Well stated!
Many, many amplifier designs exhibit the closed loop gain peak above ULGF that I was referring to. Even though that peak usually gets nowhere near up to 0dB closed-loop gain, it is a source for worry. For example, an amplifier with 6dB gain margin will virtually by definition have a 6dB closed-loop gain enhancement through the sub-unity positive feedback that will exist at the ILGF or whatever we choose to call it.
Cheers,
Bob
In Opamp theory this is called "unstable region"
Stable region 6dB/8°
Unstable region 12dB/8°
In a graph Bode the loop Gain (Avol)
Fo - First pole
Fcl - Closed loop Bandwidth
Gcl - Closed loop Gain
Fu - Unity Gain Frequency
Fcl may also be called F1 and Frequency unstable of F2 (looking at a graph).
I have not found a name for the frequency unstable region, but I think that acronym can be quite simple and objective.
Stable region 6dB/8°
Unstable region 12dB/8°
In a graph Bode the loop Gain (Avol)
Fo - First pole
Fcl - Closed loop Bandwidth
Gcl - Closed loop Gain
Fu - Unity Gain Frequency
Fcl may also be called F1 and Frequency unstable of F2 (looking at a graph).
I have not found a name for the frequency unstable region, but I think that acronym can be quite simple and objective.
Hi Edmond, thanks for chiming in here. I like your suggestion of referring the the frequency as the "Gain Margin Frequency", GMF. This seems pretty explicit and not ambiguous for those who know what gain margin is.
Unless we hear strong objection, I'll go forward with using the term GNF.
Cheers,
Bob
While I hesitate to say this is always the case, my experience has been that such 'above bandwidth peaking' is usually cos there is an inner loop that may have wonky stability. TMC and 'pure Cherry' are prone to this but even plain Miller can show this if the load on the VAS is wonky enough to affect its stability. The 'ULGF' of the inner loop is usually a lot higher than ULGF of the main loop.
I think you need to look at each internal loop if you see this. I'm also going to claim that if the Closed Loop response looks OK, all the inner loops will probably be OK.
I agree, If there is peaking in one of the local loops, it can cause peaking in the overall open loop gain, and this will inevitably result in peaking in the closed loop gain at frequencies above the ULGF.
However, and importantly, even if you have a perfectly done front end with no instability in any local loops it might have, you can also get peaking in the closed loop response above the ULGF if the amplifier has smallish gain margin. This can happen, for example, as a result of poles adding additional lagging phase in the output stage. For example, a 6dB gain margin will result in such a peak even without any local loop instability or peaking.
So there are two possible causes of closed loop peaking above the ULGF.
For purposes of this discussion, what I mean by closed loop peaking as any non-monotonic fall-off in closed loop gain above ULGF.
I like your assertion that if the closed loop gain looks OK, then the internal loops are OK. I have been leaning toward this very same assertion. If true, it is a profoundly useful observation because it is so easy to check in the lab.
Can we agree that "OK" for closed loop gain means a monotonic fall-off of closed loop gain above ULGF?
If so, and we follow this rule of thumb, there may be some cases that result in amplifiers with overly-conservative compensation as viewed by some, but it is wise to err on the conservative side, especially given the degradations that may occur in real-world situations. For example, we may have lots of amplifiers with gain margins of 10dB or more and phase margins of 70 degrees or more, but so be it.
Cheers,
Bob
Hi Bob,
I hope that GNF was a typo we wouldnt want it being confused with global negative feedback.
Gratuitously adopting acronyms in the silly fashion of the U.S. Military is uncalled for.
They certainly do have their pros and cons. Acronyms are notoriously context-dependent, especially TLAs (Three Letter Acronyms). The same acronym has totally different meanings in different industries or areas of endeavor.
I suspect that the U.S Military has nothing on the medical/healthcare industry.
Cheers,
Bob
GMF = Gain Margin Frequency.
Surely GMF is a Frequency and not a gain
Or, it is a combination of the gain margin and the frequency at which that margin occurs?
Last edited:
I was under the impression that assertion was widely viewed. I think many opamp manufacturers like Texas instruments have this view. One can find evidence of the fact in a couple of opamp datasheets. For checking stabilty network analysers are used to generate output vs frequency curves. Any peaking above 2 db usually means phase margin is around or less than 45 degrees and on the verge of instability. Any capacitive loading will cause oscilation.
I would like to draw attention to what manso posted, which in my experience is extremely important, overlooked and underappreciated. In conventional Bode plots it's easy not to see the tiny wandering of the gain curve that is the result of some sort of internal peaking. Often I see superficially sharp BW corners, which really aren't normal gain curves but an underdamped peak at the BW corner. This is why I plot group delay in LTSpice; this magnifies any such problems. Also helpful is to plot the derivative of step response, D(V(Vout)), which will often make you cry.
Not sure I've seen the assertion before in the way it is meant here.
It is very important to understand that we are not talking about the usual peaking around ULGF brought about by phase margin, like your example of peaking due to phase margin of 45 degrees. This is quite well-understood.
Here we are talking about relative peaking at higher frequencies, above the ULGF, and more in the vicinity of the frequency where gain margin is evaluated (GMF). For an amplifier with a ULGF of 1MHz, the peaking we are talking about here might occur at, say, 5-10MHz or higher. This is not peaking with respect to the nominal closed loop gain of the amplifier. These peaks may not come up anywhere near that reference. I have casually described these peaks as the closed loop gain above ULGF as not falling monotonically.
In some cases, local loop instability in the amplifier may occur at rather high frequencies, perhaps in the 10's of MHz - see Mike's post a few pages back in regard to stability of the local Miller compensation loop around a Darlington cascode VAS, wherein the signal goes through three transistors in traversing that loop.
Cheers,
Bob
Hi Bob and kgrlee,
following e.g. the discussion in the "pure Cherry" thread, I would concur with this statement. IMHO, it is really important to worry about any peaks or "sharp knees" in the global loop gain, even at frequencies well above its ULGF.
Furthermore, I would be very very careful if these artefacts already show up in the simulation. If one can confirm in the lab, that these things appear with the real layout and power supply in a reproducible way and that they are robust against load changes etc., then be it.
[I repeat myself] Independently, a careful listening evaluation will be necessary. Everybody who already has done this with good sources and speakers knows, how sensible the presentation of plosives, overall timbre and sound stage may react on things like stability margins (also of minor loops whichs ringing behaviour can influence the performance of the whole circuit under real-world conditions).
Cheers,
Matthias
Last edited:
@ Bob Cordell et al
You might be interested in some sims i did
http://www.diyaudio.com/forums/solid-state/235364-compensate-amplifiers-not-how-best.html
Yes, it's a buffer, but check out the VHF traces
You might be interested in some sims i did
http://www.diyaudio.com/forums/solid-state/235364-compensate-amplifiers-not-how-best.html
Yes, it's a buffer, but check out the VHF traces
These are all good suggestions. It is also the case that just looking at the closed loop phase will show such behavior more prominently than the amplitude curve.
Cheers,
Bob
Yes, if some funny business shows up in simulation, there is a very good chance it will be worse in the real world, due to extra parasitic components in the real amplifier, like inductances.
I have often wondered if a lot of the audible differences in amplifiers are a result of burst parasitic oscillations on perhaps a part of a cycle. Most of us have seen this in the lab when we have a border-line stable amplifier.
Think of all of the well-intentioned hi-end designers who don't even use SPICE and may not be excruciatingly thorough about checking for traces of instability in their amplifiers under a myriad of loads. These high-end amplifiers could sound different from one another, especially if you throw in some boutique cables and maybe some nasty high-end speaker loads.
Cheers,
Bob