Are you assuming that the output of the right-hand amp does not affect the output of the left hand amp?
Suppose the amp output voltages are Vl and Vr respectively, and say the correct output is Vo.
You are saying that you make Vr = (Vl - Vo) so that the voltage across the speaker is Vl - Vr = Vl - (Vl - Vo) = Vo. Perfect!
This cancellation assumes Vl is not changed by Vr. But you must agree the current through the speaker changes with Vr. So if Vl remains the same when the speaker current changes the output Z of the left amp must be zero. Which is impossible. So what happens is Vr adjusts Vl slightly and this causes a small change in Vr and this causes Vl to change and so on...and if you are lucky it converges.
Feedback is born!
I think what most people do is to try to make the left side have the least amount of distortion (or at least an agreeable distortion shape--meaning I know that people try to minimize "high order" distortion, which your design may be more prone to, but I don't know...). Then people figure that something like the right side will introduce more distortion than it cancels.
Another thing, it doesn't seem like a lot of people have actually tried this feedforward design, so perhaps you will learn something interesting (that the rest of us don't know about). This probably leads to other interesting discoveries.
JF
Another thing, it doesn't seem like a lot of people have actually tried this feedforward design, so perhaps you will learn something interesting (that the rest of us don't know about). This probably leads to other interesting discoveries.
JF
Lighter than air and faster than light...
My first observation is that the proposed topology attempts distortion cancellation by duplicating the original amplifier as a canceller and feeding it with an error signal, the cancellation occurs because the canceller is replacing ground at the load. This isn't an original concept, but its fun to work with.
The simulated THD numbers I've seen posted here won't hold up in practice, and wouldn't prove useful if they did...
1) The op-amp macro-models the Spice based simulator is using don't attempt to model distortion, and may not be simulating the internal delay times and slew rates inside the op-amp. Without the internal delays, many complex control loop topologies simulate as working well, when in fact they will oscillate. Switching to an accurate transistor based model will get both the delay and distortion. If one isn't available for a particular op-amp, choose one a full circuit transistor based model is available for. These models often have slow convergence because of the high internal gain and feedback, and if the overall circuit is unstable there may be a problem simulating it. Use a transient analysis to ramp up the DC supplies without a signal and, assuming it reaches a steady state, use the resulting node voltages to preset the bias point for the simulator for the other analyses.
2) The circuit is exploiting a level of symmetry only found in the simulator. When built, the parts will vary. Not just the resistor tolerances, the transistors, op-amps, and so on. It is important to determine the key sensitivities and their effect or do some monte-carlo simulations. This is a limiting challenge for a feedforward design that makes it difficult to create a composite device that dramatically improves on the performance of the error amplifier. As you refine your design you need limit the number of parts that need to be matched pairs, and how well matched they have to be to achieve some realistic goal for distortion. A good first step would be to determine what matching is needed to achieve a composite amplifier 10dB better than the error amplifier.
3) Non-linear effects that aren't being modelled will prevent the design from achieving numbers like -156dBc. For example; capacitors have a voltage coefficient, you can find specs for various types and its a parameter for the spice model. There is also an effect called "dielectric absorbtion" which is more difficult to get the numbers for and more complex to add to the circuit model. Dissimilar metal junctions, e.g a nickel plated binding post on a copper wire will also cause distortion. A copper oxide junction can even be a useful power rectifier if it grows thick enough. Cables, connectors, enclosures, etc. will exhibit small paramagnetic effects and create very low level harmonic distortion. There references for some common materials.
4) THD isn't the beginning and the end of distortion. Phase distortion is also important, and has a strong effect on the listeners perception of the sound (the psychoacoustics). A constant group delay is desirable, but may require additional complexity to achieve.
5) The distortion from whatever speaker is used and the room's acoustics overshadow the result of a competent audio amplifier design. Closing the feedback path beyond the transducer either by adding a sense coil to your speaker voice coil or a microphone both pose error signal quality and stability problems. Using a DSP to calibrate the feeedback transducer and then predistort the signal according to an adaptive algorithm evaluating realtime error signal samples could solve those problems and even overcome the room acoustics in at least a limited volume of the room. The design complexity is high, but some good algorithms for readily available hardware would be a considerable achievement.
My first observation is that the proposed topology attempts distortion cancellation by duplicating the original amplifier as a canceller and feeding it with an error signal, the cancellation occurs because the canceller is replacing ground at the load. This isn't an original concept, but its fun to work with.
The simulated THD numbers I've seen posted here won't hold up in practice, and wouldn't prove useful if they did...
1) The op-amp macro-models the Spice based simulator is using don't attempt to model distortion, and may not be simulating the internal delay times and slew rates inside the op-amp. Without the internal delays, many complex control loop topologies simulate as working well, when in fact they will oscillate. Switching to an accurate transistor based model will get both the delay and distortion. If one isn't available for a particular op-amp, choose one a full circuit transistor based model is available for. These models often have slow convergence because of the high internal gain and feedback, and if the overall circuit is unstable there may be a problem simulating it. Use a transient analysis to ramp up the DC supplies without a signal and, assuming it reaches a steady state, use the resulting node voltages to preset the bias point for the simulator for the other analyses.
