Re: The character of thermal distorsion
). Audio-chain behaving at infra-low frequencies does influence to our ears - recall detonations of old sound sources. Another example is an audibility of servo-chain tracking an output amplifier drift.
Global NFB reduces MD, so I care of those schematics parts which are beyond NFB loop.
We must measure THD (or, which is better, measure a spectrum) from rather low frequencies (almost from DCChrister said:
). Audio-chain behaving at infra-low frequencies does influence to our ears - recall detonations of old sound sources. Another example is an audibility of servo-chain tracking an output amplifier drift.Global NFB reduces MD, so I care of those schematics parts which are beyond NFB loop.
Re: Re: The character of thermal distorsion
But still, if my reasoning is correct, the only mentionable distorsion arising from this phenomenon would be correlated to the signal itself, manifesting itself as fundamental and 2nd harmonic. However, if what you mean is that it is not sufficient to measure THD at say 1 kHz and 20 kHz, then I agree. Thermal distorsion could be worse at low frequencies than high frequencies, so we have to sample frequencies throughout the audio bandwith at least. So yes, it could go undected in the usual THD measurements, but it would still behave as THD.
anli said:
We must measure THD (or, which is better, measure a spectrum) from rather low frequencies (almost from DC
But still, if my reasoning is correct, the only mentionable distorsion arising from this phenomenon would be correlated to the signal itself, manifesting itself as fundamental and 2nd harmonic. However, if what you mean is that it is not sufficient to measure THD at say 1 kHz and 20 kHz, then I agree. Thermal distorsion could be worse at low frequencies than high frequencies, so we have to sample frequencies throughout the audio bandwith at least. So yes, it could go undected in the usual THD measurements, but it would still behave as THD.
Christer,
To feel more, imagine such case. Very low level piano reverberations "sit on" high level contrabass note. Last one make MD, MD modulate piano reverberations. Probably, this is (one of) a reason of situations when transistor amplifiers are treated to be "sterile": common THD measurements give us 1e-100% value, but an amplifier "doesn't sing"
To feel more, imagine such case. Very low level piano reverberations "sit on" high level contrabass note. Last one make MD, MD modulate piano reverberations. Probably, this is (one of) a reason of situations when transistor amplifiers are treated to be "sterile": common THD measurements give us 1e-100% value, but an amplifier "doesn't sing"
anli said:Christer,
To feel more, imagine such case. Very low level piano reverberations "sit on" high level contrabass note. Last one make MD, MD modulate piano reverberations. Probably, this is (one of) a reason of situations when transistor amplifiers are treated to be "sterile": common THD measurements give us 1e-100% value, but an amplifier "doesn't sing"
Yes, but isn't that just IM between different frequencis in the source? MD itself just produce HD for each frequency and then these modulate each other. If this is not correct, I think there must be some flaw in the reasoning I presented earlier, but what?
I'm not deep theorist (just am angry reading published schematics, and, as a result, design-solder-listen to own ), but probably almost any distortions of kind A may be presented as "special case" of distortions of kind B. You see, I'm pragmatic here - see issue, estimate it's influence to musical illusions, eliminate it.
), but probably almost any distortions of kind A may be presented as "special case" of distortions of kind B. You see, I'm pragmatic here - see issue, estimate it's influence to musical illusions, eliminate it.Well, the point of what I posted earlier was that it is useful to know if MD is something we fail to measure with the usual methods or if it is included in the distorsion we measure. I thus thought that trying to understand theoretically what the character of MD would be might help to answer the first question. My conclusion was that it should behave like a form of harmonic distorsion and thus be difficult to distinguish from ordinary HD, but that would also mean that we do measure it, we just don't measure only MD separately. However, even if nobody finds any obvious error in my reasoning, I think it would still be useful to do the experiment with a dual transistor that I suggested. That should provide a way to estimate MD more directly and separately. If such measurements agree with my proposed theory I would be happy , but it would also tell us that this is probably the only way to measure if MD is a problem or not in practice. If the measurements clearly disagrees, there is either something wrong with the measurements, or my theory is incorrect.
, but it would also tell us that this is probably the only way to measure if MD is a problem or not in practice. If the measurements clearly disagrees, there is either something wrong with the measurements, or my theory is incorrect. 
> But still, if my reasoning is correct, the only mentionable distorsion arising from this phenomenon would be correlated to the signal itself, manifesting itself as fundamental and 2nd harmonic.
Actually, Perrot's theory (which I tend to agree with) says that :
- Humans can adjust to and compensate low-order distortion provided the distortion characteristics are constant. For instance, tube gear with some second-order distortion can sound very good.
