Bob Cordell Interview: Negative Feedback

Jump to Latest
#2,681
Re: Re: Re: Re: Re: Re: Down with NDFL

syn08 said:


Given a desired set of loop gain target results, a generic feedback network can be designed, and even worse, the process could be 100% automatized (by synthesis algorithms, and this process is in no way fundamentally different from filter synthesis).

Now, if you think about realisations and down to practical considerations, we may talk about, and here's where Scott comment fits as well. Think in terms of sensitivity rather than network configuration. If you would need a 43.54672pF capacitor and 43.54673 or 43.54671 makes the amp unstable, then the synthesis results are of course useless.

And BTW, NFB won't help for what it doesn't encompass. You and the 1942 CFB inventor, seem to be obsessed with the YAP OPS input stage distortions. Set aside I never said it couldn't be done better (and a new revision is on the board, most likely using Andy's idea) you miss an important point. The YAP OPS compensation main purpose was to a) reduce the MOSFET OPS distortion b) Keep the animal stable and allow as much phase margin. c) Get the ULG as high as possible to avoid interactions with the front end (providing the gain, this is called pole separation). You know the drill from composite opamps... Think of the YAP OPS as the second "opamp" in a composite configuration, and yes, in the next release this opamp will get better!

These three objectives are of course a) not independent and b) not necessary leading to an absolute optimum. For sure integrating the front end with the OPS and extending local loops across both the front end VAS and the OPS will lead to better performance. But what the heck, it was my decision to try and build something that is modular and allowing to experiment with various front end configuration, etc... much as I did with the PGP OPS (accepting three types of output devices and a few compensation configurations). This is DIY and this is the way I am having fun. Other have fun plugging symbols in the simulator and jerking off in front of a bunch of numbers.
Click to expand...




WTF?
Nothing in that rant addresses a single point I made regarding practical (as opposed to theoretical) equivalence of TPC Vs NDFL.
 
#2,682
Re: Re: Re: Down with NDFL

Bob Cordell said:
Hi syno8,

I kind of like TMC. I'm not disputing what you are saying, but the fact that the global feedback loop with TMC is still mainly a straight 6 dB/octave rolloff is appealing to me (as compared to TPC, for example).

Bear in mind that in principle one could combine TMC and TPC.

Cheers,
Bob
Click to expand...

Hi Bob. May I ask why you prefer a straight 6dB/oct roll-off? (As opposed to this roll-off being preceded by steeper roll-offs)
 
#2,683
Re: Re: Down with NDFL

andy_c said:


That seems right to me, with one potential caveat. That extra global loop gain you get with TPC has to come from somewhere of course. With TPC, it comes from reducing the local feedback factor of the VAS, thus reducing its local loop gain. If the VAS itself has a substantial contribution to the overall distortion of the amp, as it may in a high-power design, the improvement with TPC might not be as large as expected. TPC does add some extra loading to the VAS as well.
Click to expand...


Hi Andy. I pretty much agree. I think another important point to mention here is that when a CFB or HEC output stage is used (with an order or magnitude reduction in distortion or more) The prospects for further improving linearity (of the whole amplifier) with TMC are reduced as the OPS distortion has possibly come down to input stage / VAS distortion levels.

Regarding TPC Vs NDFL, with TPC you get higher VAS distortion, but you also get to delete the middle integrator stage required for NDFL, which itself introduces distortion.
In my sims a page or so back TPC returned the same THD as NDFL, with the TPC implemented as Miller comp. around the VAS.
However VAS loading could be further reduced by applying TPC using Bob Cordells "Miller input" compensation (returned to the inverting input).

Cheers,
Glen
 
#2,685
Re: Re: Re: Down with NDFL

G.Kleinschmidt said:
Regarding TPC Vs NDFL, with TPC you get higher VAS distortion, but you also get to delete the middle integrator stage required for NDFL, which itself introduces distortion.
Click to expand...

I agree.

I should also mention another thing. The way syn08 implemented TPC in the YAP output stage looks like it alleviates the concerns about TPC that I mentioned. By taking the feedback off the driver, the "Miller" loop gain is increased relative to taking it off the VAS collectors. And the loading issue looks like it's taken care of as well. I'm guessing he may have interpreted mentioning potential pitfalls of TPC as criticizing YAP, which I wasn't doing at all.

So, have you given up on NDFL then?
 
#2,686
Re: Re: Re: Down with NDFL

Bob Cordell said:
I kind of like TMC. I'm not disputing what you are saying, but the fact that the global feedback loop with TMC is still mainly a straight 6 dB/octave rolloff is appealing to me (as compared to TPC, for example).

Bear in mind that in principle one could combine TMC and TPC.
Click to expand...

It's not a matter of liking or disliking. Of course, I tried to simulate TMC in YAP (it happen to be in an identical topology to the schematic that Andy posted above) and I was unable to get even a half of dB of extra loop gain at 60KHz or 2 degrees of extra phase margin, comparing to TPC. To quote another living classic on DIYAudio above, then "WTF is the point?". Only to get 6dB/oct. rolloff?
 
#2,687
Re: Re: Re: Re: Down with NDFL

traderbam said:


Hi Bob. May I ask why you prefer a straight 6dB/oct roll-off? (As opposed to this roll-off being preceded by steeper roll-offs)
Click to expand...


