Feedback question

Yes, you're right about being too simplistic.

My back-of-envelope feedback equation calculation did not use the right inputs. I neglected what would happen if phase shift exceeded 180 degrees at gain >=1.

Of course cascading several stages with significant phase shifts at the cut-off frequency will add the phase shifts at that frequency, as feedback won't be able to function with gain =1. Depending on the amount of feedback used that could lead to oscillation if global feedback is used.

So simplistically :0) could we say for an n stage amplifier the unity gain ( cutoff ) phase shift in degrees per stage must be kept below 180/n ( or better for adequate phase margin, 120/n ?). Depending on amount of feedback again of course, that's for 100% feedback which is the worst case.

Non-linearities are an additional problem on top of phase shift I suppose.

"Jim Thompson" wrote in message news: snipped-for-privacy@4ax.com...

Simple-minded math doesn't take into account non-linearities. Local feedback not only reduces distortion, but broadbands the local pieces, often making it easier to get stable overall feedback.

...Jim Thompson

--
| James E.Thompson, P.E.                           |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC\'s and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
 I love to cook with wine     Sometimes I even put it in the food

A lot of difficult mathematics with a much more representative model will reveal the problem.

Each stage varies a bit from the ideal, and changes with age and temperature, and has nonlinearities that make it diverge even more at extremes of input.

Trying to wrap feedback around the whole thing is asking for disaster. Wrapping feedback around each stage makes the whole task much easier.

--
http://www.wescottdesign.com

In fact, it's very commonly done. It can aid in maintaining stability and linearity of the multi-stage amplifier, for a number of reasons.

Even in cases where there is no obvious local feedback loop, it's often there in a more subtle way (e.g. an emitter-degeneration resistor).

I'd hazard a guess that the feedback equation you're looking at, is describing the system's behavior at DC... and in this situation there isn't a lot of difference between the two.

However... consider a more complex scenario: a system handling AC signals with a fairly wide bandwidth, and in which the unity-gain bandwidth of the earlier stages of the amplifier is significantly higher than the bandwidth of the later stages. [Classic example: audio amplifier with relatively fast small-signal transistors in the front end, and slower power transistors in the finals.]

Figure out how a system of this sort (with only global feedback) would behave if hit with signals above the output stage's unity-gain cutoff frequency.

Short of oscillation (which definitely can occur in serious cases) you can run into poor behavior if the earlier stages are faster than the later stages.

You can reduce some of these problems by bandwidth-limiting the

*input* of the amplifier chain, but keeping the whole chain clean and stable is often easier if you incorporate some local feedback in the faster stages.

That can be part of it, I suppose.

Yet another reason is that using local feedback (and maintaining amplifier linearity through each individual stage) can give you an amplifier chain which recovers faster (and more cleanly) if its output stage is momentarily overdriven or overloaded in some way.

Still another reason is that using local feedback on each stage, and reducing the amount of global feedback (both in amount and in frequency range) may help reduce the amplifier chain's susceptibility to unwanted signal incursion (e.g. RF getting into the outputs).

--
Dave Platt                                    AE6EO
Friends of Jade Warrior home page:  http://www.radagast.org/jade-warrior
  I do _not_ wish to receive unsolicited commercial email, and I will
     boycott any company which has the gall to send me such ads!

Because each stage is more or less nonlinear, and the response of the whole thing could be peculiar, highly sensitive and difficult to analyse. With the global feedback only, you will have to count for the worst case, and that will not allow you to get the optimal performance.

:)))) Little mathematics. The global feedback can represent the intractable problem.

Also, the global feedback path can require multiple turns of phase vs frequency, so it should be implemented as non-trivial filter; highly sensitive, too.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

ere

een

d n

ce

ew?

Often enough, DC feedback helps prevent thermal runaway. Nothing prevents the output/input feedback from being the dominant loop in the amplifier.

Composite amplifiers, where the input op amp's feedback loop wraps around the second stage, often show this kind of bad behaviour. Looks great in a linear analysis, behaves badly in real life, even though they do have local feedback on the second stage. For some reason, just about everybody wants to do this in their first photodiode preamp. Too fancy for their own good.

Cheers

Phil Hobbs

-- Dr Philip C D Hobbs Principal ElectroOptical Innovations

55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net

Try to find a paper with the title "Nested Differentiating Feedback Loops" by Cherry. IIRC some audio power amps were made using these techniques that had (for the time) very good distortion performance.

BTW, the "differentiating" part of the title meant that the localised feedback was AC coupled; DC feedback was only applied at the global level.

Regards, Allan

ere

een

d n

ce

ew?

It is pretty hard to compensate more than two gain stages in a loop. Not impossible, since there are op amps with 3 gain stages, but often less is more, keep it simple stupid, or any number of rules of thumb comes to mind. You might want to read up on pole splitting.

amplifier?

the

will=20

Not as good of a response as i hade hoped from you. : One of the things that per stage feedback provides is per stage placement of poles and zeroes. Why is this done you ask? Simple, with the the per stage poles and zeroes can be stabilized against overall feed back, thus the total system stability is much easier to manage. Plus there is now an abundance of controllable parameters to use in tuning overal system performance. The system becomes easier to design, in some respects, at the expense of a requirement of more engineering feel for setting the initial system and per stage properties. Though that is now being reduced by simulation cpabilities, the simulations still need experienced hands for best results. .