By "constant low-order distortion", I mean : Vout = f( Vin ) with f a second-order polynomial whose coefficients don't change with time.
- However this adjustment mechanism is upset if the distortion profile changes with time : then we can't adjust to it.
This would be : Vout = f( Vin, time ) where the coefficients of f change with time.
It's the same for frequency response : we can get used to non-flat frequency responses (no speaker is 100% flat) but when we change speaker, we tend to first hear the differences between what we're used to and the new one.
So, temperature variations in transistors are a bandpass function of the dissipated power, hence we get :
Vout = f( Vin, and also the history of Vin in the past few seconds )
For instance, take a power amp with a perfectly balanced input stage and run a 1 kHz sine through it (like a piano note decaying). You'll get very low distortion. Now, add a powerful transient (like the piano attack).
The worst case is if the transient clips the amp : in this case for a few ms, the input stage will have very different dissipated powers. Until they come back to the same temperature, the input stage will be imbalanced, and the distortion spectrum of the amp will change.
So, if you could measure an instantaneous distortion spectrum, you'd get different ones depending on the instant you measure.
Actually, Perrot's theory (which I tend to agree with) says that :
- Humans can adjust to and compensate low-order distortion provided the distortion characteristics are constant. For instance, tube gear with some second-order distortion can sound very good.
By "constant low-order distortion", I mean : Vout = f( Vin ) with f a second-order polynomial whose coefficients don't change with time.
- However this adjustment mechanism is upset if the distortion profile changes with time : then we can't adjust to it.
This would be : Vout = f( Vin, time ) where the coefficients of f change with time.
It's the same for frequency response : we can get used to non-flat frequency responses (no speaker is 100% flat) but when we change speaker, we tend to first hear the differences between what we're used to and the new one.
So, temperature variations in transistors are a bandpass function of the dissipated power, hence we get :
Vout = f( Vin, and also the history of Vin in the past few seconds )
For instance, take a power amp with a perfectly balanced input stage and run a 1 kHz sine through it (like a piano note decaying). You'll get very low distortion. Now, add a powerful transient (like the piano attack).
The worst case is if the transient clips the amp : in this case for a few ms, the input stage will have very different dissipated powers. Until they come back to the same temperature, the input stage will be imbalanced, and the distortion spectrum of the amp will change.
So, if you could measure an instantaneous distortion spectrum, you'd get different ones depending on the instant you measure.
Anli,
---We must measure THD (or, which is better, measure a spectrum) from rather low frequencies (almost from DC).---
I thought of that for a long time. Maybe a THD measurement at 3 Hz could be significant. However, what would be the apparatus for such tests ?
I propose another method.
Have a DC power supply which output of is modulated by a 3 Hz sine amplitude of some volts.
The device under test is submitted to a very classical but very sensible THD test at, say, 1V 1000 Hz.
If the output has some temperature dependence, the THD value will show variations at a rate of 3 Hz.
***
Gerard Perrot wrote very interesting papers, two main subjects he dealt with were :
- the research of gain stages having very high linearity and no temperature sensitivity. He concluded that the command device of a composite gain stage should be held at nearly constant voltage and current. He then patented a sophisticated cascoded Sziklai pair where the input transistor is loaded by a constant current source
- the DC conditions of amplifiers, with the notion of two different offsets : the first one due to the input stage, and the second due the next stages (VAS and output). He concluded that to avoid the non-linearity of the input stage due any DC fed back from the output, the DC feedback loop of an amplifier should not include the input stage. He also patented the idea. See here an example of implementation of DC feedback loop outside of the input stage :
http://perso.orange.fr/francis.audio2/C25_HERVELEB.gif
It occurred to me that the two subjects are intimately related and that Perrot has disregarded one of the input gain stages he studied. It is as linear as his patented circuit, and contrarily to it, is not sensitive nor to temperature neither to input offset.
---We must measure THD (or, which is better, measure a spectrum) from rather low frequencies (almost from DC).---
I thought of that for a long time. Maybe a THD measurement at 3 Hz could be significant. However, what would be the apparatus for such tests ?
I propose another method.
Have a DC power supply which output of is modulated by a 3 Hz sine amplitude of some volts.
The device under test is submitted to a very classical but very sensible THD test at, say, 1V 1000 Hz.
If the output has some temperature dependence, the THD value will show variations at a rate of 3 Hz.
