John,
as AD844 is the current feedback opamp, open loop behavior is defined by Zol (open loop transimpedance) rather than Aol. Zol for the AD844 is 3 MOhm, first pole near to 12kHz. This means that Zol is constant in most of the audio band, not following conventional -20dB/decade (usual phase modulation problem). The ratio between open loop and closed loop is given by Zol and Zf (feedback impedance[resistor]) and does not depend on closed loop voltage gain. For my amp, Zol/Zf = 330, i.e. 50 dB. So there is 50 dB of the loop gain, independent on frequency up to 12 kHz. I assume that this and no slew rate limitation (2kV/us, current-on-demand) of the current feedback opamp might be the key to its sound quality.
Regards,
Pavel
as AD844 is the current feedback opamp, open loop behavior is defined by Zol (open loop transimpedance) rather than Aol. Zol for the AD844 is 3 MOhm, first pole near to 12kHz. This means that Zol is constant in most of the audio band, not following conventional -20dB/decade (usual phase modulation problem). The ratio between open loop and closed loop is given by Zol and Zf (feedback impedance[resistor]) and does not depend on closed loop voltage gain. For my amp, Zol/Zf = 330, i.e. 50 dB. So there is 50 dB of the loop gain, independent on frequency up to 12 kHz. I assume that this and no slew rate limitation (2kV/us, current-on-demand) of the current feedback opamp might be the key to its sound quality.
Regards,
Pavel
Yes, I meant the AD797. Almost 40 years ago, I worked extensively with the 2n697 and it gets stuck in my memory.
Page 7 of my data sheet for the AD844 says:"The open loop pole is formed by Rt (2.5Mohm) in parallel with Ct. Since Ct is typically 3pf, the open loop corner frequency occurs at about 18KHz."
For reference the open loop bandwidth of the AD797 is about 100 Hz (or less).
Page 7 of my data sheet for the AD844 says:"The open loop pole is formed by Rt (2.5Mohm) in parallel with Ct. Since Ct is typically 3pf, the open loop corner frequency occurs at about 18KHz."
For reference the open loop bandwidth of the AD797 is about 100 Hz (or less).
traderbam said:
It's not so nasty, of course i exaggerated for better seeing. What
you see is a simple assymetrical unlinearity. The transfercurve for
this would be ~ y=x + (x^2)*0.04 with x offsetted to 0-1. (x=Vs/2+0.5)
I'm not sure about the *0.04... (would be 4%)
A single asymetrical stage with very high gain gives "exactly" this
kind of distortion. If the transfercurve is a pure X^2, you have
2nd harmonics only. That's what tubes tend to do.
An asymetrical r-loaded vas with high gain for example shows
similar transfercurves.
My point was that PM and simple unlinearity seems to be somehow
the same, means that proper feedback should remove this.
Additionaly if PM generates 2nd harmonics, it's not evil ?
But you are right, as zerocrossings are not equispaced, DC is
not balanced, and i am not sure what complex signals will do
except the 2nd harmonic. But i applied these distortions (3%)
to a .WAV and listened to it, it didn't sound really distorted, just
different.
Mike
Is that the FM PM distortion is like the amp is generating signal that is not existed in the original source?
I imagine looking at a helicopter above me. The rotor blades are rotating at high speed, so we cannot see each blade's movement. But we always see there is 4 blades moving very slowly in that helicopter.
Is the FM-PM distortion is like that?
I imagine looking at a helicopter above me. The rotor blades are rotating at high speed, so we cannot see each blade's movement. But we always see there is 4 blades moving very slowly in that helicopter.
Is the FM-PM distortion is like that?
Hi lumanauw !
You are talking about interference, this effect is visible if the helicopter
is recorded with a camera using ~24fps. So these problems apply
to digital sound.
To prove my previous post see my attachement.
Distortion is 12% 2nd harmonic, other harmonics negligible.
The wave looks very similar to the PM-wave.
I forced that heavy distortions by using low current for vas. (0.4ma)
Gain of the circuit is 1:2500.
Mike
You are talking about interference, this effect is visible if the helicopter
is recorded with a camera using ~24fps. So these problems apply
to digital sound.
To prove my previous post see my attachement.
Distortion is 12% 2nd harmonic, other harmonics negligible.
The wave looks very similar to the PM-wave.
