I am happy to see we finally get a discussion on TIM. I remember
there being quite a fuzz about this back in the seventies when
Otala and others introduced this and other similar concept and
high slewrate was the catchword of the day. Not having really
followed much of the techincal aspects of audio until recently
I am surprised to see so little discussion of these aspects. Mostly
it is not mentioned at all, and some experienced designers comment
on old designs that, well, we thought that TIM was important
back then. Slone mentions TIM and seems to relate it entirely to
current starving from input stage to VAS. George Barrie at AD
says something similar (if I remember correctly) and boldly claims
that TIM was just a new name for something that was well-known
to most EEs, except in audio.
I tried myself to bring up the topic of TIM
in a thread some time ago, getting no reaction at all, leaving me
somewhat puzzled about the general opinion on the issue.
This discussion at least seems to confirm what I have managed
to assemble from different sources.
1) The input should be LP filtered to avoid high-slewrate input
signals.
2) The input diff pair can overload into the non-linear region if
the feedback signal is too delayed for step responses. This is
obvious from the basic theory, and is also discussed by Marshall
Leach in the accompanying notes on his Leach amp. This is the
motivation for his use of very heavy emitter degenration on the
input diff pair. (this may not be needed if the rest of the amp
is fast enough, though).
3) the input diff pair should have a sufficient current capacity
to feed the VAS, including compensation caps, without getting
starved for current.
4) VAS and OPS should be fast enough so the input diff pair
cannot overload.
Does this seem like a reasonably correct account of this issue?
there being quite a fuzz about this back in the seventies when
Otala and others introduced this and other similar concept and
high slewrate was the catchword of the day. Not having really
followed much of the techincal aspects of audio until recently
I am surprised to see so little discussion of these aspects. Mostly
it is not mentioned at all, and some experienced designers comment
on old designs that, well, we thought that TIM was important
back then. Slone mentions TIM and seems to relate it entirely to
current starving from input stage to VAS. George Barrie at AD
says something similar (if I remember correctly) and boldly claims
that TIM was just a new name for something that was well-known
to most EEs, except in audio.
I tried myself to bring up the topic of TIM
in a thread some time ago, getting no reaction at all, leaving me
somewhat puzzled about the general opinion on the issue.
This discussion at least seems to confirm what I have managed
to assemble from different sources.
1) The input should be LP filtered to avoid high-slewrate input
signals.
2) The input diff pair can overload into the non-linear region if
the feedback signal is too delayed for step responses. This is
obvious from the basic theory, and is also discussed by Marshall
Leach in the accompanying notes on his Leach amp. This is the
motivation for his use of very heavy emitter degenration on the
input diff pair. (this may not be needed if the rest of the amp
is fast enough, though).
3) the input diff pair should have a sufficient current capacity
to feed the VAS, including compensation caps, without getting
starved for current.
4) VAS and OPS should be fast enough so the input diff pair
cannot overload.
Does this seem like a reasonably correct account of this issue?
Christer said:
Me too.
Seems like a reasonably correct account of this issue, to me.
What pulled me in this direction, is all the discussion and lack of understanding between SS and Tube amp sound. My problem is that I don't know the right terminology to use in describing what I can see in my mind's eye.
Tube amps for music production and re-production should be kept separate, IMO. For re-production purposes a SS amp and Tube amp can be designed to accurately re-produce the original source. Second harmonic distortion can be added to either, if that is desired.
I think TIM is very important and for some reason largely ignored, by some audio engineers/technicians. If it's not ignored, then it seems to get little discussion.
Keep it coming!
Cheers
Aartificial generation of harmonics???
I have for a while been toying with an idea, but it is very long since
I took the courses on transform theory and I can't claim to ever
have had a very good understanding of this topic.
The idea is to write a computer program that takes a wav file
and adds a distorsion spectrum of choice to the signal. This
would make it possible to get an idea of how various distorsion
spectra affect the sound. To avoid the problems of computer
soundcard, these resulting files could be burnt on a CDR and
played on the equipment of choice. We might not be able to
simulate very small distorsion levels well, however, maybe it
is anyway better to use an exaggerated level of distorsion to
more clearly illustrate the difference.
The problem is how to process the signal to get an arbitraly
choosen distorsion spectrum? I glanced through my old books
but couldn't find any hint of how to do this or if it is even possible.
Is there some, not too elaborate way to do this? I am not asking
about how to write the program, just about the mathematics.
