Dan Meyer?
In any case, it's pretty easy to get the amp-induced harmonics an order of magnitude lower than the audibility numbers being bandied about. Topology is an important factor, but as PRR has pointed out, it's just one factor. If we're designing for ultra-low distortion figures, well, all of this becomes important. If we're doing something easier, like merely designing a box of gain that cannot be detected by ear, we are free to forget about the mice and concentrate on the elephants marching around in the room.
.
I highly agree with this statement.
John,
can you make this formula a little bit clearer to me? It looks nice, but I don't know where to start. Is second harmonic n=2? And what do you think is the max. amount of k2?
In my opinion there are several other important things.
The following statements are just an assumption:
1. The higher the output voltage the higher the distortion may be
2. Distortion should be independant from the load
3. Reversed error spectrum (higher output voltage/power) with decresed distortion is bad. This is typical for many class AB-amps.
4. The higher the output voltage the more higher harmonics may be present in the spectrum
5. Distortion should be independant from frequency
6. Distortion spectrum should not vary when two or more signals appear
7. With coupled rf-noise the the input distortion spectrum and levels should not vary
It's just my opinon but I think this therory is more complex than only viewin' to one (technical and therefore simplified!) signal
I highly agree with this statement.
John,
can you make this formula a little bit clearer to me? It looks nice, but I don't know where to start. Is second harmonic n=2? And what do you think is the max. amount of k2?
In my opinion there are several other important things.
The following statements are just an assumption:
1. The higher the output voltage the higher the distortion may be
2. Distortion should be independant from the load
3. Reversed error spectrum (higher output voltage/power) with decresed distortion is bad. This is typical for many class AB-amps.
4. The higher the output voltage the more higher harmonics may be present in the spectrum
5. Distortion should be independant from frequency
6. Distortion spectrum should not vary when two or more signals appear
7. With coupled rf-noise the the input distortion spectrum and levels should not vary
It's just my opinon but I think this therory is more complex than only viewin' to one (technical and therefore simplified!) signal
I think NP makes a good point in one of his papers for the viability of the single ended approach. (He's speaking of output stages, but the same should hold true throughout the chain):
Sound is single ended I.E. we don't live at 0 p.s.i. The natural state of an acoustic wave is to vary around a positive value rather than to be bipolar...........mike
Sound is single ended I.E. we don't live at 0 p.s.i. The natural state of an acoustic wave is to vary around a positive value rather than to be bipolar...........mike
The problem is we cannot build a distortion-free amp (with only one exception...)
In this way we should keep the audible distortion level below the hearing threshold level. Unfortunaltely no one know exactly what this means. So we use static signals in order to map the "audible quality" to measurable signals...
In this way we should keep the audible distortion level below the hearing threshold level. Unfortunaltely no one know exactly what this means. So we use static signals in order to map the "audible quality" to measurable signals...
Seems a good way. Maybe we can use a (calibrated) differential/instrumentation amp (fed by input and output of the power amp), sample the output of the diff-amp, do a FFT (or better continiously FFTs) over the spectrum and normalise it. Could be very interesting. Maybe the spectrum varies from FFT to FFT. If it does it could be a first clue due to different sound by several amps.
Maybe more tomorow. It's late in Europe
Maybe more tomorow. It's late in Europe
Trust me, it is difficult to get anything except a tube or class A FET amp to have an extremely low level of higher order harmonics, especially open loop and over extended frequency.
The reasons are:
Transistors are pretty darn nonlinear, and they have several different distortion producing components. These include: very non-linear Gm (voltage gain), non-linear BETA, and non-linear input capacitance (changes with voltage level on both the collector and base, referred to the emitter).
When you TRY to linearize them with local feedback (series resistor) you convert the even order harmonics into higher order odd harmonics.
If you try to use loop feedback, then you get TIM or FM modulation distortion, i.e. FIM from modulating the open loop bandwidth with amplitude changes with signal level. This is just with class A, Class AB or B is much worse.
It is a difficult problem. This is why we have developed sophisticated topologies in order to minimize the generation of distortion, over the decades.
In any case, the generation of higher order harmonics are not a good idea.
