Re: Question to ANDY_C
Jan, the reason I posted the LTSpice projects is so people can verify the data themselves, using a program that's available for free, and try out whatever they want. It's simply too much of a burden on me to simulate every configuration that anyone could come up with. My job typically takes me from 730AM to 630PM plus an hour commute each way in LA traffic. This leaves me with very little time. This thread has consumed a lot of my time already. Far too much actually.
The open-loop gains at 20 kHz were originally equalized by setting the closed-loop amplifer bandwidths equal as I've mentioned several times. Of course, this technique only works if the frequency of interest is well above the open-loop pole of both configurations, which it is. Later, I changed over to use a simplified version of the Middlebrook technique for measuring the loop gain, which is described at:
http://www.spectrum-soft.com/news/spring97/loopgain.shtm
That's what the voltage source in the feedback loop of the later schematics is about. The full version of the Middlebrook technique requires both a voltage source and a current source to be injected. The simplified version only requires a single voltage source. The simplified version gives exact results, but only if the current through the source is zero. Since the VCVS in the loop guarantees this, the technique is justified. The loop gains as matched via the equalization of closed-loop bandwidth ended up being off by 0.4 dB when measured by the Middlebrook technique. As I've already mentioned, subsequent equalization of the loop gains resulted in the distortion figures being within 0.1 dB of their previous values. So I didn't repeat them. The distortion was for 20 kHz fundamental.
janneman said:
Jan, the reason I posted the LTSpice projects is so people can verify the data themselves, using a program that's available for free, and try out whatever they want. It's simply too much of a burden on me to simulate every configuration that anyone could come up with. My job typically takes me from 730AM to 630PM plus an hour commute each way in LA traffic. This leaves me with very little time. This thread has consumed a lot of my time already. Far too much actually.
The open-loop gains at 20 kHz were originally equalized by setting the closed-loop amplifer bandwidths equal as I've mentioned several times. Of course, this technique only works if the frequency of interest is well above the open-loop pole of both configurations, which it is. Later, I changed over to use a simplified version of the Middlebrook technique for measuring the loop gain, which is described at:
http://www.spectrum-soft.com/news/spring97/loopgain.shtm
That's what the voltage source in the feedback loop of the later schematics is about. The full version of the Middlebrook technique requires both a voltage source and a current source to be injected. The simplified version only requires a single voltage source. The simplified version gives exact results, but only if the current through the source is zero. Since the VCVS in the loop guarantees this, the technique is justified. The loop gains as matched via the equalization of closed-loop bandwidth ended up being off by 0.4 dB when measured by the Middlebrook technique. As I've already mentioned, subsequent equalization of the loop gains resulted in the distortion figures being within 0.1 dB of their previous values. So I didn't repeat them. The distortion was for 20 kHz fundamental.
Re: question and suggestion
Eric,
I just used the cursor, doing some zooming first. Not elegant, but it works. In the LTSpice Yahoo user's group, there's a program posted by Helmut Sennewald that allows conversion of the binary .raw file to text for processing. This allows bringing it into Excel or whatever. This takes lots of time though. I've used this to analyze crossover distortion of output stages. I did a nonlinear DC transfer curve, then exported the data. I did a linear regression to find the best fit straight line through the data, then subtracted the best fit line from the nonlinear transfer curve. This allowed viewing the deviation from linearity in high resolution.
But things like VAS distortion are to some extent dynamic. The collector-base capacitance is a function of voltage, which is a function of time. So the DC transfer curve doesn't tell the whole story (and I admit it doesn't tell the whole story for my crossover distortion example either). Also, removing the feedback makes the amplifier DC unstable, since there's now nothing to set the voltage at the junction of the VAS collectors.
capslock said:
Eric,
I just used the cursor, doing some zooming first. Not elegant, but it works. In the LTSpice Yahoo user's group, there's a program posted by Helmut Sennewald that allows conversion of the binary .raw file to text for processing. This allows bringing it into Excel or whatever. This takes lots of time though. I've used this to analyze crossover distortion of output stages. I did a nonlinear DC transfer curve, then exported the data. I did a linear regression to find the best fit straight line through the data, then subtracted the best fit line from the nonlinear transfer curve. This allowed viewing the deviation from linearity in high resolution.