A simpler way could be to just compute an arbitrary polynomial
function of the input, although that makes it more difficult to
control exactly what distorsion spectrum one gets.
Opinions on the idea and hints on the maths are welcome.
I have for a while been toying with an idea, but it is very long since
I took the courses on transform theory and I can't claim to ever
have had a very good understanding of this topic.
The idea is to write a computer program that takes a wav file
and adds a distorsion spectrum of choice to the signal. This
would make it possible to get an idea of how various distorsion
spectra affect the sound. To avoid the problems of computer
soundcard, these resulting files could be burnt on a CDR and
played on the equipment of choice. We might not be able to
simulate very small distorsion levels well, however, maybe it
is anyway better to use an exaggerated level of distorsion to
more clearly illustrate the difference.
The problem is how to process the signal to get an arbitraly
choosen distorsion spectrum? I glanced through my old books
but couldn't find any hint of how to do this or if it is even possible.
Is there some, not too elaborate way to do this? I am not asking
about how to write the program, just about the mathematics.
A simpler way could be to just compute an arbitrary polynomial
function of the input, although that makes it more difficult to
control exactly what distorsion spectrum one gets.
Opinions on the idea and hints on the maths are welcome.
That sounds pretty interesting. But an output spectrum by itself is not enough information. Suppose you had a simple nonlinear transfer function like below:
vo = a1*vi + a3*vi3
You might pick the input voltage to be a sinusoid so that the third harmonic was, say, 80 dB down from the fundamental. But in the weakly nonlinear region, if you reduced the input signal by 1 dB, the third harmonic would now be 82 dB down from the fundamental. So the spectrum is not enough information. One would have to define some reference signal level and then figure out the Taylor series expansion that would give the desired harmonics at that level. And that only takes into account static nonlinearities which of course don't depend on frequency.
vo = a1*vi + a3*vi3
You might pick the input voltage to be a sinusoid so that the third harmonic was, say, 80 dB down from the fundamental. But in the weakly nonlinear region, if you reduced the input signal by 1 dB, the third harmonic would now be 82 dB down from the fundamental. So the spectrum is not enough information. One would have to define some reference signal level and then figure out the Taylor series expansion that would give the desired harmonics at that level. And that only takes into account static nonlinearities which of course don't depend on frequency.
Andy,
Yes, I suppose you are right. On the other hand, since the distorsion
in amps anyway mainly comes from nonlinear behaviour we should
anyway get a relative distorsion spectrum that differs with signal
level. So I guess the change in harmonic distribution with the level
of the fundamental is yet a parameter to take into account. Not
that it makes thing simpler.
A simple version would be to just be able to set the polynomial
coefficients and be able to do an fft at varios levels to see what
spectrum one gets and then use trial and error to get interesting
spectra for comparison.
Yes, I suppose you are right. On the other hand, since the distorsion
in amps anyway mainly comes from nonlinear behaviour we should
anyway get a relative distorsion spectrum that differs with signal
level. So I guess the change in harmonic distribution with the level
of the fundamental is yet a parameter to take into account. Not
that it makes thing simpler.
A simple version would be to just be able to set the polynomial
coefficients and be able to do an fft at varios levels to see what
spectrum one gets and then use trial and error to get interesting
spectra for comparison.
I agree. So it wouldn't be "adding a distortion spectrum", but rather "imposing a nonlinearity that has a certain distortion spectrum at some chosen reference level". The variation in distortion products with level should be pretty predictable from the mathematics, provided the circuit is not overdriven to very high distortion levels. The cubed term gives distortion products at the third harmonic that are proportional to the input amplitude cubed, so they will increase 3 dB for every 1 dB increase in the fundamental. But the cubed term will also contribute distortion at the fundamental frequency (compression if a3 is negative, the usual case). So if it's driven too hard, the compression of the fundamental will mess up the simple relationship of relative distortion level to input level.
nov 03 jaes has a article on distortion perception/measurement correlation
for volterra math to model distortion in audio:
http://www.tele.ntnu.no/akustikk/meetings/DAFx99/schattschneider.pdf
"A variety of computational models have been proposed for digital
simulation of nonlinear systems with memory [1, 2, 3, 4].
They are dealing with different aspects of the problem, like methods
for identification, avoiding aliasing and fast convolution algorithms.