Hi Brian,

To be honest, it may just be a gut feeling. I have often asserted that different equally smart engineers make different designs largely because of what they fear (differently). Different engineers have different regions of comfort.

For me, I'm a little more fearful of messing up the global loop, possibly by detracting a bit from its phase margin. The global feedback loop is of course suceptible to the vaguaries of different nasty loads. Of course, one could argue that with TMC, the local loop formed is also subject to such vaguaries if the transition does not take place until too high a frequency.

Cheers,
Bob
 
#2,688
Re: Re: Re: Re: Down with NDFL

syn08 said:


It's not a matter of liking or disliking. Of course, I tried to simulate TMC in YAP (it happen to be in an identical topology to the schematic that Andy posted above) and I was unable to get even a half of dB of extra loop gain at 60KHz or 2 degrees of extra phase margin, comparing to TPC. To quote another living classic on DIYAudio above, then "WTF is the point?". Only to get 6dB/oct. rolloff?
Click to expand...


That is rather interesting. Just a little while ago you were saying that all these feedback tricks represented by 3 letter acronyms were functionally the same and everything else is irrelevant.
🙄 🙄 🙄
 
#2,691
Re: Re: Re: Re: Re: Down with NDFL

Bob Cordell said:
Hi Brian,

To be honest, it may just be a gut feeling. I have often asserted that different equally smart engineers make different designs largely because of what they fear (differently). Different engineers have different regions of comfort.

For me, I'm a little more fearful of messing up the global loop, possibly by detracting a bit from its phase margin. The global feedback loop is of course suceptible to the vaguaries of different nasty loads. Of course, one could argue that with TMC, the local loop formed is also subject to such vaguaries if the transition does not take place until too high a frequency.
Click to expand...

There is no easy escape. Well, my particular fear is instability. Makes me shudder, some of the circuits I see posted around the place.
:Ohno:
 
#2,693
TMC vs TPC

Bob Cordell said:
Hi Brian,

To be honest, it may just be a gut feeling. I have often asserted that different equally smart engineers make different designs largely because of what they fear (differently). Different engineers have different regions of comfort.

For me, I'm a little more fearful of messing up the global loop, possibly by detracting a bit from its phase margin. The global feedback loop is of course suceptible to the vaguaries of different nasty loads. Of course, one could argue that with TMC, the local loop formed is also subject to such vaguaries if the transition does not take place until too high a frequency.

Cheers,
Bob
Click to expand...

Hi Bob,

Only 'a gut of feeling'? I'm afraid you have overlooked one important aspect and I'm pretty sure you know it very well.

Concerning TPC, the steep roll-off and the accompanying phase dip somewhere between 10...100kHz results in overshoot of the step response, typically more than 10%. Of course we don't like this and it seems to me that the only way to get rid it, apart from bluntly masking it by means a LP filter in front of the input, is putting a cap or a series R-C across global NFB resistor. Now, what have we gained by this exercise? Nothing, as this additional phase compensation thingie increases the ULG frequency of the global FB loop considerably.
Apart from stability issues which will certainly arise, probably the same amount of distortion reduction would be accomplished by simply increasing the ULGF without all that TPC hassle. In this regard, TPC is nothing more than putting the cart before the horse.
Needless to say that TMC hasn't this kind of issues.

Cheers,
Edmond.
 
#2,695
Re: TMC vs TPC

Edmond Stuart said:
Concerning TPC, the steep roll-off and the accompanying phase dip somewhere between 10...100kHz results in overshoot of the step response, typically more than 10%.
Click to expand...

Do you mean the CL step response? I don't follow you. I think overshoot is caused by poor phase margin rather than maximum OL slope...a TPC scheme doesn't have to have poor phase margin, does it?
Brian
 
#2,696
Re: Re: TMC vs TPC

traderbam said:
Do you mean the CL step response? I don't follow you. I think overshoot is caused by poor phase margin rather than maximum OL slope...a TPC scheme doesn't have to have poor phase margin, does it?
Brian
Click to expand...

Of course I mean CL step response, what else?
The overshoot is not caused by a poor phase margin, rather by the phase dip around 30kHz or so, see below. In a next post I'll show the step response obtained by using ideal components (VCCS etc.)

Regards,
Edmond.
 

Attachments

  • tpc1.png
    tpc1.png
    8.2 KB · Views: 440
#2,698
Re: Re: TMC vs TPC

traderbam said:


Do you mean the CL step response? I don't follow you. I think overshoot is caused by poor phase margin rather than maximum OL slope...a TPC scheme doesn't have to have poor phase margin, does it?
Brian
Click to expand...

Overshooting is always because of a second order transfer function, feedback has no direct relevance here. Feedback doesn't change the system order but it does change the poles positions in the complex plane. Understanding the poles trajectories when closing the loop is essential here, to predict the CL overshoot.

TPC creates a second order system (and so does TMC or NDFL, etc... at least locally in the frequency domain or complex plane) and each of these can create a significant overshoot if poorly designed. To assert that TPC is always generating an overshoot is certainly a stretch, as much as asserting that TMC never creates an overshoot.
 
You must log in or register to reply here.
Top Bottom