But things like VAS distortion are to some extent dynamic. The collector-base capacitance is a function of voltage, which is a function of time. So the DC transfer curve doesn't tell the whole story (and I admit it doesn't tell the whole story for my crossover distortion example either). Also, removing the feedback makes the amplifier DC unstable, since there's now nothing to set the voltage at the junction of the VAS collectors.
millwood said:
1) I'm not sure, but I think that ANY circuit will generate both even and odd harmonics, because no circuit is ever fully assymmeteric or fully symmetric. But if a circuit has a reasonable symmetry, the waveform distortion will be reasonable symmetric, which means predominantly odd order thd. But such a circuit will also generate even order harmonics because it is not perfectly symmetric.
2) No, squarewaves (ideal) consist strictly of odd harmonics, IMMSMW in the ratio related to their order, in other words a 1kHz square wave of 1 V peak consists of the following sine waves: 1V 1kHz, 1/3V 3kHz, 1/5v 5kHz, 1/7v 7kHz etc. I think.
Now, if the squarewave has not exactly 50% duty cycle, iow if it isn't exactly symmetric, you will start to see even order harmonics as well. Isn't this beautiful?
Jan Didden
janneman said:
That's the right ratio for the harmonics, but the amplitudes all need to be multiplied by 4/pi. This gives the weird result that the fundamental has a higher amplitude than the square wave itself. There are harmonics which are out of phase with the fundamental that reduce the total amplitude when you sum them all up.
Re: No Logic, just subjective
This is an interesting result! The cascode of the VAS should reduce distortion quite a bit, while the cascode of the input diff amp shouldn't affect distortion much at all. By any chance, does your amp have an input low-pass filter? The reason I mention this is that cascoding the diff amps will reduce the input capacitance of the amplifier. If the input stage has a high voltage gain, it will reduce the input capacitance a lot. If an input low-pass filter is present, this will increase the cutoff frequency of that filter by virtue of the reduced input capacitance. That might explain the difference in sound you experienced.
Dave S said:
This is an interesting result! The cascode of the VAS should reduce distortion quite a bit, while the cascode of the input diff amp shouldn't affect distortion much at all. By any chance, does your amp have an input low-pass filter? The reason I mention this is that cascoding the diff amps will reduce the input capacitance of the amplifier. If the input stage has a high voltage gain, it will reduce the input capacitance a lot. If an input low-pass filter is present, this will increase the cutoff frequency of that filter by virtue of the reduced input capacitance. That might explain the difference in sound you experienced.
If I may be permitted to dispute a statement that I recall Nelson Pass making about the virtues of asymmetrical design:
The claim that because musical waveforms are asymmetrical, amplifiers should be asymmetrical to sound natural is false. To reproduce an asymmetrical waveform correctly requires symmetrical reproduction of the asymmetrical waveform! Otherwise you would be adding extra asymmetry.
I still think symmetrical amps sound less "phasey", for want of a better word!
The claim that because musical waveforms are asymmetrical, amplifiers should be asymmetrical to sound natural is false. To reproduce an asymmetrical waveform correctly requires symmetrical reproduction of the asymmetrical waveform! Otherwise you would be adding extra asymmetry.
I still think symmetrical amps sound less "phasey", for want of a better word!
owdeo said:
Maybe this is your word for what the Pass XA product literature calls a "fluffy coloration" in the single-ended Alephs. I can hear this when comparing an Aleph 3 to another balanced MOSFET design which has an almost pure 3rd harmonic residual. Still, taken on its own, I think the A3 has a beguiling sound. Overall, the balanced sound strikes me as more neutral but at the same time less forgiving of imperfect source material.
janneman said:
Jan: my point to your prior post is that it is not a natural extension to go from "asymmetrical waveforms have odd/even harmonics" to "asymmetrical topologies generate odd/even harmonics".
For example, a sraight wire with gains is perfectly symmetrical (in whatever way you measure it), and it generates no harmonics of its own. and if you have a perfect asymmetrical amp mimicing the straight wire, it will generate no harmonics of its own either.
I am not convinced that it is by definition that asymmetrical topology generates odd harmonics and symmetrical topology even harmonics. There doesn't seem to be natural links here.
"I then added cascodes to the LTPs and VASs. The cascodes on the LTPs made the sound worse and I heard very little difference with/without them on the VASs.
This is an interesting result! The cascode of the VAS should reduce distortion quite a bit, while the cascode of the input diff amp shouldn't affect distortion much at all. By any chance, does your amp have an input low-pass filter? The reason I mention this is that cascoding the diff amps will reduce the input capacitance of the amplifier. If the input stage has a high voltage gain, it will reduce the input capacitance a lot. If an input low-pass filter is present, this will increase the cutoff frequency of that filter by virtue of the reduced input capacitance. That might explain the difference in sound you experienced."