In this paper we shortly sum up some of the common
approaches and present a straightforward method for bandlimited
discrete-time realization of analog nonlinear audio effects, like
tube amps, exciters etc., using off-time digital cross correlation
measurements. From these measurements we obtain a rather inefficient
Wiener representation of the unknown nonlinearity. We
then reduce the number of required coefficients significantly on the
basis of multi-dimensional Laguerre transformation of the related
Volterra kernels to allow real-time implementation on a digital signal
processor"
for volterra math to model distortion in audio:
http://www.tele.ntnu.no/akustikk/meetings/DAFx99/schattschneider.pdf
"A variety of computational models have been proposed for digital
simulation of nonlinear systems with memory [1, 2, 3, 4].
They are dealing with different aspects of the problem, like methods
for identification, avoiding aliasing and fast convolution algorithms.
In this paper we shortly sum up some of the common
approaches and present a straightforward method for bandlimited
discrete-time realization of analog nonlinear audio effects, like
tube amps, exciters etc., using off-time digital cross correlation
measurements. From these measurements we obtain a rather inefficient
Wiener representation of the unknown nonlinearity. We
then reduce the number of required coefficients significantly on the
basis of multi-dimensional Laguerre transformation of the related
Volterra kernels to allow real-time implementation on a digital signal
processor"
Yes, my original suggestion was a simplification. What would be
interesting to do is to simulate, possibly in an exaggerated way,
what different types of amplifier behaviours would do to the sound.
If a decent such program could be written, it might be possible to
play around with the progam and get an idea of how different
types of amplifiers would sound, at least from a distorions point
of view.
It is trivial to process a wav file. The problem is to figure out how
to process it in a way that gives some idea of how amplifier
distorsion will affect the sound.
interesting to do is to simulate, possibly in an exaggerated way,
what different types of amplifier behaviours would do to the sound.
If a decent such program could be written, it might be possible to
play around with the progam and get an idea of how different
types of amplifiers would sound, at least from a distorions point
of view.
It is trivial to process a wav file. The problem is to figure out how
to process it in a way that gives some idea of how amplifier
distorsion will affect the sound.
Hi, all,
Right now I want to report my simple experiment. I will study the expenations above later.
In my experimental amp (using full differential bipolars), when I remove the Re, the sound seems tobe more detailed and bright. Is this because of TIM was introduced, or it is because the differential has more gain?
When I put Re up to 300ohm (like leach paper), the sounds seems tobe more dull.
Just wondering, inspite of there is TIM (reduced by Re), why so many amp design do not use Re at all?
About miller cap in VAS, between C-B. I also experimented some values of it from 5pf to 470pf. The bigger this C, the sound seems tobe less detailed/bright especially in high frequencies.
But in my amp, eliminating this C doesn;t make the amp oscilate (where there is said that this C is important to keep amp stable).
What is the drawback if I dont use this C at all?
Note that I conducted these experiments without any distortion analyzer whatsoever, I dont know whats going on in the cct. Just observing by amp and speaker only.
Right now I want to report my simple experiment. I will study the expenations above later.
In my experimental amp (using full differential bipolars), when I remove the Re, the sound seems tobe more detailed and bright. Is this because of TIM was introduced, or it is because the differential has more gain?
When I put Re up to 300ohm (like leach paper), the sounds seems tobe more dull.
Just wondering, inspite of there is TIM (reduced by Re), why so many amp design do not use Re at all?
About miller cap in VAS, between C-B. I also experimented some values of it from 5pf to 470pf. The bigger this C, the sound seems tobe less detailed/bright especially in high frequencies.
But in my amp, eliminating this C doesn;t make the amp oscilate (where there is said that this C is important to keep amp stable).
What is the drawback if I dont use this C at all?
Note that I conducted these experiments without any distortion analyzer whatsoever, I dont know whats going on in the cct. Just observing by amp and speaker only.
This is a very good idea. A transistors gain for me is RC/RE.
I can make the differential has no voltage gain, set the RE to match RC.
I think I get the point. Have the differential only do differential (no voltage gain) and having VAS do the voltage gain (but just enough, not over) may lead to good sounding amp. While the output is only gaining current.
The gains in VAS only. If I want a close loop gain of 10X, while the differential has gain of 1, how many gain in the VAS is "GOOD"?
