Feedback & audio amp

jcx

Member
2003-02-17 7:38 pm
..
will that be global or local?

Feedback is limited by stability requirements, at some high frequency all physical systems have excess phase shift which turn the desired negative feedback into oscillation causing positive feedback

For audio power amps with conventional output devices/packaging the unity loop gain intercept is usually limited to (optimistically) <20MHz for mosfets or <5MHz for bjts, the choice of how you roll off loop gain to reach the unity gain intercept then determines your allowable audio frequency loop gain

Single pole dominant compensation is most popular and the most conservative, limiting loop gain roll off to 20bB/decade which limits loop gain to 40-60 dB at high audio frequencies (work backwards from 20 MHz; 20KHz to 20 MHz is 3 decades of room to roll off the audio frequency loop gain)

Bode’s maximum feedback can be approximated with multiple pole-zero compensation and increases gain slope to 30dB/decade; 80-90 dB loop gain would be possible – but you never see this type of compensation

2-pole compensation abandons Bode’s unconditional stability and may require inspired clamping/limiting of internal nodes to avoid bad clipping/current limiting behavior but allows over 100 dB loop gain at audio frequencies

Additional local feedback loop gain is possible, nested feedback loops have been used around the output stage so that the output stage may be wrapped in 100 + dB of local gain which is not visible to the outer loop – your 200 dB number might be approached in this way by a Halcro amp's output stage
 

tlf9999

Disabled Account
2005-05-25 2:25 am
none
jcx said:
Feedback is limited by stability requirements, at some high frequency all physical systems have excess phase shift which turn the desired negative feedback into oscillation causing positive feedback



I disagree with that: as long as you have small enough of a gain (less than unity) when the feedback turns positive, you will have a stable system.

Instability isn't inherent to negative feedback. It is the combination of gain and phase shift that cause feedback to be unstable.
 

1001

Member
2004-07-19 9:38 pm
jcx said:
will that be global or local?

Feedback is limited by stability requirements, at some high frequency all physical systems have excess phase shift which turn the desired negative feedback into oscillation causing positive feedback

For audio power amps with conventional output devices/packaging the unity loop gain intercept is usually limited to (optimistically) <20MHz for mosfets or <5MHz for bjts, the choice of how you roll off loop gain to reach the unity gain intercept then determines your allowable audio frequency loop gain

Single pole dominant compensation is most popular and the most conservative, limiting loop gain roll off to 20bB/decade which limits loop gain to 40-60 dB at high audio frequencies (work backwards from 20 MHz; 20KHz to 20 MHz is 3 decades of room to roll off the audio frequency loop gain)

Bode’s maximum feedback can be approximated with multiple pole-zero compensation and increases gain slope to 30dB/decade; 80-90 dB loop gain would be possible – but you never see this type of compensation

2-pole compensation abandons Bode’s unconditional stability and may require inspired clamping/limiting of internal nodes to avoid bad clipping/current limiting behavior but allows over 100 dB loop gain at audio frequencies

Additional local feedback loop gain is possible, nested feedback loops have been used around the output stage so that the output stage may be wrapped in 100 + dB of local gain which is not visible to the outer loop – your 200 dB number might be approached in this way by a Halcro amp's output stage

I refer to global, overall, whole feedback factor, of course

Why do I never see multiple pole-zero compensation? What do you mean? Doesn't exist commercial amp (pre amp, headphone amp, opamp, etc.) with this type of compensation?

2-pole compensation allows over 100 dB loop gain at audio frequencies would mean as well 200-300 dB?

very interesting... so an Halcro amp's has 100 dB of FB in the output stage and 100dB in the global loop, is it true? :bigeyes:
 

jcx

Member
2003-02-17 7:38 pm
..
tlf9999;

feel free to be more precise in a short quick post, be sure to mention minimum phase/nonminimum phase, device carrier transist time, Bode's integral of feedback...


1001;

I'm merely speculating based on what little is public about Halcro's amp - he claims to use "video op amps" in local loops around the output devices - probably only 60-80 dB local V gain - what the final number is is only known by Halcro himself - but if you want to spend the $25K it is perfectly legal to reverse engineer one!

Prof Ed Cherry wrote about nested feedback loop audio amplifiers in the 70's

Bode's maximum feedback has a s^1.5 slope - it requires many R-C stages to approximate this slope - see "pink noise" filters for an idea of fractional roll off approximation with discrete parts

high feedback is a tool with limitations, no one has demonstrated that 200 + dB feedback is necessary for audio - few audio playback signal sources can deliver 120 dB S/N

as a precision instrumentation designer I know how many layers of implementation detail have to be addressed to get to even 16-bit resolution - high feedback is only a minor part - uOhm amplifier out Z at the feedback terminal is meaningless with 100s of mOhm cable and 1-10 mOhm connectors to mention only one limitation - measuring Halcro's ppm distortion is an exercise in defining where the amplifier’s output is - to avoid nonlinear load currents in interconnect z from swamping the measurement

if you really want to pursue high feedback then b j Lurie's "Feedback Maximization" is the book for you - after completing a undergrad controls curriculum –the pricey “Classical Feedback Control” by Lurie is more readable but also full of publishing errors – not the book to learn the material from cold
http://www.luriecontrol.com/ClassicalFeedbackControl.htm
 
Hi jcx,

"what the final number is is only known by Halcro himself - but if you want to spend the $25K it is perfectly legal to reverse engineer one!"