This is the most interesting discussion Iv'e seen in long time, and I'm totally snowed under with work right now, but early this am, I played around in Circuitmaker a little, and yes, cascoding the diff amp increase distortion big time. Probably beacuse it decreases the diff amp collector load due to emitter loading?
Very quick findings, single ended Self type topology, just diff amp and VAS, no output stage, 1.2 mA per diff tranny, 12 mA for VAS, 100 ohm diff tails, 100 ohm in current mirror emitters, x10 closed loop gain, all 2N3904/6 devices, Cdom is 20 pF, 68 ohm in VAS emitter to rail:
1 - Cascoding the diff current source and VAS current source show by far the biggest improvements. Cascoding the VAS shows a much smaller improvement.
2 - Emitter follower between diff amp and VAS shows second best improvement, opposite sex than VAS device being slightly better than same sex, and about the same as a super pair VAS.
3 - CFP configuration for the diff amp trannies also improve things slightly, secondary effect is that DC offset decreases dramatically, due to higher current gain, thus loading feedback network and input bias resistor less. Probably linearises the diff amp quite a bit.
4 - The simulation with all these implemented, shows 2nd harmonic about 122 dB down at 10 KHz, after that about 130 dB down for ALL harmonics, odd and even, up to 10th.
Darn AC analysis craps out on me, and I don't have teh time to chase it down.
Back to work
Lukas
This is an interesting result! The cascode of the VAS should reduce distortion quite a bit, while the cascode of the input diff amp shouldn't affect distortion much at all. By any chance, does your amp have an input low-pass filter? The reason I mention this is that cascoding the diff amps will reduce the input capacitance of the amplifier. If the input stage has a high voltage gain, it will reduce the input capacitance a lot. If an input low-pass filter is present, this will increase the cutoff frequency of that filter by virtue of the reduced input capacitance. That might explain the difference in sound you experienced."
This is the most interesting discussion Iv'e seen in long time, and I'm totally snowed under with work right now, but early this am, I played around in Circuitmaker a little, and yes, cascoding the diff amp increase distortion big time. Probably beacuse it decreases the diff amp collector load due to emitter loading?
Very quick findings, single ended Self type topology, just diff amp and VAS, no output stage, 1.2 mA per diff tranny, 12 mA for VAS, 100 ohm diff tails, 100 ohm in current mirror emitters, x10 closed loop gain, all 2N3904/6 devices, Cdom is 20 pF, 68 ohm in VAS emitter to rail:
1 - Cascoding the diff current source and VAS current source show by far the biggest improvements. Cascoding the VAS shows a much smaller improvement.
2 - Emitter follower between diff amp and VAS shows second best improvement, opposite sex than VAS device being slightly better than same sex, and about the same as a super pair VAS.
3 - CFP configuration for the diff amp trannies also improve things slightly, secondary effect is that DC offset decreases dramatically, due to higher current gain, thus loading feedback network and input bias resistor less. Probably linearises the diff amp quite a bit.
4 - The simulation with all these implemented, shows 2nd harmonic about 122 dB down at 10 KHz, after that about 130 dB down for ALL harmonics, odd and even, up to 10th.
Darn AC analysis craps out on me, and I don't have teh time to chase it down.
Back to work
Lukas
millwood said:
Well, if you put it that way, there *is* a natural link here in that symmetically-only distorted waveforms (both polarities flattened or peaked to the same order), when picked apart, will show only odd harmonics. So, the expectation is that if you have a symmetrical amp where both polarities distort (in the same way), the waveform will show symmetrical distortion which according to mr fft IS odd order thd.
I realise that in the real world nothing is only symmetrical or asymetrical of course, but basically the relationship is the way I described.
(BTW You have them switched in your statement, symmetrical>odd, assymmetrical>even, which is counter-intuitive. we think of symmetrical as " better" and also even as " better". another trick we play on ourselfs).
Jan Didden
LukasLouw said:
Lukas, you're absolutely right. I didn't realize until later that the Miller integrator presents such a low load impedance to the input diff amp at high frequencies that the voltage gain of the diff amp will be more with cascoding than without. Serves me right for posting late at night again.
I need to figure out how to get lower residual distortion out of LTSpice.
Thanks for the info Lukas...
janneman said:
so let's say your have an amp, that does 1:1 amplification until it clips. and it clips at 1v input level.
so you feet a 2v sine wave into this box, and you got a lot (more) of odd harmonics out of it.
Question: what design is this box, symmetrical or assymmetrical? The inside of the box is actually quite simple but the way,
This thread is interesting because it constantly throws up the blind spot we humans suffer in all our technological endeavours.
