AndrewT said:
Yes, it keeps the OLG flat up to a higher frequency and then it falls of quicker above that since you really have two poles kicking in.
Slone likes to use it a lot, and he acknowledges the stability problems. He comments in his book that he mostly uses it conservatively and only maintain a flat response up to around 10 kHz, to improve stability for difficult loads, while it could have been used to go much higher.
AndrewT said:
Difference between the lines is the extra feedback ....
/sreten.http://www.audiodesignline.com/howto/205207238
My thoughts:
1. There are differences between amps designed for class A and biased in class A.
2. Any active element distort. Even passive elements distort, for example when I worked on thick film IC power amps I've found thermal distortions of resistors in feedback well measurable. However, "real" resistors have bigger time constant of thermal distortions, so on high frequencies they add unmeasurable distortions, but the point is about any active element we know about.
3. We amplify power. It means, if using deep feedback by voltage we decrease output resistance of the amp and increase it's input resistance a lot, anyway it's output and input resistances are non linear and frequency dependent
4. Due to need for compensation for stability of amps with deep feedback and more than one stage we have to roll of highs, i.e. to increase non-linearity of input and output resistances.
5. And, the last, if Douglas said that some distortions are negligible in respect to other kinds of them he could not say that they don't exist at all.
Recipe:
1. As less as possible of stages, with power amplification in mind.
2. As fast and as linear as possible each stage, probably with local feedbacks.
If to optimize the design using this recipe the amp never will be made as a classical opamp with a differential input stage, VAS stage, complementary emitter followers biased to class a.
1. There are differences between amps designed for class A and biased in class A.
2. Any active element distort. Even passive elements distort, for example when I worked on thick film IC power amps I've found thermal distortions of resistors in feedback well measurable. However, "real" resistors have bigger time constant of thermal distortions, so on high frequencies they add unmeasurable distortions, but the point is about any active element we know about.
3. We amplify power. It means, if using deep feedback by voltage we decrease output resistance of the amp and increase it's input resistance a lot, anyway it's output and input resistances are non linear and frequency dependent
4. Due to need for compensation for stability of amps with deep feedback and more than one stage we have to roll of highs, i.e. to increase non-linearity of input and output resistances.
5. And, the last, if Douglas said that some distortions are negligible in respect to other kinds of them he could not say that they don't exist at all.
Recipe:
1. As less as possible of stages, with power amplification in mind.
2. As fast and as linear as possible each stage, probably with local feedbacks.
If to optimize the design using this recipe the amp never will be made as a classical opamp with a differential input stage, VAS stage, complementary emitter followers biased to class a.
Wavebourn said:
...like an amp designed to drive real speakers in order to fool human imaginations.
I forgot to add to the recipe that distortions and their dynamics should be weighted against audibility of them.
Actually, I think this is one of the most sensible answers to the question. I am afraid the advice is not that simple to follow though.
Otherwise we have the excellent component called voltage controlled voltage source. Unfortunately the manufacturers still seem to have problems putting it on the market.
Lumba Ogir said:
It is generally accepted that the practical limit for OLG at 20KHz is about 20dB. The point is that regardless of OL bandwidth it is in effect the amount of gain at 20, 50, or 100KHz that is the limiting factor. Once you have hit the stability limits for OLG at HF the actual bandwidth becomes a moot point.
AndrewT said:
OK, it should exactly read OLG - CLG = 20dB @ 20kHz
I admit I'm generalising but it was meant to read: As a rule of thumb 20dB of feedback at 20KHz is adequate to reduce distortion but unlikely to cause instability problems.
DSelf and Bob Cordell have both thrown this figure about so I've taken it on board as good advice for a novice builder / designer.
I hope the above explanation is more plausible.
Hi,
The feedback factor at 20kHz is mentioned by Self, and with single
pole compensation he has used up to 34dB, but prefers somewhat
less, 20dB though is a low figure, lower than used for most amplifiers.
He is vague on the subject but suggests 40dB is asking for trouble.
Becomes fairly meaningless with other compensation
schemes, e.g. for 2-pole it can be > 60dB at 20Khz.
/sreten.
The feedback factor at 20kHz is mentioned by Self, and with single
pole compensation he has used up to 34dB, but prefers somewhat
less, 20dB though is a low figure, lower than used for most amplifiers.
He is vague on the subject but suggests 40dB is asking for trouble.
Becomes fairly meaningless with other compensation
schemes, e.g. for 2-pole it can be > 60dB at 20Khz.
/sreten.VHF man said:
OK, now that I've re read the remainder I can see it's confusing. My argument was really concerned with open loop bandwidth so if the original statement from Sandy K was referring to closed loop bandwith then I'm in agreement with it 100%.
