That second TPC should be TMC.
Waly said:
I think that is more or less what Bob has just said. You are in violent accord.
So what are the appropriate apples when the loops are structured somewhat differently?
Best wishes
David.
Hi Waly,
I thought I made it clear that the overall stability of a TMC amplifier was not simply determined by the gain and phase margin as seen with the global feedback loop, but rather must include all of the net feedback seen by the output stage around it. I'm sorry if I did not make this clear.
I do believe that TMC is overall superior to TPC in most amplifier arrangements that I have encountered, but that is just my experience. We always must allow for exceptions that are unique to a particular amplifier that may be designed differently from the ones one of us has studied, or with the particular component values used. I agree that we must all be cautious of overly general and broad statements made without caveats or qualifications.
Cheers,
Bob
Hi David,
>I don't see what it contributes.
I included this square wave test, because, as mentioned in the original post, it is common practice to torture amps with a square wave
Cheers,
E.
Maybe, or maybe not. I am constantly bothered by the way people interpret the first order aspect of the TMC global loop gain phase. "Look, I got so many extra gain dBs and the phase margin is still 80 degrees!". This is flat wrong, and I don't think the hardcore TMC promoters did enough to explain that there are no free lunches.
Implicitly, or even explicitly like Mr. Stuart here
maintaining the illusion that TMC is a) a first order system and b) as a first order system it somehow creates some extra loop gain, compared to Miller, to improve the amplifier performance, is flat wrong.
TMC is just another second order compensation method, with pros and cons (and some were just quoted by Mr. Stuart, regarding the VAS vs. IPS loading), and should be treated accordingly. There is no magic involved and no secret sauce.
Sometimes I think its Mike’s mission in life to prove Edmond and me wrong in regard to TMC vs TPC.
No.
I have never urgued that TPC and "TMC" are the same. And, yes, C1, C2 and R1 can be the same for both techniques; they have to be for the same forward path unity gain frequency. Yes, TPC subjects the whole amplifier to enhanced loop gain at HF, while "TMC" does the same for only the TIS and the output stage.
True.
Not true. The input stage poles are typically well beyond the unity gain frequency of any competently compensated amplifier.
Last edited:
Hi kgrlee,
You are exactly right here. I often believe that one of the things that makes amplifiers sound different is that they misbehave differently under different conditions. The burst of parastic oscillations you mention are especially insidious. I have seen cases in my own lab where such brief burts at high frequencies have caused the amplifier to exhibit higher, but still reasonable, THD - it can be subtle in a lab measurement environment. Maybe the difference between 0.02% and 0.002%, for example.
Your point about unconditional stability under virtually and load goes in part to the matter of whether or not an output inductor is used. In my experience, it is extremely difficult to satisfy that criteria without an output coil. I advocate the use of an output coil, albeit one as small as possible to do the job. Some high-end designers with good intentions eschew the use of an output coild because they think it degrades the sound quuality; in some cases their amplifiers may sound "different" because of the situations you mention.
Even 1uH in a well-designed amplifier can do the job, and I believe even John Curl agreed at one point that 1uH was unlikely to degrade sound quality. It is notable that a 1uH inductor has a reactance of about 0.13 ohm at 20kHz, so it will limit the HF damping factor to about 62. This is more than adequate for all but the wierdest loudspeakers.
As discussed in my book, I also advocate the use of distributed Zobel networks on the output stage side of the L-R network. These Zobels should be placed close to the output transistors and must be non-inductive. This becomes more important as faster and/or more output transistors are used. Multiple smaller Zobels can also be more convenient from a physical design point of view, since the power dissipation can be distributed across several 1 or 2-watt film resistors. Similarly, the capacitors in each Zobel network can be smaller and inevitably with less self-inductance (e.g., several 0.01uF polypropylene caps).
Cheers,
Bob
This is really not necessary: only a single properly calculated LCR network is required at the amplifier's output.
These oscillations are not necessarily symmetric nor limited to a single condition.