This is actually about how we look at the problem, and how we draw conclusions from our highly subjective interpretations. Amplifier topology is probably as much about meta-knowledge as knowledge.
In essence, odd order harmonic distortions, which are less musical than even order harmonic distortions, are produced, according to Monsieur Fourier, when the distortion profile affects both positive and negative half cycles the same way, or 'symmetrically' if you prefer.
Even order harmonic distortion, which for the foregoing reason are preferable, are produced by topologies which affect the entire cycle asymmetrically; typically by sharpening the positive half cycle and broadening the negative half cycle, or vice versa.
Global negative feedback around a typical SS amplifier shows decreasing effectiveness with frequency, and in any case the Bode-Nyquist constraints require that voltage gain be pulled down to below unity by the pole frequency of the amplifier, which is typically greater than 400KHz. The result of this is that the odd order harmonic distortions, H3, H5, H7, H9 etc, will be less effectively corrected out of the mix than the marginally slower even order distortions.
Now the interesting bit. Humans go for symmetry in a big way. They are almost blind to asymmetry, and yet it pervades Nature. We might be bilaterally symmetrical, but we still have one mouth, one nose, one stomach, one spleen, liver, pancreas, etc. In fact surgeons tell me that when you open up the abdomen there is very little symmetry. That a symmetrical topology intrinsically produces more odd than even order distortion is one of technology's grand jokes. It is hardly surprising that this produces some discomfort; we are used to symmetry and we all love to adjust our CROs to show off this pleasing aspect of audio.
Cascodes. The low and non-ohmic impedance the lower transistor is working into (Zin is equal to 26/mA, the reciprocal of the transconductance of the upper transistor) naturally causes non-linearities. The linearity of the cascode can be improved, particularly at lower clip, by inserting a small resistor between the transistors. I found this is the effective way to make this interesting topology work well so that at lower clip it does not produce a spray of odd-order artefacts.
Those of you who like math will notice that there is none in this analysis. That's because I don't believe the math helps too much here; you don't want to understand this process by studying a model; you want to understand it by carefully examining what is happened. I humbly suggest this approach is just as effective.
Cheers,
Hugh
This is actually about how we look at the problem, and how we draw conclusions from our highly subjective interpretations. Amplifier topology is probably as much about meta-knowledge as knowledge.
In essence, odd order harmonic distortions, which are less musical than even order harmonic distortions, are produced, according to Monsieur Fourier, when the distortion profile affects both positive and negative half cycles the same way, or 'symmetrically' if you prefer.
Even order harmonic distortion, which for the foregoing reason are preferable, are produced by topologies which affect the entire cycle asymmetrically; typically by sharpening the positive half cycle and broadening the negative half cycle, or vice versa.
Global negative feedback around a typical SS amplifier shows decreasing effectiveness with frequency, and in any case the Bode-Nyquist constraints require that voltage gain be pulled down to below unity by the pole frequency of the amplifier, which is typically greater than 400KHz. The result of this is that the odd order harmonic distortions, H3, H5, H7, H9 etc, will be less effectively corrected out of the mix than the marginally slower even order distortions.
Now the interesting bit. Humans go for symmetry in a big way. They are almost blind to asymmetry, and yet it pervades Nature. We might be bilaterally symmetrical, but we still have one mouth, one nose, one stomach, one spleen, liver, pancreas, etc. In fact surgeons tell me that when you open up the abdomen there is very little symmetry. That a symmetrical topology intrinsically produces more odd than even order distortion is one of technology's grand jokes. It is hardly surprising that this produces some discomfort; we are used to symmetry and we all love to adjust our CROs to show off this pleasing aspect of audio.
Cascodes. The low and non-ohmic impedance the lower transistor is working into (Zin is equal to 26/mA, the reciprocal of the transconductance of the upper transistor) naturally causes non-linearities. The linearity of the cascode can be improved, particularly at lower clip, by inserting a small resistor between the transistors. I found this is the effective way to make this interesting topology work well so that at lower clip it does not produce a spray of odd-order artefacts.
Those of you who like math will notice that there is none in this analysis. That's because I don't believe the math helps too much here; you don't want to understand this process by studying a model; you want to understand it by carefully examining what is happened. I humbly suggest this approach is just as effective.
Cheers,
Hugh
@ Andy, Lukas, jcx
Lukas, Andy,
this beats me, can you explain some more? Of course, the cascode on the LTP increases its output impedance, thereby increasing gain, if only a little because of the low input impedance of the following VAS input. But why would this result in increased distortion??