If you can build an amplifier with a wide closed loop bandwidth that's stable into any load (within reason) then there's a good chance that it will also have have low distortion at 20KHz. However, in the real world there are practical limits imposed by circuit layout as well as design topology and device selection etc..
sreten said:Hi,
The feedback factor at 20kHz is mentioned by Self, and with single
pole compensation he has used up to 34dB, but prefers somewhat
less, 20dB though is a low figure, lower than used for most amplifiers.
He is vague on the subject but suggests 40dB is asking for trouble.
Becomes fairly meaningless with other compensation
schemes, e.g. for 2-pole it can be > 60dB at 20Khz.
OK thanks for the info and it makes sense. Actually my own poweramp is a derivative of a D Self design with single pole comp and I'm happily getting away with about 30dB of fb at 20KHz.
I have experimented with 2 pole C but found that the VAS distortion rise mitigated against it and I preferred the sound of single pole. (The VAS was simple / single ended so it could have been improved on). I also suspect that the amount of rail decoupling becomes even more critical when implementing 2 pole C in this design.
I can see recommending 200-500 KHz loop gain intercept if you want to avoid load stabilizing networks, or are offering diy beginner circuits where they may not even have a 'scope
but feedback amplifier loop gain crossover frequency should be able to be made much higher with modern output devices
BJT with 30-60MHz ft should be easy to keep stable at 2 MHz
Mosfets can in principle go lot higher but packaging parasitics and ordinary layout/construction mean most people don't push them more than 2-3x higher for ~ 5MHz loop gain intercept - like Bob's Mosfet Amp
these 2-5 MHz numbers give 40 dB loop gain @20 KHz even with single pole compensation, 2-pole would easily allow 60 dB
but feedback amplifier loop gain crossover frequency should be able to be made much higher with modern output devices
BJT with 30-60MHz ft should be easy to keep stable at 2 MHz
Mosfets can in principle go lot higher but packaging parasitics and ordinary layout/construction mean most people don't push them more than 2-3x higher for ~ 5MHz loop gain intercept - like Bob's Mosfet Amp
these 2-5 MHz numbers give 40 dB loop gain @20 KHz even with single pole compensation, 2-pole would easily allow 60 dB
VHF man said:
Hi,
For Class aB I'd say the extra 2nd harmomic distortion is well worth
the trade off against the reduction of higher harmonics (on paper).
(also causes some extra slew rate limiting, but it is not drastic.)
Rail decoupling ? according to Self ? where he shows the PSRR for
the negative rail (also see many op-amp specs) follows the the
feedback factor completely, this is the main reduction mechanism.
The implication for 2-pole is therefore higher negative rail PSRR
with 2 pole compared to 1 pole due to more feedback factor at hf.
For those not familiar Self loses around 1V on the negative rail
to symmetrise the clipping point, and uses this to decouple the
negative rail to bring the PSRR close to the positive rail value.
/sreten.edit : Self 2 pole is around a "darlington" 2 transistor Vas, the most
elegant way would probably to also cascode the input to the Vas
and use the cascode input as the compensation reference point.
jcx said:
Does it really matter that the output devices are high ft? From a VAS point of view the falling hfe of the EF output stage looks like a capacitive impedance. As long as the VAS has a low enough output impedance and can source the output current then the output stage fT should be very flexible.
the ultimate limits in applying feedback are "nonminimum phase" effects like carrier transit time
often the "earlier" practical limit is the margins you have to allow for uncertainty in the location of higher order poles
this uncertanty allowance has to accomodate the varying bias conditions of the output transistors with ft being both current and voltage sensitive as well as there being device production parameter variations
add in the unspecified loudspeaker and cable loading effects and I think you have to build in the robustness by backing off of ft by ~ an order of magnitude - at least this appears to be common practice
it's also nice to actually get power gain out of your output devices - although not strictly requred at the gain intercept, I think in some designs fast drivers actually "feedthrough" the slow ouput device Cbe to improve high frequency phase beyond the output Q ft
Halcro has pursued both higher frequency gain intercept and higher order compensation so "typical practice" doesn't prove some of these effects are insurmountable limits - just mostly too much effort for most applications
often the "earlier" practical limit is the margins you have to allow for uncertainty in the location of higher order poles
this uncertanty allowance has to accomodate the varying bias conditions of the output transistors with ft being both current and voltage sensitive as well as there being device production parameter variations
add in the unspecified loudspeaker and cable loading effects and I think you have to build in the robustness by backing off of ft by ~ an order of magnitude - at least this appears to be common practice
it's also nice to actually get power gain out of your output devices - although not strictly requred at the gain intercept, I think in some designs fast drivers actually "feedthrough" the slow ouput device Cbe to improve high frequency phase beyond the output Q ft
Halcro has pursued both higher frequency gain intercept and higher order compensation so "typical practice" doesn't prove some of these effects are insurmountable limits - just mostly too much effort for most applications
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.