There are issues with Error Correction and the relation to feedback/feedforward that I don't quite understand yet.
I have read Bob Cordell's explanation of Hawksford EC as an embedded feedback loop but neither he or Hawksford consider the fact that the EC loop is not ideal.
Lipshitz and Vanderkooy did a comparison of EC with feedback and apparently concluded that feedforward needed to be included.
That seems to involve some of the same issues that worry me but I don't know how they relate until I read the paper.
Unfortunately it's a conference paper, and my library only has the Journal of the AES, not full access.
Anyone have a link or understand the paper?
Lipshitz and Vanderkooy "Is zero distortion possible with feedback?"
76th Convention of the AES, 1984, New York.
Best wishes
David
I don't have the latter Hawksford but I did try to show some of poor thinking about HEC, how it should just be viewed as global negative feedback around a gain boosting positive feedback "error correction" loop:
http://www.diyaudio.com/forums/solid-state/240712-cfa-topology-audio-amplifiers-229.html#post3723022
collects a few post of mine, I have more sims showing various ways to see that it fits in Horowitz' formalism in Bob's HEC thread
http://www.diyaudio.com/forums/solid-state/240712-cfa-topology-audio-amplifiers-229.html#post3723022
collects a few post of mine, I have more sims showing various ways to see that it fits in Horowitz' formalism in Bob's HEC thread
I tend to see it much as you do, I hope that is because accurate analysis should arrive at the same, correct answer.
But I must admit that it was one of your sims that first made me take a closer look at this, so perhaps I have been influenced to use a certain viewpoint.
I have started to read over the Cordell EC thread but there are >4,000 posts and I haven't reached your contribution there yet.
Best wishes
David
Common Emitter Error Correction?
I have had a look at the Hawksford papers and had an idea, why not apply Error Correction to a common emitter OPS?
I have never seen this but most ideas in audio have already been tried, any of the experts here aware of any prior art?
Alternatively, anyone here have such a circuit😉
Best wishes
David
I have had a look at the Hawksford papers and had an idea, why not apply Error Correction to a common emitter OPS?
I have never seen this but most ideas in audio have already been tried, any of the experts here aware of any prior art?
Alternatively, anyone here have such a circuit😉
Best wishes
David
Hi David,
The FF EC implementation, used by Takahashi and Tanaka, is rather universal - I suppose, it will work in CE (CS) configuration, the same way, as it works in CC (CD) one. My experiments with this kind of approach have shown good results in differemt BJT and MOSFET based OPS arrangements, although I didn't touch any CE ones at that time.
I refer to their paper "Design and Construction of a Feedforward Error-Correction Amplifier", presented at the 65th Convention of the Audio Engineering Society, London, 1980 February 25-28.
EC approaches, desigbed by Giovanni Stochino are also external to the amplifier, being corrected, so those will work for CE OPS as well.
Cheers,
Valery
The FF EC implementation, used by Takahashi and Tanaka, is rather universal - I suppose, it will work in CE (CS) configuration, the same way, as it works in CC (CD) one. My experiments with this kind of approach have shown good results in differemt BJT and MOSFET based OPS arrangements, although I didn't touch any CE ones at that time.
I refer to their paper "Design and Construction of a Feedforward Error-Correction Amplifier", presented at the 65th Convention of the Audio Engineering Society, London, 1980 February 25-28.
EC approaches, desigbed by Giovanni Stochino are also external to the amplifier, being corrected, so those will work for CE OPS as well.
Cheers,
Valery
Hi Valery
I mentioned the Hawksford papers but I should have been more clear that I meant a Hawksford, so-called Error Correction CE OPS.
True FForward EC is indeed rather universal, and also has less theoretical constraints, so I should study it more but not quite ready yet.
The Stochino FF is very impressive but I want to avoid the complexity of his transformer based "feed-forward summer" so I will see what I can do with Hawksford s-c EC first.
