WaltJ said:
Walt,
I assume that this PIM contribution (provided we use opamps
suitable for audio purposes) to resulting sound is often overlayed by opamp output stage imperfection. This is just a result of numerous listening tests on a great number of well known audio components. In case you use a signal cable you must use a buffer, otherwise the sonic result is not good.
janneman said:
Jan,
I am an EE as well and from the strictly engineering point of view you are probably right. I would have been argueing the same way as you 3 years ago in this buffer case. But I have been verifying the case in praxis, as I always do. My 25 years lasting engineering experience has made me humble rather than believing in trivial explanations. There is a real world and the world of models. We have to improve our models and it is a never ending story. In case that the model does not describe the real world exactly I am not going to condemn the real world.
PMA said:
Beg to differ, I think that the trivial explanations are the ones like 'I have no clue why this would be such-and-such, but I feel it is better'. Sounds like a cop-out to me in the face of some hard questions. Especially in the face of our collective knowledge about perception and how utterly untrustworthy that is. Sorry for being so blunt, but that is how I see it.
Jan Didden
janneman said:
No. For me the trivialization looks different. Neglecting HF interference as inaudible, neglecting cross-over AB class distortion of THD 0.05% as inaudible etc. This is from my point of view the chair-designer's attitude. The comparison of different solutions shows that it's wrong.
PMA said:
My personnel perception and that of at least 100 another people in listenning tests, in this case there is no Gaussian distribution
Pavel,
You stated that the buffer is needed because the opamp was thermally and current stressed. When ask a hard question you cop out saying you have no real explanation but 'it sounds good'. Now there is another 100 persons that say 'it sounds good'. So now what, I come back and say I have 200 people stating it sounds bad? Is that the state of serious discussions?
Why is it so damn hard for intelligent people to admit that maybe things are not quite as they thought when seriously discussed?
Jan Didden
Discrete Buffer
Well a good discrete buffer with FET inputs along the lines of Walt's buffer and a AD817 opamp internals does not harm the sound and can easily drive the cable. It sounds better than a BUF634 even in wide bandwidth mode......
Of course just my personal listening impression on Klipschorns, no listening panel present.....
Pavel,
A double terminated 50 Ohm cable is a tough load for an opamp. No wonder the buffer sounds better.
Does you idea with the cable sound better than say a 100k termination at the receiving end?. I presumed reflections are only important for SPDIF or radio frequencies as in TV, satellite and video signals.
Hi Jan,janneman said:
Well a good discrete buffer with FET inputs along the lines of Walt's buffer and a AD817 opamp internals does not harm the sound and can easily drive the cable. It sounds better than a BUF634 even in wide bandwidth mode......
Of course just my personal listening impression on Klipschorns, no listening panel present.....
Pavel,
A double terminated 50 Ohm cable is a tough load for an opamp. No wonder the buffer sounds better.
Does you idea with the cable sound better than say a 100k termination at the receiving end?. I presumed reflections are only important for SPDIF or radio frequencies as in TV, satellite and video signals.
Chair designers
Though I do not sometimes agree with John Curl or Nelson Pass, for example, I appreciate greatly their work and results. Contrary to those that I would call chair-designers, who dispute the phenomenae they are unable to describe in math equations. We use the same math, I guess. The basic problem is that the models do not include enough input variables of the real life circuits. They fit to models only.
Though I do not sometimes agree with John Curl or Nelson Pass, for example, I appreciate greatly their work and results. Contrary to those that I would call chair-designers, who dispute the phenomenae they are unable to describe in math equations. We use the same math, I guess. The basic problem is that the models do not include enough input variables of the real life circuits. They fit to models only.
Pavel, what I was suggesting very subtly was that perhaps a discussion of output stage problems/solutions didn't belong in this thread. It is confusing enough as it is.
Understand that I don't disagree with you as to the seriousness of various distortions possible in an output stage. Perhaps a thread like "Driving problems with IC op amp outputs" might be worthwhile, eh?
Also, see the other post on using video grade parts to drive lines. It works for at least two of us now!
wj
I received an anonymous email to the effect that the above date of "late 1969" was wrong. Sorry, not so, at least for the CA3080, the original OTA. Another useful reference on the CA3080 (which validates my quoted date) is "OTA Obsoletes Op Amp", by C. F. Wheatley and H. A. Wittlinger, NEC Proceedings, Dec. 1969. Reprinted as RCA ST-4159, Dec. 1969. The CA3094 came out later in the 70's, so if anyone wishes to hold my feet to the fire on including it with the CA3080, feel free.
Of course, the *exact* date has nothing at all to do with the technical point that this was published long ago. Hopefully not forgotten.
Walt Jung
Some preconceived ideas from an old fuddy-duddy (me)
Back in this post http://www.diyaudio.com/forums/showthread.php?postid=493105#post493105 I plotted Gilbert's equations for an op-amp operating at a frequency of 1/100 of the gain-bandwidth product at a gain of 10. The phase error as defined in that post and in this one is the phase shift with input stage nonlinearities included minus the phase shift assuming a linear system. Output voltage ranges from 0.1 V peak to 10 V peak in 0.1 Volt steps. That post showed that when the phase error reached 1 degree, the third harmonic was at a level of about 0.5 percent.
Suppose we are looking at a very good op-amp such as the AD797, with a gain-bandwidth product of 100 MHz. The plot of the post referenced above corresponds to a frequency of 1 MHz for the AD797, since the test frequency is 1/100 of the gain-bandwidth product. This begs the question of what happens to the AD797 at a frequency of 20 kHz? I had the equations in MathCad already, and the results checked with Gilbert's for the previous post (that's why I chose this frequency originally, to allow comparison with his results). All I had to do was change the frequency to 1/5000 of the gain-bandwidth product in the MathCad worksheet and re-run it. The results are plotted below. The closed-loop gain value is 10 as before.
It's hard to see the exact value on the graph, but at an output voltage of 2 V peak, the phase modulation is 1.7 * 10-6 degrees. That's not a typo.
Based on this data, I'd conclude that the whole PIM issue as far as op-amps with a very wide gain-bandwidth product (like the AD797) are concerned, can best be described here.
Back in this post http://www.diyaudio.com/forums/showthread.php?postid=493105#post493105 I plotted Gilbert's equations for an op-amp operating at a frequency of 1/100 of the gain-bandwidth product at a gain of 10. The phase error as defined in that post and in this one is the phase shift with input stage nonlinearities included minus the phase shift assuming a linear system. Output voltage ranges from 0.1 V peak to 10 V peak in 0.1 Volt steps. That post showed that when the phase error reached 1 degree, the third harmonic was at a level of about 0.5 percent.
Suppose we are looking at a very good op-amp such as the AD797, with a gain-bandwidth product of 100 MHz. The plot of the post referenced above corresponds to a frequency of 1 MHz for the AD797, since the test frequency is 1/100 of the gain-bandwidth product. This begs the question of what happens to the AD797 at a frequency of 20 kHz? I had the equations in MathCad already, and the results checked with Gilbert's for the previous post (that's why I chose this frequency originally, to allow comparison with his results). All I had to do was change the frequency to 1/5000 of the gain-bandwidth product in the MathCad worksheet and re-run it. The results are plotted below. The closed-loop gain value is 10 as before.
It's hard to see the exact value on the graph, but at an output voltage of 2 V peak, the phase modulation is 1.7 * 10-6 degrees. That's not a typo.
Based on this data, I'd conclude that the whole PIM issue as far as op-amps with a very wide gain-bandwidth product (like the AD797) are concerned, can best be described here.
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.