Usually VAS only have RE without any RC. How can we adjust the gain of VAS? By putting RC to ground or some other way?
lumanauw said:
Those are cool experiments! Besides the TIM issue, there's also the gain-bandwidth product of your amp that you're changing by changing RE and C. The gain-bandwidth product is proportional to 1 / (re + RE)C. The resistance re is equal to 26mV / IE where IE is the DC emitter current of one side of the diff amp (half the total). So if you start with, say, RE = 300 Ohms, then reduce it so that re + RE is, say, cut in half, then double C, you'll increase (theoretically) the TIM distortion while holding the gain-bandwidth product constant. It would be really interesting to find out about the results of that. By doing so, you can keep the issues of gain-bandwidth product and TIM separate.
lumanauw said:
I don't think the conventional differential amp front end is necessary. Most tube amps don't have a differential amp front end. Just start off with a voltage amp and drive the current amp. Put local feedback around just the voltage amp. If you want some global feedback, sum it with the local feedback. This way you can adjust the amount of each feedback - 25 percent global and 75 percent local, 50 percent global and 50 percent local, etc.
Thanks AndyC for the experiment guide. re= internal transistor resistance, and C is miller cap? You're pointing the right way, if I wanted to know what is the effect of TIM, we should hold the gain bandwith constant. Does anyone have done this? What is the difference in sound?
In the Leach paper, there is a value of 0.057V (57mV) that is said tobe the value of a differential begin to not working properly. Where is this value come from? Is that in a differential, front transistor and back transistor can differ so much in their base voltage?
I dont have any measurement device, is there any way to measure open loop gain, closed loop gain, gain bandwith product, with only scope and signal generator?
In the Leach paper, there is a value of 0.057V (57mV) that is said tobe the value of a differential begin to not working properly. Where is this value come from? Is that in a differential, front transistor and back transistor can differ so much in their base voltage?
I dont have any measurement device, is there any way to measure open loop gain, closed loop gain, gain bandwith product, with only scope and signal generator?
lumanauw said:
Open loop gain is hard to measure. But the closed loop gain and gain bandwidth product should be pretty easy. For gain bandwidth product, measure the -3dB bandwidth of the amp with a sine wave, taking as the input the non-inverting input of the amp, right at the diff amp base (to eliminate any effect of input low-pass filter if there is one). Multiply this bandwidth by the nominal gain as set by the ratio of feedback resistors and that's your gain bandwidth product. This assumes there's no peaking in the frequency response due to marginal stability, and that it behaves as a simple first-order filter regarding its frequency response. The closed loop gain is just a fancy name for the voltage gain of the amp and is called that just to distinguish between the gains with and without feedback.
Yes, the re is just 26mV divided by the current in one transistor of the diff amp. So if you have a total tail current of 2 mA in the diff pair, you'll have 1 mA in each transistor, so re will be 26 Ohms. For best accuracy, C should be the sum of the compensation capacitor and your estimate of the transistor's collector-base capacitance (Cob) from the transistor data sheets if it's available.
Okay, now I found it, in Figure 7. He says "The linear range is taken to be the region between the dots where the currents vary between 5% and 95% of the maximum value". So what he's saying is that it takes +/-57mV to drive a diff amp with no emitter resistors to the point where the output current goes from 5% of its maximum value to 95% of its maximum value. In other words, the current goes almost from rail to rail with only +/- 57 mV when there's no emitter degeneration.
time delay in power amp
In the Leach paper there is also a writing that wonders me. Somewhere it is said that from the input to the output it takes about 1.5uS delay. Is this true for all amps? Output are not instaneously from input? How about 2 stage power amp, does it takes less time? Or what about 3 stages or more power amp, does it takes longer than 1.5uS delay?
In the Leach paper there is also a writing that wonders me. Somewhere it is said that from the input to the output it takes about 1.5uS delay. Is this true for all amps? Output are not instaneously from input? How about 2 stage power amp, does it takes less time? Or what about 3 stages or more power amp, does it takes longer than 1.5uS delay?
Re: time delay in power amp
A 1.5uS delay? Couldn't be true for all amps. Thinks about the propagation delay in digital circuits. One inverter may have 10 ns delay and so 5 should have 50 ns delay. Current flow in PCB copper and silicon is less than the speed of light, but by how much? Seems like 5 transistors in an analog circuit would cause more delay to a signal than 1 transistor. Just my guess.
lumanauw said:
A 1.5uS delay? Couldn't be true for all amps. Thinks about the propagation delay in digital circuits. One inverter may have 10 ns delay and so 5 should have 50 ns delay. Current flow in PCB copper and silicon is less than the speed of light, but by how much? Seems like 5 transistors in an analog circuit would cause more delay to a signal than 1 transistor. Just my guess.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.