Best wishes
David
David, you can look this kind of EC suggested by smms73. http://www.diyaudio.com/forums/solid-state/260308-gainwire-ngnfb-classb-poweramp.html#post4018576
Damir
Damir
The hyperbole used around this topic is amusing. To clarify:
You can cancel some distortion by summing
You can reduce some distortion by degenerative feedback
By far the most effective approach is summing but it is limited by the availability of components with accurate complementary distortion.
Feedback is an attempt to spend gain to buy accuracy. But there is only so much gain to spend and there are classes of distortion that are not for sale. So feedback's effectiveness is always limited.
And remember this: time cannot be turned backwards. Nothing can fix a time delay.
You can cancel some distortion by summing
You can reduce some distortion by degenerative feedback
By far the most effective approach is summing but it is limited by the availability of components with accurate complementary distortion.
Feedback is an attempt to spend gain to buy accuracy. But there is only so much gain to spend and there are classes of distortion that are not for sale. So feedback's effectiveness is always limited.
And remember this: time cannot be turned backwards. Nothing can fix a time delay.
Hi David,
I did not want to show OPS, but EC part, and that could be used with any OPS type.
I didn't consider CE OPS, I don't see any advantage over CC OPS. I could be wrong but for me it's not of high importance to have a real to real swing.
BR Damir
The rail to rail means improved efficiency, which is nice but not my main interest.
EF is basically local feedback but with CE I can use feedback to improve the whole loop.
EC is, in a sense the continuation of this idea, local positive feedback to improve gain that can be used on the whole loop.
It is best if EC can be included as a very close-coupled loop, so I want the circuit of the EC and OPS to be as unified as possible.
Bob Cordell does this well with his EF version and I wanted to see if it had already been done before I tried a CE variation.
I will have a closer look at your circuit to see if it can incorporate a CE OPS neatly.
Best wishes
David
Just a random thought about the title of this thread...
Whether the output devices are connected in common collector or common emitter, if they are current driven, the result should be the same, at least at low frequencies, except polarity is inverted of course.
I mean, if you drive current into the base of the NPN, for instance, you get hFe times that on the output, no matter if it's wired in emitter follower or common collector.
The big difference comes at HF : common collectors pass the drive current to the output through their BE capacitance, which allows the output to be driven directly at HF even when the transistors run out of speed. However, in common emitter mode, when this happens, drive current passes through Cbc capacitors, and the output stage no longer provides inverting current gain, which creates an "interesting" phase shift...
Whether the output devices are connected in common collector or common emitter, if they are current driven, the result should be the same, at least at low frequencies, except polarity is inverted of course.
I mean, if you drive current into the base of the NPN, for instance, you get hFe times that on the output, no matter if it's wired in emitter follower or common collector.
The big difference comes at HF : common collectors pass the drive current to the output through their BE capacitance, which allows the output to be driven directly at HF even when the transistors run out of speed. However, in common emitter mode, when this happens, drive current passes through Cbc capacitors, and the output stage no longer provides inverting current gain, which creates an "interesting" phase shift...
"Current driven OPS"
A good idea if the distortions you care about are lesser in current mode than in transconductance mode.
You could make a current-driven OPS that drives a shunt resistor - certainly no larger than 1 ohm to give a low impedance to the speaker. More power more distortion, tho. Otherwise, you have to turn to the Dark Side and use global feedback to control the output voltage.
In practice, pure current drive is not possible, moreso at high f where parasitics will lower the Z. But a well-designed, open-loop, output shunted amp could sound pretty good.
To be realistic, all power transistors are hives of distortion no matter how you drive or arrange them. The topology will depend on the specific device(s) and your specific output requirements.
A good idea if the distortions you care about are lesser in current mode than in transconductance mode.
You could make a current-driven OPS that drives a shunt resistor - certainly no larger than 1 ohm to give a low impedance to the speaker. More power more distortion, tho. Otherwise, you have to turn to the Dark Side and use global feedback to control the output voltage.
In practice, pure current drive is not possible, moreso at high f where parasitics will lower the Z. But a well-designed, open-loop, output shunted amp could sound pretty good.