2) The circuit is exploiting a level of symmetry only found in the simulator. When built, the parts will vary. Not just the resistor tolerances, the transistors, op-amps, and so on. It is important to determine the key sensitivities and their effect or do some monte-carlo simulations. This is a limiting challenge for a feedforward design that makes it difficult to create a composite device that dramatically improves on the performance of the error amplifier. As you refine your design you need limit the number of parts that need to be matched pairs, and how well matched they have to be to achieve some realistic goal for distortion. A good first step would be to determine what matching is needed to achieve a composite amplifier 10dB better than the error amplifier.
3) Non-linear effects that aren't being modelled will prevent the design from achieving numbers like -156dBc. For example; capacitors have a voltage coefficient, you can find specs for various types and its a parameter for the spice model. There is also an effect called "dielectric absorbtion" which is more difficult to get the numbers for and more complex to add to the circuit model. Dissimilar metal junctions, e.g a nickel plated binding post on a copper wire will also cause distortion. A copper oxide junction can even be a useful power rectifier if it grows thick enough. Cables, connectors, enclosures, etc. will exhibit small paramagnetic effects and create very low level harmonic distortion. There references for some common materials.
4) THD isn't the beginning and the end of distortion. Phase distortion is also important, and has a strong effect on the listeners perception of the sound (the psychoacoustics). A constant group delay is desirable, but may require additional complexity to achieve.
5) The distortion from whatever speaker is used and the room's acoustics overshadow the result of a competent audio amplifier design. Closing the feedback path beyond the transducer either by adding a sense coil to your speaker voice coil or a microphone both pose error signal quality and stability problems. Using a DSP to calibrate the feeedback transducer and then predistort the signal according to an adaptive algorithm evaluating realtime error signal samples could solve those problems and even overcome the room acoustics in at least a limited volume of the room. The design complexity is high, but some good algorithms for readily available hardware would be a considerable achievement.
Re: Lighter than air and faster than light...
Why do tubes sound good, though they are fully magnetic...
I know of all that sim problems, but I believe it is useful to test for warmer and colder.
If I have a circuit and get 120dB and after adding an error amp I get 150dB, that makes me believe it is the right direction.
Soon I will build it and we will see...
polychrome said:
Why do tubes sound good, though they are fully magnetic...
I know of all that sim problems, but I believe it is useful to test for warmer and colder.
If I have a circuit and get 120dB and after adding an error amp I get 150dB, that makes me believe it is the right direction.
Soon I will build it and we will see...
That reminds me on my tweeters*
I disagree with this. In my experience an amp design that is commonly called "competent" contains types of distortion that are extremely unnatural and thus more noticable/intrusive than those of "competent" speakers and certainly of room effects. After you've spent $5000 on your CD source your next highest priority is the amp.
a possible explanation of the distortion at a single frequency is that the cancellation scheme does work - at a single frequency where the op amp and ouput phase shifts "line up", generally such schemes have a single sharp OL gain peak at the "zero sensitivity" frequency
so a big question is wether the claimed distortion numbers hold up with varying frequencies, say 1, 2, 5, 10 KHz
because phase shift is inherent in any real circuit it is begging the question to include any zero phase shift "idealized" difference amp - as you can derive infinite gain from such a zero pahse shift gain block
so a big question is wether the claimed distortion numbers hold up with varying frequencies, say 1, 2, 5, 10 KHz
because phase shift is inherent in any real circuit it is begging the question to include any zero phase shift "idealized" difference amp - as you can derive infinite gain from such a zero pahse shift gain block
,
FEEDBACK
is when you apply a correction signal to the circuit related to the difference between target output and actual output.
FEED-FORWARD
is when you apply a correction signal to the output related to the target output and a model of the circuit's error function and the load being driven.
PRE-DISTORTION
is like feed-forward except that you apply the correction signal at the input, added to the input signal.
ADAPTIVE PRE-DISTORTION
is where the model of the error function is adjusted using feedback (typically much slower than the active signal bandwidth) to minimize output error. This is used in telecommunications.
is when you apply a correction signal to the circuit related to the difference between target output and actual output.
FEED-FORWARD
is when you apply a correction signal to the output related to the target output and a model of the circuit's error function and the load being driven.
PRE-DISTORTION
is like feed-forward except that you apply the correction signal at the input, added to the input signal.
ADAPTIVE PRE-DISTORTION
is where the model of the error function is adjusted using feedback (typically much slower than the active signal bandwidth) to minimize output error. This is used in telecommunications.
Towards -200dB distortion?
Being an armchair amp designers myself, I pretty much symphatize with Bernhard's quest for perfection by finding *the* topologoy.
Even considering, that these simulations cannot be materialized in real circuits, as was rightly argumented several times, it is an intellectual challenge in itself.
But let's have a reality check with human hearing: Do you really think, you can hear something which is 200dB below another sound? Either the distortion is below any sensible hearing threshold or the signal will destroy the ears.
To pick a number, I would suggest, that 130dB will be enough margin between distortion and signal, in any case.
Regards,
Peter Jacobi
Being an armchair amp designers myself, I pretty much symphatize with Bernhard's quest for perfection by finding *the* topologoy.
Even considering, that these simulations cannot be materialized in real circuits, as was rightly argumented several times, it is an intellectual challenge in itself.
But let's have a reality check with human hearing: Do you really think, you can hear something which is 200dB below another sound? Either the distortion is below any sensible hearing threshold or the signal will destroy the ears.
To pick a number, I would suggest, that 130dB will be enough margin between distortion and signal, in any case.
Regards,
Peter Jacobi
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.