Would like to see if the next edition could possibly expand on local oscillations and remedies.
Thanks
-Antonio
Last edited:
An EF3 using fast output devices is easy to provoke into oscillation. You need the Zobel (I have not tried Bob's distributed approach but have no reason doubt it) and you need the base and collector stoppers. Don't forget also, if you're using cascodes, to provide some base stoppers as well. Ferrite beads work wonders.
I see things with my 200 MHz bandwidth analog scope and 1Gs/s Rigol that were never visible on my 10 MHz Hitachi scope, so I suspect a lot of designs have issues that are just not picked up without the right gear.
I see things with my 200 MHz bandwidth analog scope and 1Gs/s Rigol that were never visible on my 10 MHz Hitachi scope, so I suspect a lot of designs have issues that are just not picked up without the right gear.
Last edited:
It may not be really necessary on many amplifiers, but it can only help, especially depending on the layout and speed of the output transistors. You are always free to not use a good idea.
Cheers,
Bob
Hi Waly,
All that has been said, as far as I know, is that TMC provides a first-order open-loop gain as seen by (only) the global feedback loop. I don't think anyone has said it is is overall a first-order system.
There is no free lunch, but some are tastier than others, at least to some
Cheers,
Bob
That was 3 years back and and more is understood since then.
I took Bob's summary as a statement of the current consensus that TMC has a second order loop, as has TPC.
Now let's try to understand a bit more.
Best wishes
David
Cross posted this but will let it stand.
Last edited:
In fairness to TPC, let me say one thing here, just thinking out loud. I could be wrong.
With TPC, I think it is clearer and more easy to see what you get in terms of overall stability. The price paid in stability is easily seen in the conventional gain and phase margin analysis of the major loop.
With TMC, one may be able to stumble into more trouble because the similar price to be paid in stability may not be explicitly seen in the conventional measurement of phase and gain margin of the major loop.
Cheers,
Bob
With TPC, I think it is clearer and more easy to see what you get in terms of overall stability. The price paid in stability is easily seen in the conventional gain and phase margin analysis of the major loop.
With TMC, one may be able to stumble into more trouble because the similar price to be paid in stability may not be explicitly seen in the conventional measurement of phase and gain margin of the major loop.
Cheers,
Bob
Hi Mike,
I think you are over-simplifying here. There are many different combinations of C1, C2 and R1 that can yield the same forward path unity gain frequency for either TPC or TMC. Some combinations that yield the same forward path unity gain frequency may be more optimal for one technique versus the other.
Secondly, sole focus on the forward path unity gain frequency is not necessarily the proper metric for evaluating overall stability. It is phase and gain margin of the path being considered. For example, one of these techniques may be able to achieve the same degree of stability but with a higher forward path unity gain frequency because it incurs less excess phase.
Cheers,
Bob
I asked about the distributed Zobel in some other thread but perhaps you missed it.
In an OPS with multiple EF transistors, it makes sense to me to put the distributed Zobel resistors on the emitter side of the emitter resistors. Does that seem reasonable to you?
Best wishes
David
I don't think so, have you read the link and quote I posted?
Anyways, other than perhaps an interesting observation, what would be the relevance of "a first-order open-loop gain as seen by (only) the global feedback loop" in the case of TMC?
Here's the one and only I was able to figure out: no overshoot in the step response. While I don't think this is anywhere a performance metric for audio, it may create a false sense of security when it comes to global stability (since traditionally overshoots are associated (not necessary always correctly) with lower stability margins).
One issue is excess phase from any RHP zeroes.
In simple Miller comp this is too small to be a problem.
Presumably similar for TMC and TPC but worth consideration.
Best wishes
David
Actually if you have sufficient stability margins with TPC, then all you have to do is connect R1 to the output and you'll have the same stability margins within the local loop about the the TIS and the output stage with "TMC", provided you don't change the component values when moving from TPC to "TMC".
Last edited:
The RHP zero is only a problem in amplifiers with a MOSFET TIS; this is due to their moderate transconductance compared to BJTs.