I forced that heavy distortions by using low current for vas. (0.4ma)
Gain of the circuit is 1:2500.
Mike
Attachments
PMA, your data sheet is newer than mine. I accidently found my data sheet that was originally sent to me 15 years ago or so. Remembering that YOU liked the AD844, I thought to look more carefully at this data sheet.
To my happy surprise, I found the 18KHz number. From the same file, I also had the AD797 data sheet, so I compared. These are two excellent examples of IC amplifier design. Is there any significant sonic difference between them?
Also, let's not try to merge FM distortion with AM distortion. It doesn't prove much, just like equating Doppler distortion and AM distortion in loudspeakers. They are actually different mechanisms, even if they share certain characteristics.
This is phase 2 of Murphy's law. 'It exists, but is not important'
To my happy surprise, I found the 18KHz number. From the same file, I also had the AD797 data sheet, so I compared. These are two excellent examples of IC amplifier design. Is there any significant sonic difference between them?
Also, let's not try to merge FM distortion with AM distortion. It doesn't prove much, just like equating Doppler distortion and AM distortion in loudspeakers. They are actually different mechanisms, even if they share certain characteristics.
This is phase 2 of Murphy's law. 'It exists, but is not important'
lumanauw,
Here's something that may help you visualize what this PIM stuff is all about. Suppose we have the simple RC low-pass filter as shown in my attachment. Let's make its input a sine wave whose frequency is large enough that we get some phase shift through the circuit. We know that for a sine wave input at a fixed frequency, the phase shift will depend on R and C.
Suppose we replace C with a trimmer cap. If we look at the output of the circuit on a scope, we'll see that we can vary (modulate) the phase of the output sine wave by adjusting the trimmer cap. The amplitude of the sine wave will change too, but let's ignore that and concentrate on the phase. When we talk about modulation, we usually refer to a "carrier" and a "modulating signal". In this case, the carrier is the sine wave, and the "modulating signal" is our hand rotating the trimmer cap to adjust the phase.
Now suppose we replace the trimmer cap with a varactor diode. The varactor diode acts like a capacitor whose capacitance value depends on the voltage across it. Now the output signal will look sinusoidal, but will contain distortion. Since the output voltage is varying with time, so is the capacitance of the varactor diode. Since the capacitance is varying with time, so is the phase shift through the circuit. What is the carrier and the modulating signal in this case? They are one and the same! The carrier is modulating its own phase as it moves up and down, changing the capacitance of the varactor.
Going back to the trimmer cap example, suppose we were able to adjust the capacitance back and forth (clockwise, then counterclockwise) in a completely repeatable way at a rate of once per second. Suppose the frequency of the sine wave is fc. If we looked at the output of the circuit on a spectrum analyzer, we'd see a spike at fc, but also spikes ("sidebands") at fc +/- 1Hz, fc +/- 2Hz, etc. In other words, we see spikes at the carrier, plus and minus integer multiples of the modulation frequency.
Going back to the varactor diode example, we found that the carrier and the modulating signal were one and the same. So on a spectrum analyzer, we would once again see spikes at the carrier, plus and minus integer multiples of the modulating signal. But since the carrier and the modulating signal are the same, the "sidebands" become harmonics, plus a possible DC shift.
So how does this relate to the PIM problem? Well, for a conventional voltage feedback op-amp, the gain-bandwidth product (in radians/second) is gm/Ccomp, where gm is the transconductance of the input diff amp and Ccomp is the Miller compensation capacitor. The bandwidth of an op-amp is just the gain-bandwidth product divided by the gain that's set by the feedback resistor ratio. The problem comes in when you get a time-varying voltage between the inverting and non-inverting inputs of the op-amp (usually pins 2 and 3 if it's a single op-amp in a DIP package). Let's call this the error voltage. As the error voltage varies, the gm of the input diff amp varies. This problem is worst with bipolar input diff amps with no emitter degeneration. Since gm varies as the error voltage varies, the gain-bandwidth product of the op-amp varies too. Therefore the closed-loop bandwidth of the op-amp also varies. For a sine wave of a high enough frequency, this means the phase shift from input to output will also vary with the signal in much the same way as the varactor diode low-pass filter example above.