Another beneficial effect of the cascode on the LTP would be to reduce common mode distortion, even if this is probably not too significant for >+/- 30 V rails and a close loop gain of around 20.
jcx, I believe you said somewhere (headphone thread?) that the Skilazi CFP output configuration has higher input impedance than a double emitter follower. Is this really the case?
Regards,
Eric
andy_c said:
Lukas, you're absolutely right. I didn't realize until later that the Miller integrator presents such a low load impedance to the input diff amp at high frequencies that the voltage gain of the diff amp will be more with cascoding than without. Serves me right for posting late at night again.
I need to figure out how to get lower residual distortion out of LTSpice.
Thanks for the info Lukas...
Lukas, Andy,
this beats me, can you explain some more? Of course, the cascode on the LTP increases its output impedance, thereby increasing gain, if only a little because of the low input impedance of the following VAS input. But why would this result in increased distortion??
Another beneficial effect of the cascode on the LTP would be to reduce common mode distortion, even if this is probably not too significant for >+/- 30 V rails and a close loop gain of around 20.
jcx, I believe you said somewhere (headphone thread?) that the Skilazi CFP output configuration has higher input impedance than a double emitter follower. Is this really the case?
Regards,
Eric
"this beats me, can you explain some more? Of course, the cascode on the LTP increases its output impedance, thereby increasing gain, if only a little because of the low input impedance of the following VAS input. But why would this result in increased distortion??"
The cascode emitters load the heck out of the diff amp collectors, and that attenuation means that the cascode stage has to make up the gain. so your open loop gain does not necessarily increase. These common base stages are not as linear as one might want them to be.
The subjective improvement does not come from the so called increased gain or reduced THD, but by linearising teh diff amps with more constant low voltage across them.
Lukas
The cascode emitters load the heck out of the diff amp collectors, and that attenuation means that the cascode stage has to make up the gain. so your open loop gain does not necessarily increase. These common base stages are not as linear as one might want them to be.
The subjective improvement does not come from the so called increased gain or reduced THD, but by linearising teh diff amps with more constant low voltage across them.
Lukas
"The cascode emitters load the heck out of the diff amp collectors, and that attenuation means that the cascode stage has to make up the gain. so your open loop gain does not necessarily increase. These common base stages are not as linear as one might want them to be.
The subjective improvement does not come from the so called increased gain or reduced THD, but by linearising teh diff amps with more constant low voltage across them."
In fact, the same thing happens with the VAS, but to a lesser degree, to the point where it may not be noticable unless you go looking for it. Simulate 2 identical power amps in teh same design, and make changes one at a time in amp, with teh other one as reference, and these differences can easily be discerned.
The small resistors that Dean Hugh had mentioned between a stage collector and cascode emitter simply degenerate the cascode device, thus linearizing it, which helps for the VAS, but not for the diff amps in my experience, unless they are so large to teh piont where they introduce even more massive attenutaion and DC volt drop.
Lukas
The subjective improvement does not come from the so called increased gain or reduced THD, but by linearising teh diff amps with more constant low voltage across them."
In fact, the same thing happens with the VAS, but to a lesser degree, to the point where it may not be noticable unless you go looking for it. Simulate 2 identical power amps in teh same design, and make changes one at a time in amp, with teh other one as reference, and these differences can easily be discerned.
The small resistors that Dean Hugh had mentioned between a stage collector and cascode emitter simply degenerate the cascode device, thus linearizing it, which helps for the VAS, but not for the diff amps in my experience, unless they are so large to teh piont where they introduce even more massive attenutaion and DC volt drop.
Lukas
I think that if a cct element has to 'work' to achieve it's function then this will almost certainly be heard.
In my self taught world of electronics I have the idea that it's good to follow a device with high o/p impedance ( like a current mirror or indeed a cascode ) with a device with an even higher i/p impedance i.e. a mosfet. typically i/p stage to driver stage.
this means it can operate with ease, without loss of gain and without increase in distortion.
with the extra effortless gain the sound of the amp should then be more dominated by the quality of the components, and layout of the feedback cct
what do the experts think of this policy ?
mike
In my self taught world of electronics I have the idea that it's good to follow a device with high o/p impedance ( like a current mirror or indeed a cascode ) with a device with an even higher i/p impedance i.e. a mosfet. typically i/p stage to driver stage.
this means it can operate with ease, without loss of gain and without increase in distortion.
with the extra effortless gain the sound of the amp should then be more dominated by the quality of the components, and layout of the feedback cct
what do the experts think of this policy ?
mike
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.