***
Gerard Perrot wrote very interesting papers, two main subjects he dealt with were :
- the research of gain stages having very high linearity and no temperature sensitivity. He concluded that the command device of a composite gain stage should be held at nearly constant voltage and current. He then patented a sophisticated cascoded Sziklai pair where the input transistor is loaded by a constant current source
- the DC conditions of amplifiers, with the notion of two different offsets : the first one due to the input stage, and the second due the next stages (VAS and output). He concluded that to avoid the non-linearity of the input stage due any DC fed back from the output, the DC feedback loop of an amplifier should not include the input stage. He also patented the idea. See here an example of implementation of DC feedback loop outside of the input stage :
http://perso.orange.fr/francis.audio2/C25_HERVELEB.gif
It occurred to me that the two subjects are intimately related and that Perrot has disregarded one of the input gain stages he studied. It is as linear as his patented circuit, and contrarily to it, is not sensitive nor to temperature neither to input offset.
Anli,
"Memory distorsion" was first named "thermal distorsion" by Perrot.
It seems that Kinergetics Research amplifiers had the same kind of ideas and called it "hysteresis distorsion".
Varying the power supply can vary the thermal dissipation a lot more than that the input signal can do, magnifying all the effects.
"Memory distorsion" was first named "thermal distorsion" by Perrot.
It seems that Kinergetics Research amplifiers had the same kind of ideas and called it "hysteresis distorsion".
Varying the power supply can vary the thermal dissipation a lot more than that the input signal can do, magnifying all the effects.
Justcallmedad said:
Yes, I also made the analogy with DA. The shift of the fundamental of the thermal distorsion seems to be a similar effect, although it adds to the main fundamental of the signal. However, yes you are right that we don't capture this with THD measurements. It was perhaps sloppy of me to use the term THD, which strictly speaking is just a single figure. I was rather thinking of looking at the amplitude and phase spectrum.
peufeu said:
That makes a lot of sense. It seems intuitively reasonable that the brain should have much easier to deal with a constant distorsion profile than a varying one. Of course, the distorsion profile will never be quite constant, since it will vary with amplitude and other factors, but of course, the less variation, the easier it ought to be for the brain. It remains, thus, to show that thermal distorsion is of a more varying nature. However, there seems to be one fundamental difference between thermal distorsion and the distorsion inherent in the transfer function of a device. The distorsion components of the transfer function will have a fixed phase shift, while the components of the thermal distorsion will have a frequency dependent phase shift, due to the thermal impedance. I suppose this is what you have been saying all the time, but in different words. I just want to try making it more precise what we are dealing with and talking about.
peufeu said:
Forgive me for being stubborn, but I do wish we can make things a bit more precise than that, and replace such loose terminology with equations on the level of transistor models. If we discuss on the level on such terms as "memory", "history" etc. without properly defining them it is very easy to get deluded by false intuitions. I am absolutely not saying you are wrong, I just want us to get more precise and arrive at something that can be directly related to semiconductor physics.
peufeu said:
Seems reasonable. Howver we still need to be more precise. Are we talking about a strongly assymetric transient here that disrupts the power dissipation, or rather a small amplitude HF signal with a sudden onset of a simultaneous large amplitude LF signal? I think the problem may arise in both cases, but the reasoning might be different for the two cases.
I have however realized at least one errror in my previous reasoning. I did fall into the trap of sloppily trusting my intuition without thinking further. I was assuming we mainly considered the input stage of an amp and thus class A. Unfortunately I did the mistake of assuming the average power disspation for a symmetric signal equals the quiescent power disspation, which is clearly wrong if just bothering to jot down a few equations and do some calculations. In reality the average power is lower than the Q power, so the temperature should actually decrease when a signal is applied (unless I made some error again). This is one very good example of why one should be very careful with trusting the instinct and not check the equations.
This does, of course, have the consequence that the temperature will not only vary on a cycle by cycle basis but also on a slower time scale. Yes, you have been saying that, but I also like to know why, which I think I do now.
peufeu said:
Probably yes. However, that will not only happen with thermal distorsion. Suppose we could have only the distorsion arising from the the transfer function. Then use an input signal with a large amplitude LF signal and a low amplitude HF signal. Then the distorsion spectrum for the HF signal would vary depending on where in the cycle of the LF signal we are. However, contrary to the case of thermal distorsion it would be correlated to the LF signal.
BTW, why must we try to measure MD in terms of spectrum? Why must we do long, hard and not intuitive job of such concepts converting 
Out ears have nothing common with spectrum-meter.
Probably, some square wave curve deformation observing is more appropriate and much simpler (of course, any hf ring must be eliminated by lp filter, i.e., input rising/falling must not be too fast to invoke any hf artifacts).
Out ears have nothing common with spectrum-meter.Probably, some square wave curve deformation observing is more appropriate and much simpler
(of course, any hf ring must be eliminated by lp filter, i.e., input rising/falling must not be too fast to invoke any hf artifacts).- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.