Here's something that may help you visualize what this PIM stuff is all about. Suppose we have the simple RC low-pass filter as shown in my attachment. Let's make its input a sine wave whose frequency is large enough that we get some phase shift through the circuit. We know that for a sine wave input at a fixed frequency, the phase shift will depend on R and C.
Suppose we replace C with a trimmer cap. If we look at the output of the circuit on a scope, we'll see that we can vary (modulate) the phase of the output sine wave by adjusting the trimmer cap. The amplitude of the sine wave will change too, but let's ignore that and concentrate on the phase. When we talk about modulation, we usually refer to a "carrier" and a "modulating signal". In this case, the carrier is the sine wave, and the "modulating signal" is our hand rotating the trimmer cap to adjust the phase.
Now suppose we replace the trimmer cap with a varactor diode. The varactor diode acts like a capacitor whose capacitance value depends on the voltage across it. Now the output signal will look sinusoidal, but will contain distortion. Since the output voltage is varying with time, so is the capacitance of the varactor diode. Since the capacitance is varying with time, so is the phase shift through the circuit. What is the carrier and the modulating signal in this case? They are one and the same! The carrier is modulating its own phase as it moves up and down, changing the capacitance of the varactor.
Going back to the trimmer cap example, suppose we were able to adjust the capacitance back and forth (clockwise, then counterclockwise) in a completely repeatable way at a rate of once per second. Suppose the frequency of the sine wave is fc. If we looked at the output of the circuit on a spectrum analyzer, we'd see a spike at fc, but also spikes ("sidebands") at fc +/- 1Hz, fc +/- 2Hz, etc. In other words, we see spikes at the carrier, plus and minus integer multiples of the modulation frequency.
Going back to the varactor diode example, we found that the carrier and the modulating signal were one and the same. So on a spectrum analyzer, we would once again see spikes at the carrier, plus and minus integer multiples of the modulating signal. But since the carrier and the modulating signal are the same, the "sidebands" become harmonics, plus a possible DC shift.
So how does this relate to the PIM problem? Well, for a conventional voltage feedback op-amp, the gain-bandwidth product (in radians/second) is gm/Ccomp, where gm is the transconductance of the input diff amp and Ccomp is the Miller compensation capacitor. The bandwidth of an op-amp is just the gain-bandwidth product divided by the gain that's set by the feedback resistor ratio. The problem comes in when you get a time-varying voltage between the inverting and non-inverting inputs of the op-amp (usually pins 2 and 3 if it's a single op-amp in a DIP package). Let's call this the error voltage. As the error voltage varies, the gm of the input diff amp varies. This problem is worst with bipolar input diff amps with no emitter degeneration. Since gm varies as the error voltage varies, the gain-bandwidth product of the op-amp varies too. Therefore the closed-loop bandwidth of the op-amp also varies. For a sine wave of a high enough frequency, this means the phase shift from input to output will also vary with the signal in much the same way as the varactor diode low-pass filter example above.
Attachments
Andy,
great post
Great to see someone here with a solid technical understanding being able to explain things in an easy way and trying to shed light over the matter.
I am pretty tired of the "argument" - all opamps are bad.
- Why?
- Because they use feedback. And - oh well they are just plain and simply bad. Period.
great post
Great to see someone here with a solid technical understanding being able to explain things in an easy way and trying to shed light over the matter.I am pretty tired of the "argument" - all opamps are bad.
- Why?
- Because they use feedback. And - oh well they are just plain and simply bad. Period.
Andy,
Very lucid, thanks for the clear explanation. What I found particularly interesing from a technical point of view is this: If the carrier provides its own modulation (as in the level dependent Gm example), the sidebands are harmonically related to the carrier. Now, if in such I case I look at the output spectrum, can I distinguish harmonics caused by this from harmonics caused by 'normal' non-linear devices? If I cannot, feedback cannot either. Thus, feedback would be equally effective to suppress PIM and FM generated harmonics as the usual ones.
One more point: FM would also generate Fc -Fm, I assume that is the harmonic with a 180 degress phase difference to the sum-harmonic. So we can have in theory the case that this cancels a 'normal' harmonic, at least partly? The mind starts to boggle!
Jan Didden
Very lucid, thanks for the clear explanation. What I found particularly interesing from a technical point of view is this: If the carrier provides its own modulation (as in the level dependent Gm example), the sidebands are harmonically related to the carrier. Now, if in such I case I look at the output spectrum, can I distinguish harmonics caused by this from harmonics caused by 'normal' non-linear devices? If I cannot, feedback cannot either. Thus, feedback would be equally effective to suppress PIM and FM generated harmonics as the usual ones.
One more point: FM would also generate Fc -Fm, I assume that is the harmonic with a 180 degress phase difference to the sum-harmonic. So we can have in theory the case that this cancels a 'normal' harmonic, at least partly? The mind starts to boggle!
Jan Didden
john curl said:
John, I agree, they are excellent IC amplifiers. As I wrote before, I have been doing a listening comparison of several very good opamps. Just me and an experienced Czech reviewer. The AD797 was one of the best among the VFB opamps, but still we were hearing trace of smear on bow instruments and brass instruments. The only opamp able to play it clean, with perfect resolution and without smear was the AD844, fast CFB opamp with flat Zol across most of the audio band.
Folks, may I recommend a 'Google' search of key words for those of you who don't understand a certain concept. I typed 'FM distortion' and got all kinds of results. I even got to HEAR examples of FM Distortion. This saves time and effort of others, (like me).
I do think that we are on the right track, to understand more fully, the hidden compromises in op amps.
I do think that we are on the right track, to understand more fully, the hidden compromises in op amps.
janneman said:
Let me play the devil's advocate here. We know it's possible to test the slew rate of an op-amp with a large-signal high-frequency sine wave. The end result is just a sine wave with a lot of harmonic distortion, right? Shouldn't the feedback have suppressed this?
The problem is that I've exceeded the dynamic range of the feedback mechanism itself. Of course, this is the extreme case.
Hi, AndyC,
Thanks for the explenation. It is very clear.
After reading your explenation, somehow I came up with questions like Janneman. Are these distortions (PM distortion, transistor non-linearities distortion) cannot be eliminated by feedback?
What does feedback do?
Hi, Mike,
Your example shows upper part of sinusoidal is higher than lower part. What is the name of this distortion?
Thanks for the explenation. It is very clear.
After reading your explenation, somehow I came up with questions like Janneman. Are these distortions (PM distortion, transistor non-linearities distortion) cannot be eliminated by feedback?
What does feedback do?
Hi, Mike,
Your example shows upper part of sinusoidal is higher than lower part. What is the name of this distortion?
lumanauw said:
The PIM distortion we've been discussing is part of the feedback system itself and can't be cleanly separated the way other non-linearities can. This is because it relates to the gain-bandwidth product variation with signal level, and the gain-bandwidth product is a "feedback-oriented" parameter. Keep in mind though, that the amount of this distortion can be made extremely low in a well-designed amplifier. Much lower than other similar types of distortion in amplifiers which have no global feedback. One reason we discuss them so much is because of the "By the fanatics, for the fanatics" motto of this site. Also, people who dislike feedback will tend to overemphasize the problems which are unique to feedback systems, so they end up getting discussed a lot.
For instance, the phase distortion example of the resistor and varactor diode happens with common-emitter amplifiers driven from high source impedances, feedback or not. The varactor diode is nothing but a reverse-biased PN junction. This is exactly the same thing as the collector-base junction of a common-emitter amp, whose collector sees the entire output voltage swing of the amplifier. So that's an example of "AM to PM conversion" (what PIM really is) in an amplifier with only local feedback (just its emitter resistor). In terms of how much distortion there is, it's way worse than any PIM from a feedback amplifier, unless the feedback amplifier is very poorly designed (like a 741).
Regarding your question "what does the feedback do?", that question is way too broad to answer in anything less than an entire book
.
Hi, AndyC,
I'm far from as clever as you, but I get your point.
Can I say that VAS in power amplifier is fragile of this distortion? I read that one way to make better VAS is to Voltage-Cascode it, like freezing the VAS transistor in some voltage (2-3V) so it cannot experienced the full swing.
Is this cascoding helps minimize the PM distortion?
I'm far from as clever as you, but I get your point.
Can I say that VAS in power amplifier is fragile of this distortion? I read that one way to make better VAS is to Voltage-Cascode it, like freezing the VAS transistor in some voltage (2-3V) so it cannot experienced the full swing.
Is this cascoding helps minimize the PM distortion?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.