To be realistic, all power transistors are hives of distortion no matter how you drive or arrange them. The topology will depend on the specific device(s) and your specific output requirements.
Yes, there are really two issues that are a bit interconnected.
First I started to think about CE OPS then that made me reconsider the whole subject of how the OPS is driven.
My ideas on current drive are very fuzzy, mainly Cherry's observation in JAES article about the different implications of Beta non-linearity compared to Vbe/Ic non-linearity.
This seems to fit in with some notes from TU Delft about current drive.
Just want to kick the idea around, maybe learn a few references or links.
My ideas on CE OPS are clearer, have found quite a few patents for IC circuits that discuss aspects, mainly how to bias.
The failure of CE to be minimum phase has occurred to me, not a problem for an OPS with typical bias I would think, thanks for the reminder to check it.
Best wishes
David
Output capacitance of NJL4281/4302 is <1000pF, should be OK if the gain of the OPS is reasonable, say 5 would avoid too much Miller multiplication,
Last edited:
I've not read this entire thread, but the way I'd think of this is to take the output stage as
everything past the VAS. I think of a virtual ground input where the VAS drives current into
it. Think of the output stage as an inverting OP amp with a virtual ground input.
I've not thought of a best way to implement this.
Reviewing the old Marantz 250 power amp gave some suggestion to this idea but the input
is referenced to the rails which are essentially ground. I don't think that it is a very good
design by the way.
everything past the VAS. I think of a virtual ground input where the VAS drives current into
it. Think of the output stage as an inverting OP amp with a virtual ground input.
I've not thought of a best way to implement this.
Reviewing the old Marantz 250 power amp gave some suggestion to this idea but the input
is referenced to the rails which are essentially ground. I don't think that it is a very good
design by the way.
The VAS becomes an AS, not sure what AS is?
The VAS would not have to swing a lot if any voltage, that alone should
be an advantage. I have not really put much thought into the output
stage, yes the obvious thing would be for it to have its own VAS but
perhaps some CFP with gain configuration could be set up with a
virtual ground input. The Tiger output stage might provide some
ideas but it is not configured as a virtual ground input. The Bryston
with more (current) gain might also provide some ideas. Quite the
complex output stage BTW.
Many years ago (as a teenager) I thought what is the big deal about fully
nested feedback? Each block does not know how the other block got its
transfer function, feedback just changes the transfer function. I bring
this up since there would be nested feedback.
The VAS would not have to swing a lot if any voltage, that alone should
be an advantage. I have not really put much thought into the output
stage, yes the obvious thing would be for it to have its own VAS but
perhaps some CFP with gain configuration could be set up with a
virtual ground input. The Tiger output stage might provide some
ideas but it is not configured as a virtual ground input. The Bryston
with more (current) gain might also provide some ideas. Quite the
complex output stage BTW.
Many years ago (as a teenager) I thought what is the big deal about fully
nested feedback? Each block does not know how the other block got its
transfer function, feedback just changes the transfer function. I bring
this up since there would be nested feedback.
and this is of course the advantage of the most important type of current driven OPS ... pure Cherry 🙂
Great Guru Baxandall first drummed this into me in some Jurassic correspondence.peufeu said:
IME, this is the most important case of non-minimum phase behaviour in PAs. It usually shows itself around the VAS.
IIRC, Damir shows how slight changes in compensation can move the LG from MP to NMP.
_________________
I've shown a number of very simple designs which (at least in SPICE world) have 1ppm THD20k at 50W. Guru Zan has criticised these as having zillion GHz ULGF.
But other (much more complex) designs with 1ppm THD20k 50W are most likely zillion GHz ULGF too.
The advantage of my simple designs is that 'real life' is MUCH more likely to come close to SPICE world.
These designs rely on 'current drive' of the OPS via pure Cherry compensation ... which also gives loadsa extra LG 😀
Last edited:
- Status
- Not open for further replies.