The graph in question was made with an HP3581 wave analyzer with its corresponding HP plotter, owned and maintained by a Finnish Government Laboratory. It is usually operated by a trained technician who did 100's of measurements and at least 50 for me, personally, with this specific TIM test. I came home in 1976 with a briefcase full of these measurements, made on full sized, similar normal writing, paper.
The idea that this plot could arbitrarily be off 50Hz or so, relative to its MEASURED AND DENOTED adjacent frequencies is almost impossible. Since, such a deviation is necessary to 'explain' the FM hypothesis away, who came up with this 'conclusion'?
The idea that this plot could arbitrarily be off 50Hz or so, relative to its MEASURED AND DENOTED adjacent frequencies is almost impossible. Since, such a deviation is necessary to 'explain' the FM hypothesis away, who came up with this 'conclusion'?
You're digging a very large hole for yourself. The non-harmonics are not possible and Ron will not find any. Did you read the patent? The plot was off by a small fraction probably because it was copied, all tones are accounted for by A*F1+-B*F2 a list was posted long ago. BTW it was a magnitude plot so all phase information was destroyed therefore the AIM and PIM can not be separated. FM sidebands are all at A*F1+-B*F2 same as the AM ones.
Last edited:
Long ago I believe I posted a possible cause would be the lateral PNP transistor in a 741 opamp could saturate and cause such a result depending on level and time (1/frequency). So I can believe the graphs are correct, but anyone listening under those conditions would also find a non subtle distortion.
Repeating the results with a sampling of amplifiers would be more interesting. At the moment I am overbooked.
ES
Ed we're talking about causality, this is physically impossible unless the amplifier is actually oscillating creating alien frequencies, but then there would be a huge spray of junk.
I think andy_c, Jan, and I wasted some time to finally settle this. I took the plot and counted pixels in photoshop. I think Jan posted the frequencies, some involved 9th or 11th harmonic mixing and were very close actually.
the only difference is that with "AM only" distortion the A,B are integers that add up to the order of the nonlinearity - square law nonlinearity gives only 2 IMD product lines
with "FM" or "PIM" intermodulation product the IM products are at "all" integer multiples - which at first appears alarming
but the amplitudes of the higher order IM products declines by a power series (well a Bessel series for high level nonlinearities)
for "AM" IMD with 1 KHz, 10KHz test tone, square law nonlinearity you just get 9,11 KHz IM products
for "FM", "PIM" there are additional IM lines at all integer combinations - but if the level of the distortion of the 9,11 KHz components were -60 dB then the 8,12 KHz would be ~-120 dB, 7,13 KHz ~-180 dB...
so for a mostly linear open loop amp (the only responsible starting point for application of negative feedback) the "extra" IMD products amplitudes are so low that practically it should be hard to see higher order "FM" IMD product spectral components
and John really should give up the "Big Lie" technique with respect to Gilbert's "Are Op Amps Really Linear" article - it explicitly treats low bias, non degenerated inputs with signal 1/10 the GBW - very like using the original 741 for audio
any monolithic op amp recommended today for audio beats Gilbert's pedagogical example by orders of magnitude using a combination of degeneration, fet inputs, alternative compensation and simple speed
It is perfectly consistent for Gilbert to dismiss the importance of Otala’s analysis for (competently) purpose designed discrete audio power amps and still present the material in the article to educate Analog’s customers on op amp limitations/misapplication - I have had to redesign a cookbook "engineer"'s circuit which was using a TL072 with 5 MHz carrier signal
Last edited:
Ed, even with the 741, a 5V pk-pk or 1.8V rms out, with a 30KHz filter at the input, is not really impossible for the 741 to be able to stay out of clipping. It is just a representative example of 'reasonable worst case' that might be expected from a uA741 used in audio. I know that IF severely aberrant behavior were present in the graph, we would have just used a lower output pk-pk, voltage such as 3V pk-pk or 1V rms graph. We did plots over the whole range. The distortions grow like 'grass' between the test signals. How high the grass is, depends on level, UNTIL slew rate limiting or clipping is reached. 5V pk-pk is just a representative sample, showing a potential problem with the 741 used in audio, as it was at the time, and still is today, by the way, in slightly different form.
If I can I'll look for my notes, but I thought the possible mechanism is the apparent effect of "stored charge" in a lateral PNP causing an increase in the base width changing one of the time/gain factors which MIGHT seem to stretch a superimposed waveform. But Scott it has been a very long time since I played with semiconductor physics, I think the possible issues are more your ballgame. I don't think IM is the explanation.
I don't doubt the traces, just what is the cause.
OK I'll bow out now and wait and see. It HAS to be 2f2-8f1, aren't you listening to me or jcx or others? So where is the 2f2-8f1 component?
FM signal can be represented as:-
v = ac sin(wct + m sin wmt )
The frequency spectrum can be found by rewriting the above expression as a sum of components of constant frequency using the properties of the Bessel Functions. This gives:-
v = ac{Jo (d) sin(wct)
+ J1 (d)[sin(wc + wm)t - sin( wc - wm)t]
+ J2 (d)[sin(wc + 2wm)t + sin( wc - 2wm)t]
+ J3 (d)[sin(wc + 3wm)t - sin( wc - 3wm)t]
+ ...
Scott
I have trouble understanding why it would be so large if it were caused by higher order harmonics.
ES
John,
Since you brought up, again, the beaten-to-death horse you KNOW is wrong, does that mean you have no answer to:
/quote
I hope John would clarify what he thinks about the [Quan] method, whether it uncovers hithero unknown phenomena or whether it provides a new view into known shortcomings (which can be very valuable to be sure).
/unquote
?
jan didden
Last edited:
Jan, I really don't appreciate being called a liar. I just rechecked my calculations AND the graph again, just this morning. This must be the third time. The 2f2-8f1 frequency SHOULD BE: 4.56 KHZ, but it is not there. Why should it be there? IF I am incorrect in my estimate of 2f2-8f1, please give me the proper number and how you calculated it.
John I'm not calling you a liar, but why start again on this route, while we were discussion Quan's paper. You repeatedly said you believed Quan 'was on to something' or words to that effect. Pray tell, what DID he find what we don't know already? What DID he find that will help us to build better amps, and/or to better understand how amps work?
jan didden
Ron Quan and Bob Cordell have had significant correspondence since the AES. Perhaps, many of your questions should be directed to Bob. I will stand by my position on the graph in the TIM paper, presuming that FM sidebands are being mistaken for higher order IM sidebands. This is a pesky problem like the perihelial movement of Mercury was to the Newtonian and the general theory of relativity. The ANSWER is in the details. If the frequencies don't fit, then another explanation is necessary.
ahem, engineers don't aruge endlessly over questionable data...
...they just repeat the test, and with better equipemnt
good data is so much more convincing - which amps show this today? - just putting up the data file from a "pro-sumer" soundcard measurement would allow independent analysis
any of something popular like ESP project amps likely to show these artifacts? – anyone have one and a ESI Julia@ or better soundcard? 192K fast enough?
...they just repeat the test, and with better equipemnt
good data is so much more convincing - which amps show this today? - just putting up the data file from a "pro-sumer" soundcard measurement would allow independent analysis
any of something popular like ESP project amps likely to show these artifacts? – anyone have one and a ESI Julia@ or better soundcard? 192K fast enough?
Last edited:
- Status
- Not open for further replies.