I have designed a simple amplifier with only 1 chip and one nested discrete output loop which achieves THD < 0.00005%, possibly lower than Halcro, but what's the point, when audiophiles want euphonic distortions of maybe 0.1% or so. Better to just engineer out the nasties and include the nice ones.
So we need to focus on PSRR to get rid of those commutation artefacts while nurturing the soft low order CM 'bloom'. Nice to take the vagaries of output stage crossover and load dependent transcondductance variations out of the equation.


Cheers,
Greg
 

jcx

Member
2003-02-17 7:38 pm
..
Greg;

maybe you're just not charging enough to get the "the price IS the object" snob appeal

Halcro may not have broken new amplifier design/performance ground - but his marketing seems to be working

1001;

a further practical limitation on high feedback is the noise of the feedback measurement itself - putting enough power through the feedback resistor network to get >~150 dB S/N starts to cause nonlinear response from power dissipation in the feedback resistors!
 

1001

Member
2004-07-19 9:38 pm
Well, I realize (by your post) it's impossible to build a stable amp with 200-300 dB of overall feedback, but If It could be possible it could have the best electrical (sonic?) performance regard to THD and noise, is it true?

Greg;
An amp whit THD of 126dB is an extraordinary amp but of course you haven't broken new amplifier design/performance ground. Halcro amp's have a better (maybe the best, maybe...) THD value.:D
 

jcx

Member
2003-02-17 7:38 pm
..
Bode's maximum feedback loop gain curve’s slope is set to give a fixed minimum phase margin at any frequency - 30-60 degrees is the typical recommendation

this is to assure that for nonlinear operation that reduces the effective amplifier gain without much affecting the phase shift the feedback won't cause the reduced gain amplifier to oscillate/behave badly - tube amps are apparently especially problematic since their gain sweeps from zero to the operating gain slowly each time the amp is powered up and the phase shift is largely determined by parasitic C that doesn’t change too much with bias point

with 2 pole compensation you have 180 degrees phase shift in the feedback network = zero (or negative if the amp contributes any phase shift) phase margin over the region where the loop gain slope is near 40dB/decade –this region of the Nyquist diagram plot “pokes up” above the 180 phase axis and as the amp gain grows radially from 0 to the final gain locus there is an intermediate gain region where the magnitude 1, 180 degree point is encircled by the gain locus

for ss amps the turn on transient can be much faster and more complicated - and semiconductor junction C is 1st order sensitive to bias point - if the operating bias point is reached much faster than a cycle of the potential feedback oscillation frequency there may not be a problem

but nonlinear operation such as clipping or slew rate limiting reduces the effective gain of the amplifier viewed at the fundamental frequency of the input frequency and the amplifier can show jump resonance (= “hysteric” clipping, you have to reduce the input level well below the initial clipping level for the amp to return to normal linear operation) or even self sustaining large signal oscillation that is triggered by input in the frequency region of negative phase margin even though the linear stability margins are fine at the much higher loop gain intercept frequency

despite the potential problems 2-pole compensation seems to work in practical amplifiers (the region of pure 2-pole slope shouldn’t be large in audio amps so the full 180 degree feedback phase shift isn’t usually reached) – but you really need to address the potential clipping behavior if you try to go even further – ever see a “3 pole” loop gain amplifier?
 
jcx said:
...but nonlinear operation such as clipping or slew rate limiting reduces the effective gain of the amplifier viewed at the fundamental frequency of the input frequency and the amplifier can show jump resonance (= “hysteric” clipping, you have to reduce the input level well below the initial clipping level for the amp to return to normal linear operation) or even self sustaining large signal oscillation that is triggered by input in the frequency region of negative phase margin even though the linear stability margins are fine at the much higher loop gain intercept frequency

despite the potential problems 2-pole compensation seems to work in practical amplifiers (the region of pure 2-pole slope shouldn’t be large in audio amps so the full 180 degree feedback phase shift isn’t usually reached) – but you really need to address the potential clipping behavior if you try to go even further – ever see a “3 pole” loop gain amplifier?

Yes...Roberge attempts to show potential problems with double-pole compensation....such as jump-resonance...etc..

But these appear to be specific to his implementation....and are not generaly applicable...in my experiance at least...
 
1001 said:


2-pole compensation allows over 100 dB loop gain at audio frequencies...


Exceedingly unlikely...unless said compensation includes the output stage....

In which case the frequency of unity loop-transmission would exceed 20MHZ! :bigeyes:

Recipe for intractable minor-loop instability, unless the dominant singularities of your output stage reside beyond 20MHz....

Only power MOSFETs need apply...methinks? :scratch2: