line with his feedback analysis. Don't have the article title at my fingertips*. But it's out there if anyone wants to dig for it.
Leach was also leary of using a lead cap in feedback, being concerned with
the amp's load possibly having an interaction with the cap that would affect compensation.
*Might have been in his sensitivity analysis set of papers. I have these, but don't have time to dig them out and find it, right now.
Last edited: