Hi,
Pre-amplifier stage
Power amplifier stage:-
-Input stage
-Driver stage
-Power stage
From what I understand the input stage will be bandwidth limited with feedback often applied from power to the inverted input stage. The feedback is bandwidth limited to prevent oscillation of the power amplifier stage. The power stage should be linear in operation.
Does this mean that feedback in amplifier is ultimately the driver of sound quality ? (rather than esoteric parts)
Pre-amplifier stage
Power amplifier stage:-
-Input stage
-Driver stage
-Power stage
From what I understand the input stage will be bandwidth limited with feedback often applied from power to the inverted input stage. The feedback is bandwidth limited to prevent oscillation of the power amplifier stage. The power stage should be linear in operation.
Does this mean that feedback in amplifier is ultimately the driver of sound quality ? (rather than esoteric parts)
Bandwidth is generally determined by the initial recording.
Tape, CD, Phono, Radio are all bandwidth limiting.
Feedback enhances overall linearity.
Some feel it can be detrimental.
Others believe it fixes more than it hurts.
YMMV.
Tape, CD, Phono, Radio are all bandwidth limiting.
Feedback enhances overall linearity.
Some feel it can be detrimental.
Others believe it fixes more than it hurts.
YMMV.
Oh, global feedback can definitely be detrimental to the sound. No doubt about it. Basic feedback theory doesn't mention this because basic feedback theory is based on linear systems.
The problem with audio amps is their non-linearities, of course.
Designers use feedback in audio amps as a technique for trading gain for linearity. But like any technique it must be used with care to avoid unwanted side effects. There are a lot of potential unwanted side-effects.
One solution is to avoid feedback wherever possible. This tends to lead to big, heavy, energy wasteful, expensive designs that are costly to manufacture as they require careful component matching across many parts.
Another solution is to use feedback with proper care. This tends to lead to small circuits with fewer components, relatively low power wastage, cheaper to buy and cheaper to manufacture because there are less parts.
Whether feedback is used or not don't lose sight of the fact that it is the non-linearities of the components that ultimately determines the sound.
The problem with audio amps is their non-linearities, of course.
Designers use feedback in audio amps as a technique for trading gain for linearity. But like any technique it must be used with care to avoid unwanted side effects. There are a lot of potential unwanted side-effects.
One solution is to avoid feedback wherever possible. This tends to lead to big, heavy, energy wasteful, expensive designs that are costly to manufacture as they require careful component matching across many parts.
Another solution is to use feedback with proper care. This tends to lead to small circuits with fewer components, relatively low power wastage, cheaper to buy and cheaper to manufacture because there are less parts.
Whether feedback is used or not don't lose sight of the fact that it is the non-linearities of the components that ultimately determines the sound.
A perfectly linear transistor doesn't exist. This is why they are used so as the Q-point doesn't move very much. This gives them more linear operation, but still needs some quantity of error correction. For output devices the Q-point must move in order to have efficiency so non-linearity arises that must be corrected. Otherwise, you might as well go with good old vacuum tubes. A non-feedback transistor amp must be designed around the specific devices used, and are hard to duplicate for a production line and cheap mass production.
I agree with Traderbam.
There is no free lunch.
Feedback trades even order harmonics (non linearity) for odd harmonics, which are distributed over the higher frequencies.
The types of distortion that remain after feedback (the odd higher order ones) are very objectionable. I suppose the question is, are these distortion components still audiable?
I think final quality has to depend on the initial design, and how much work the feedback is expected to do.
The effects of feedback are very noticable in musical instrument/studio applications, because things do clip very often, as the source signals have wide dynamic range.
HiFi amps should be able to avoid this type of input clipping, but the problem still remains- when you use feedback to correct for non linearity, you do create HF distortion.
I don't hate feedback, but I do hate high odd order distortion, no matter how well the amp measures
There is no free lunch.
Feedback trades even order harmonics (non linearity) for odd harmonics, which are distributed over the higher frequencies.
The types of distortion that remain after feedback (the odd higher order ones) are very objectionable. I suppose the question is, are these distortion components still audiable?
I think final quality has to depend on the initial design, and how much work the feedback is expected to do.
The effects of feedback are very noticable in musical instrument/studio applications, because things do clip very often, as the source signals have wide dynamic range.
HiFi amps should be able to avoid this type of input clipping, but the problem still remains- when you use feedback to correct for non linearity, you do create HF distortion.
I don't hate feedback, but I do hate high odd order distortion, no matter how well the amp measures
wildswan said:
This is why if you are interested in sound quality to allow lots of headroom for the actual average power you intend to listen to. Music isn't a single sine wave but has harmonics that produce spikes and peaks that can be way above the average power output. This is particularly true of clasical music. Of course most people who listen to clasical tunes don't listen at 1000Wrms.
ash_dac said:
Yes, in a nutshell. Transistors (and valves) are inherently non-linear. There's no getting around it. You can operate them around a ridiculously small range, to minimise the non-linearities, or else you can utilise feedback to linearise them.
Of course feedback doesn't always have to be global. each stage can be linearised a bit using emitter degeneration resistors as feedback elements, but the general principles always apply. There's no such thing as a free lunch.
Cheers,
Suzy
Or you can create a linear component by combining two conjugate non-linear components, or some combination of multiple components.
traderbam said:
Or you can create a linear component by combining two conjugate non-linear components, or some combination of multiple components.
Indeed. Feedforward has the added benefit of being inherently stable, as well.
Cheers,
Suzy
Re: Re: Feedback in amplifier is ultimately the driver of sound quality ?
Nicely said.
Besides from feedback application as we know it from Harry Black, there exist also feedforward error correction popularised by P.J. Walker and feedback error correction by M.O.J. Hawksford.
Nicely said.
Besides from feedback application as we know it from Harry Black, there exist also feedforward error correction popularised by P.J. Walker and feedback error correction by M.O.J. Hawksford.
Feedforward is a tricky challenge for audio power amps because the load, the speaker, can have wildly varying impedance. The amp is designed with a low output Z to control the speaker voltage. A feedforward system would have to drive both the speaker and the amp to correct the voltage. So you end up needing a feedforward amp that is powerful enough to do this ...catch 22.
If, say, the speaker was a resistor, then the main amp could be a transconductance amp, supplying the bulk of the power and the feedforward amp would supply the low Z and would need much less power to correct the voltage and would not be "fighting" the main amp. But speakers are not normally like this. You might be able to make a really good headphone amp this way.
If, say, the speaker was a resistor, then the main amp could be a transconductance amp, supplying the bulk of the power and the feedforward amp would supply the low Z and would need much less power to correct the voltage and would not be "fighting" the main amp. But speakers are not normally like this. You might be able to make a really good headphone amp this way.
I still don't understand of what is called feedforward. In Quad 405, which is the feedforward? I see current dumping, but cannot see the feedforward. The 47pf from VAS collector to inverting-input seems to reduce the ability of differential pair to fix the output stage for higher frequencies (in trade for stability) instead of helping the output stage to be more linear in higher frequencies.
http://www.quadesl.org/Amplifiers/amplifiers.html
The Quad is a great example, thanks for raising it. Basically, the 405 dealt with the problem by splitting the main amp and feedforward amp by frequency. As I understand it "current dumping" consisted of putting a series inductor on the output of the high power class AB amp (a low pass current filter) and then using a higher precision, low power amp to provide faster current. The inductor raised the output Z of the main amp as frequency increased and the FF amp took increasing control. Again, without the inductor the amps would fight one another.
This is a clever idea but relies on the bulk of the current/power demand occuring at low frequencies. At high frequencies the Quad was very power-contrained. When they launched the produt they had to be specific about what sort of speaker impedance characteristic was acceptable...there were problems with some speakers. The Quad sounded very clean but lacked the slam and treble power we expect today. The technology was sharp for the early 70's when fast, reliable power transistors were not available like they are today.
The Quad is a great example, thanks for raising it. Basically, the 405 dealt with the problem by splitting the main amp and feedforward amp by frequency. As I understand it "current dumping" consisted of putting a series inductor on the output of the high power class AB amp (a low pass current filter) and then using a higher precision, low power amp to provide faster current. The inductor raised the output Z of the main amp as frequency increased and the FF amp took increasing control. Again, without the inductor the amps would fight one another.
This is a clever idea but relies on the bulk of the current/power demand occuring at low frequencies. At high frequencies the Quad was very power-contrained. When they launched the produt they had to be specific about what sort of speaker impedance characteristic was acceptable...there were problems with some speakers. The Quad sounded very clean but lacked the slam and treble power we expect today. The technology was sharp for the early 70's when fast, reliable power transistors were not available like they are today.
wildswan said:I agree with Traderbam.
There is no free lunch.
Feedback trades even order harmonics (non linearity) for odd harmonics, which are distributed over the higher frequencies.
............
I think final quality has to depend on the initial design, and how much work the feedback is expected to do.
I don't hate feedback, but I do hate high odd order distortion, no matter how well the amp measures
Greetings from Norfolk (UK)
Where does the idea that feedback 'trades' even order harmonics for odd order harmonics. Non linearity can show its self as both odd and even order harmonics, and properly applied feedback will reduce all amplifier non linearities, (odd or even).
A properly designed amplifier will start from a circuit exhibiting minimal distortion, and then have feedback applied which will then significantly reduce this distortion. This is part of the art of amplifier design.
Feedback is the designer's (and the user's) friend, not his enemy, but like all good things must be used correctly
Richard
I imagine John Curl has many memories of the birth of the 405.