The reasons are:
Transistors are pretty darn nonlinear, and they have several different distortion producing components. These include: very non-linear Gm (voltage gain), non-linear BETA, and non-linear input capacitance (changes with voltage level on both the collector and base, referred to the emitter).
When you TRY to linearize them with local feedback (series resistor) you convert the even order harmonics into higher order odd harmonics.
If you try to use loop feedback, then you get TIM or FM modulation distortion, i.e. FIM from modulating the open loop bandwidth with amplitude changes with signal level. This is just with class A, Class AB or B is much worse.
It is a difficult problem. This is why we have developed sophisticated topologies in order to minimize the generation of distortion, over the decades.
In any case, the generation of higher order harmonics are not a good idea.
I agree with using music as the ultimate test. It's woolly, in that it's not objective, but since assessment of amp quality is often made en masse in the marketplace by consumers (who are usually less interested in the technology than the 'sound'), it seems logical to design with simple topologies then 'voice' with careful listening tests over large samples to achieve the goal.
You could argue that intense, spec driven engineering has created amps of vanishingly low distortion, but little apparent correlation with 'acceptable' sonics, at least from the point of view of listener (read: consumer) popularity.
PRR's point about a steadily decreasing harmonic structure, maybe around 12dB/octave, sits very well with me. It might also imply that distortion profiles increasing with frequency are not too important as all the harmonics beyond about 10KHz fundamental are inaudible anyway. Further, any harmonic content less than about 80dB below is probably inaudible in the average urban sitting room. I've enjoyed some success with very simple topologies - even including boostrapping the VAS load - by careful attention to layout, diff pair balance, component choice and dimensioning. Speed of the VAS is important, the dead zone at crossover is critical, and lag compensation is crucial.
The very controversy of just why certain designs sound as they do would indicate design detail and empirical refinement are the major factors, just as PRR opines, and music is the real test.
And Fred, I'll be most interested in the results with your 'oldie but goody'. Keep us posted!
Cheers,
Hugh
You could argue that intense, spec driven engineering has created amps of vanishingly low distortion, but little apparent correlation with 'acceptable' sonics, at least from the point of view of listener (read: consumer) popularity.
PRR's point about a steadily decreasing harmonic structure, maybe around 12dB/octave, sits very well with me. It might also imply that distortion profiles increasing with frequency are not too important as all the harmonics beyond about 10KHz fundamental are inaudible anyway. Further, any harmonic content less than about 80dB below is probably inaudible in the average urban sitting room. I've enjoyed some success with very simple topologies - even including boostrapping the VAS load - by careful attention to layout, diff pair balance, component choice and dimensioning. Speed of the VAS is important, the dead zone at crossover is critical, and lag compensation is crucial.
The very controversy of just why certain designs sound as they do would indicate design detail and empirical refinement are the major factors, just as PRR opines, and music is the real test.
And Fred, I'll be most interested in the results with your 'oldie but goody'. Keep us posted!
Cheers,
Hugh
I'm a bit surprised that no one has mentioned one of the
advantages of single-pair inputs biased by a constant
current source and driving an SE 2nd stage which is loaded
with a constant current source - namely the bias stability that
such arrangements provide. It is easy to make a stable CCS,
and this stabilizes the operation of the gain devices.
Usually the output stage bias is derived from the 2nd stage
DC bias, so further down the chain we also see more output
stage stability as a result.
I'm not necessarily advocating single vs dual diff pairs, as I
use either when I feel like it, but having a convenient CCS
by which you anchor the circuit helps explain one of the virtues
of some earlier solid state amplifiers. I bet a few more Tigers or
Zillas would still be working today if they had taken that
approach.
advantages of single-pair inputs biased by a constant
current source and driving an SE 2nd stage which is loaded
with a constant current source - namely the bias stability that
such arrangements provide. It is easy to make a stable CCS,
and this stabilizes the operation of the gain devices.
Usually the output stage bias is derived from the 2nd stage
DC bias, so further down the chain we also see more output
stage stability as a result.
I'm not necessarily advocating single vs dual diff pairs, as I
use either when I feel like it, but having a convenient CCS
by which you anchor the circuit helps explain one of the virtues
of some earlier solid state amplifiers. I bet a few more Tigers or
Zillas would still be working today if they had taken that
approach.
