Two weeks ago, I had a conversation with Mr. Kazutaka Tsuda, head of Esoteric, Japan. Nowdays, their CD player seems to have high praise from many different reviews. They have deep research/measurement on mechanism vibration, etc, making their transport is used by other brand like DCS.
In one of their higher class product, they use quite expensive connector. Even the screws are different for different places. I asked him, with so many complicated measurement devices they have, what is the difference between this connector and another connector, what is the measurement difference. The answer quite shock me. They cannot detect a difference in measurement, they pick it by listening test. It seems many things still "unmeasureable" but "listenable", even with current SOTA measurement devices
In one of their higher class product, they use quite expensive connector. Even the screws are different for different places. I asked him, with so many complicated measurement devices they have, what is the difference between this connector and another connector, what is the measurement difference. The answer quite shock me. They cannot detect a difference in measurement, they pick it by listening test. It seems many things still "unmeasureable" but "listenable", even with current SOTA measurement devices
Bob Cordell said:
Bob,
I guess we are all pretty busy and this is not a question of several minutes to analyze.
Anyway, I tried several simulations, not only on PM-A1 amp, but also on more traditional and less traditional feedback designs with both conventional output stage and highly biased AB stage. I came to the same conclusion, global NFB is not a very good cure for CCIF IM f1 + f2 2nd order component. Simulated 3+4kHz, 6+7kHz and 13+14kHz. I would like to emphasize that I have not explored ideal math functional blocks, but rather real amplifiers.
Regards,
Pavel
PMA said:
Thanks for looking at this and thanks for the reminder. This issue is intriguing. If the loop gain is falling with frequency, the beneficial effect of NFB on the f1+f2 component will be less, but I think this is not what you are saying. I think you are saying that the NFB tends to increase the f1+f2 component, regarless of frequency dependence of loop gain. I'll see if I can check this out this week.
Note that the f1+f2 will lie above the audio band for conventional 19+20 kHz CCIF, but that we should not necessarily take comfort in this. We need to allow for the possibility of lower frequencies being involved, where f1+f2 could lie in the middle of the audio band.
Cheers,
Bob
Today we had interesting discussion on our local forum and I have realized that some designers do not understand the fact that input differential voltage / v - v(i) / of amplifier with 1 dominant pole OLG response is 90 degrees phase shifted from output voltage / v(out) /, almost everywhere where openloop gain plot A(w) decreases with -20dB/decade slope. This may cover frequency range like 100Hz - 100kHz, e.g.. I hope it is obvious and I add the illustrating image.
- v(i) / of amplifier with 1 dominant pole OLG response is 90 degrees phase shifted from output voltage / v(out) /, almost everywhere where openloop gain plot A(w) decreases with -20dB/decade slope. This may cover frequency range like 100Hz - 100kHz, e.g.. I hope it is obvious and I add the illustrating image.Attachments
PMA said:Today we had interesting discussion on our local forum and I have realized that some designers do not understand the fact that input differential voltage / v
Yes, this is a very good point that some may miss.
Cheers,
Bob
PMA said:Today we had interesting discussion on our local forum and I have realized that some designers do not understand the fact that input differential voltage / v
I am following this with great interest though not commenting much. But regarding the above: Yes, indeed. Not wanting to sound too wise, but this intrigued me long ago. We glibly go compensating with Cdom, and that is fine as far as it goes - but what about the 90 degrees phase shift? In a way this implies that one is going with 90 degrees less stability margin - OK, a short-cut generalisation, but the point is valid.
That is one reason I do not like op-amps in hi-fi. Oh they do the job, but when I am considering stability and the road is bad enough, I do not also want to be saddled with an obligatory phase shift. My personal preference toward discreets and wide band so that I am free to tailor stability requirements.
PMA said:Today we had interesting discussion on our local forum and I have realized that some designers do not understand the fact that input differential voltage / v
I wonder if the phase shift is why GNFB is audibly less desirable?
It's not quite the same as a constant time delay.
Frankly speaking, I am thinking about the same. But firstly we must consider that it is not the phase shift between output and input signal, but between output voltage and input differential voltage.
For Cdom compensated open loop gain plot, this results in:
- Vdif increases with frequency
- higher the signal slope, higher the Vdif
- crossover distortion residuals, fast and steep, are emhasized in Vdif
Last - as amplitude of Vdif changes with frequency and speed, there is a question of possible phase modulations.
Regards,
Pavel
P.S.: such amplifier has high vector error. We usually assume only non-linear distortion. Is it correct? Music is a transient, time variant signal with instant amplitude values. We try to substitute it by summation of steady state sine waves.
For Cdom compensated open loop gain plot, this results in:
- Vdif increases with frequency
- higher the signal slope, higher the Vdif
- crossover distortion residuals, fast and steep, are emhasized in Vdif
Last - as amplitude of Vdif changes with frequency and speed, there is a question of possible phase modulations.
Regards,
Pavel
P.S.: such amplifier has high vector error. We usually assume only non-linear distortion. Is it correct? Music is a transient, time variant signal with instant amplitude values. We try to substitute it by summation of steady state sine waves.
PMA said:
Moreover Pavel, we must be extremely careful about our wording or risk misundertanding.
It is true the differential signal goes about 90deg. out of phase inside the OpAmp, but of course the global transfer is reasonably flat and phase shift free within the audio band even for mediocre designs.
I understand what you mean is, given the rollercoaster this differential signal is subject to along the internal signal path, then it is vulnerable to low level frequency dependent mutations, and this does really do harm in the form of that seemingly innocent comb of low level harmonics (and attendant IM products in complex signals).
Rodolfo
Now, the error voltage at the output is:
Verror = A(closed loop gain) x Vdif
for image in
http://www.diyaudio.com/forums/showthread.php?postid=1310234#post1310234
it is more than 2V! The vector error. This is 10kHz damped sine burst, 10Vp output amplitude, CL gain = 10. Yes, uA741 + nasty classB output stage. Yes one can imagine better device and better output stage, the error will be reduced, but still exist, and not negligible at all.
Verror = A(closed loop gain) x Vdif
for image in
http://www.diyaudio.com/forums/showthread.php?postid=1310234#post1310234
it is more than 2V! The vector error. This is 10kHz damped sine burst, 10Vp output amplitude, CL gain = 10. Yes, uA741 + nasty classB output stage. Yes one can imagine better device and better output stage, the error will be reduced, but still exist, and not negligible at all.
PMA said:
I finally got around to doing some 19+20 kHz CCIF simulations to see if I could duplicate your observation that the NFB increased the f1+f2 distortion component.
Although I plan to do the experiment around a Class-AB output stage, I first did the experiment around a single BJT CE stage, as pictured below. I ran the stage at 3V peak output under four conditions: no NFB, 10 dB loop gain, 20 dB loop gain, and 30 dB loop gain.
The results were as follows:
In all cases the test tones were at +0.5 dB on the FFT spectrum.
no NFB: -17.4 dB for both 1 khz and 39 kHz
10 dB: -28 dB for both 1 kHz and 39 khz
20 dB: -37 dB for both 1 kHz and 39 kHz
30 dB: -46 dB for both 1 khz and 39 kHz
This is almost textbook behavior, so at least for this basic case I was not able to duplicate your results.
Cheers,
Bob
Attachments