Let's consider this feed-forward issue some more. If you accept that it is impractical to correct the amplifier at its output because of the low impedance requirement across the whole audio band, then you might decide that the next best place would be to inject a correction signal at the input of the output stage. This is still feed-forward, just to a point one or two devices before the output.
Suppose you have a perfect op-amp and use this to measure the input output error and feed a correction signal to the output stage. You have to bandwidth limit your op-amp to avoid oscillation because the output stage has some phase shift. Ok, you try this and it seems to make an improvement.
Then you might notice that your feed-forward op-amp is working in parallel with the LTP/VAS stage of the amp and you suddenly realise that the original input stage is not really doing anything useful - you perfect op-amp, with a little gain adjustment, can do both the input signal amplification AND the feedforward correction at the same time, better than the original circuit. You remove the original and just use the op-amp. It sounds even better now. But your feedforward system seems to have suddenly become a proportional-integral feedback system...😱
Congratulations, you have just invented the topology of the Naim NAP250. Of course you are over 30 years behind Naim. Actually, this approach was originally published in an RCA databook. It is of course just a very simple op-amp topology.
Sorry, this was rather a swift bit of lateral thinking. Have a pause for thought at this point.
So my question to the deep thinkers out there is which approach is best - feeding a correction to the output stage output or feeding a correction to the output stage input?
Let's consider this feed-forward issue some more. If you accept that it is impractical to correct the amplifier at its output because of the low impedance requirement across the whole audio band, then you might decide that the next best place would be to inject a correction signal at the input of the output stage. This is still feed-forward, just to a point one or two devices before the output.
Suppose you have a perfect op-amp and use this to measure the input output error and feed a correction signal to the output stage. You have to bandwidth limit your op-amp to avoid oscillation because the output stage has some phase shift. Ok, you try this and it seems to make an improvement.
Then you might notice that your feed-forward op-amp is working in parallel with the LTP/VAS stage of the amp and you suddenly realise that the original input stage is not really doing anything useful - you perfect op-amp, with a little gain adjustment, can do both the input signal amplification AND the feedforward correction at the same time, better than the original circuit. You remove the original and just use the op-amp. It sounds even better now. But your feedforward system seems to have suddenly become a proportional-integral feedback system...😱
Congratulations, you have just invented the topology of the Naim NAP250. Of course you are over 30 years behind Naim. Actually, this approach was originally published in an RCA databook. It is of course just a very simple op-amp topology.
Sorry, this was rather a swift bit of lateral thinking. Have a pause for thought at this point.
So my question to the deep thinkers out there is which approach is best - feeding a correction to the output stage output or feeding a correction to the output stage input?
Hi, Traderbam,
Thanks for the hint. All this time I see ampA as the main amp and ampB as the followeramp. If it is tilted, ampA as the follower amp and ampB as the mainamp, it makes sense 😀
Thanks for the hint. All this time I see ampA as the main amp and ampB as the followeramp. If it is tilted, ampA as the follower amp and ampB as the mainamp, it makes sense 😀
You need to use a larger input signal to the to get the same output when you use feedback which is sure to cause more odd order distortion than the open loop case.
Btw I didn't mean to suggest that feedback affects only even harmonics. It's just that the reduction in odd harmonics results from having a smaller output signal, and this could be achieved by attenuating the signal.
Feedback does not operate on linear devices anyway and the feedback signal is not just an amplified version of the input signal, which actually increases distortion at lower feedback ratios. Baxandall suggested this in 1978 I believe.
Unless the input section of your amplifier is perfectly linear, the signals will be partially multiplied and produce higher order distortion components. The solution- more feedback. And then there's TIM...
Now it really does sound like I hate feedback😀
wildswan said:
Feedback does not operate on linear devices anyway and the feedback signal is not just an amplified version of the input signal, which actually increases distortion at lower feedback ratios. Baxandall suggested this in 1978 I believe.
Unless the input section of your amplifier is perfectly linear, the signals will be partially multiplied and produce higher order distortion components. The solution- more feedback. And then there's TIM...
Now it really does sound like I hate feedback😀
Thankyou everyone for your replies.
I think what you are saying is that the distortion at the output stage, and thus correction signal is not equal to the distortion at the input stage.
But if the amplifier was ideal (perfectly linear) there would be no feedback signal by definition. So as already mentioned feedback compensates for the non-linearity of components used in the amplifier.
Also I understand that it takes a finite time for the feedback loop to operate so corrections occur after the initial event but never before.
So in conclusion component parts should be selected so as to minimise non-linearities in the amplifier, and thereby reduce the amount of feedback necessary in the system.
...the input stages is paramount to good power amplifier design
All designs I have looked at use a long tail pair differential input. Is there anything better ?
- Status
- Not open for further replies.