I agree that higher order harmonics should be avoided, even if they are very very low in magnitude. The other important issue is an intermodulation with HF signals and spikes, like D/A conversion residuals at the output of CD players. This is perfectly audible and can be quite effectively minimized by filtration and bandwidth reduction to a reasonable limit, say some 100 kHz.
Tigers and Zillas
You mean Ampzillas?
Mine has constant current sources for the input differential pairs.
It's still working after 28 years!
Hi Nelson,Nelson Pass said:I'm a bit surprised that no one has mentioned one of the
advantages of single-pair inputs biased by a constant
current source and driving an SE 2nd stage which is loaded
with a constant current source - namely the bias stability that
such arrangements provide. It is easy to make a stable CCS,
and this stabilizes the operation of the gain devices.
Usually the output stage bias is derived from the 2nd stage
DC bias, so further down the chain we also see more output
stage stability as a result.
I'm not necessarily advocating single vs dual diff pairs, as I
use either when I feel like it, but having a convenient CCS
by which you anchor the circuit helps explain one of the virtues
of some earlier solid state amplifiers. I bet a few more Tigers or
Zillas would still be working today if they had taken that
approach.
You mean Ampzillas?
Mine has constant current sources for the input differential pairs.
It's still working after 28 years!
I highly interesting thread...
John, it's interesting that your results are nearly the same as mine. Distortion is mainly produced by the output stage when VAS and diff-pair is designed as Doug Self would call blameless. Most distortion comes nearly from crossover when leaving class A and moving into class AB. This introduces high order harmonics. In this case a class AB produces at least 10 or 20 times the distortion of a well designed class A stage.
Worsed in this case is a CFP, followed by a bipolar EF in class AB. Better is a MOSFET in class A and best the bipolar class A because of the higher transconductance of the bipolars. Unfortunately this is only true when a bipolar is driven by a low impedance. Obviously a VAS stage has a high output and typical bipolar stage a very non-linear input impedance witch is the main mechanism of crossover distortion in a bipolar output stage.
The only way to come out of this problem is to implement an additional "driver-stage" to convert the high impedance from the VAS to a low-impedance stage which drives the output stage. This can be done with a simple emitter follower. Or by a tripplet as output. Or something more sophisticated like Nelson does
Has anybody measures what happens to the distortion spectrum when introduced an additional signal? If we have an amp witch has a distortion spectrum at 600Hz like A=(n-1)!/2 or which decreases with 12dB/octave, will the spectrum vary if we have another signal maybe of 7kHz? I do not mean something like intermodulation, the first order intermodulation products should be masked by the ear. Will an amp in presence of two signals also have the same spectrum of A=(n-1)!/2?
John, it's interesting that your results are nearly the same as mine. Distortion is mainly produced by the output stage when VAS and diff-pair is designed as Doug Self would call blameless. Most distortion comes nearly from crossover when leaving class A and moving into class AB. This introduces high order harmonics. In this case a class AB produces at least 10 or 20 times the distortion of a well designed class A stage.
Worsed in this case is a CFP, followed by a bipolar EF in class AB. Better is a MOSFET in class A and best the bipolar class A because of the higher transconductance of the bipolars. Unfortunately this is only true when a bipolar is driven by a low impedance. Obviously a VAS stage has a high output and typical bipolar stage a very non-linear input impedance witch is the main mechanism of crossover distortion in a bipolar output stage.
The only way to come out of this problem is to implement an additional "driver-stage" to convert the high impedance from the VAS to a low-impedance stage which drives the output stage. This can be done with a simple emitter follower. Or by a tripplet as output. Or something more sophisticated like Nelson does
Has anybody measures what happens to the distortion spectrum when introduced an additional signal? If we have an amp witch has a distortion spectrum at 600Hz like A=(n-1)!/2 or which decreases with 12dB/octave, will the spectrum vary if we have another signal maybe of 7kHz? I do not mean something like intermodulation, the first order intermodulation products should be masked by the ear. Will an amp in presence of two signals also have the same spectrum of A=(n-1)!/2?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.