Bob, interesting reply.
You can tell PIM from TIM WHEN that mathematical relationships of some of the 'IM' products do NOT match an integer algebraic sum or difference of the test tones. Please note that several 'blips' or whatever they are called, do NOT match anything, in Fig. 3 of the 1976 paper, presented at AES-NY. This is because they are PIM, not TIM. Yet, they are the same height as many other 'blips' on the screen, that we are testing for, and expect. Is this to be ignored?
Show Barrie Gilbert, where he went wrong, as well. You can't just handwave that away.
Hi John,
Your comment about the distinction between PIM and TIM is an interesting head-scratcher that some more mathematical types here may be able to answer better than I.
I have two conjectures or perhaps caveats. First, I have always understood that the spectral difference between am modulation and narrow-band fm modulation was that the phase of the upper and lower sidebands was different between the two, and that the amplitude of the spectral lines was essentially the same. That is why I did not think that one can distinguish between TIM and PIM on a spectrum analyzer. I believe that, to a very good approximation, PIM is no different than narrow-band fm; it is just that the phase modulation is the integral of the frequency modulation, which is a linear difference. I will freely admit that we are at the edge of my knowledge here. In any case, I believe that this view does not provide for the production of any "new" spectra by PIM of any significant amplitude.
This view of PIM vs TIM spectra may break down in the case of extremely high values of PIM, but I assume we are not talking about that here.
The second caveat is that the integer algebraic sum of difference approach to looking for expected spectral lines, especially with a complex stimulous like the sine-square, may not account for all spectra created in reality, especially in a complex circuit (or there may be high-order N/M spectra that were not accounted for).
Barrie Gilbert did not go wrong. As far as he went, his results simply do not support your assertions. You and Matti assert that low open-loop bandwidth increases PIM. Barrie did not say anything about this, as far as I recall. That is where you and Matti are wrong.
For a given amount of input stage nonlinearity, low CLOSED LOOP bandwidth increases PIM. Two amplifiers, both with the same closed loop bandwidth, but with different open-loop bandwidth, will tend to exhibit about the same amount of PIM. This is simply because incremental gain variations in the input stage modulate the open-loop gain, which in turn makes the closed loop bandwidth move around, which changes the amount of phase that the closed loop pole puts in the audio band.
Moreover, Barrie did not discuss pre-existing PIM in the open loop of the amplifier that always exists in reality. This element of PIM is usually REDUCED by the application of NFB.
Cheers,
Bob
John, world doesn't start and end with audio. In telecom applications (which most of these products are targeted at) it is the THD+N that characterizes the dynamic behaviour of a circuit. Nobody gives a rat's *** on crossover distortions (or anything alike) if it's buried in noise, noise will take extra bandwidth (or jitter the phase) anyway.
I'll let you explain why it should be different in audio, and why crossover distortions buried at -100dB in noise could be important. Also, to explain what kind of averaging or phase lock-in mechanism your ears have, that allows you to discriminate the crossover distortions from the noise.
It's pretty clear you don't pay much attention to what some of us have to say. We discussed this at length weeks ago (and posted the pictures from the AES preprint, AGAIN).
The extra pins allow Dick to use it open-loop, you said that too, weeks ago.
Here's another one. You can put an external PNP on the AD744 (base to comp pin, emitter to output and collector to V-). This bypasses the lousy verticle PNP in the output stage. We actually used to sell this as a hybrid.
EDIT - I had forgotten that this was the first part with a de-compensation connection. 225kHz at g =1000, probably greatly increased open-loop BW.
Last edited:
Scott, I was just hoping to get you talking in a serious way to Dick Sequerra about this subject of feedback and quadrature. You don't take me very seriously, and this might get you thinking. It also helps others to come to know that many successful designs are created by thinking outside the app note, with the same IC.
For everyone else, now that it is 'obvious' is:
Dick effectively split the AD797, using pin 8 not as originally intended, but as a place to put a reactive load, in order to create an RIAA correction network. The output follower devices are used as a BUFFER. This is brilliant! But why did Dick Sequerra do this with a $25,000 preamp that he sold, a few years ago? Why did he NOT follow Scott's technical instructions? Any ideas, out there? Has anybody been 'paying attention'?
For everyone else, now that it is 'obvious' is:
Dick effectively split the AD797, using pin 8 not as originally intended, but as a place to put a reactive load, in order to create an RIAA correction network. The output follower devices are used as a BUFFER. This is brilliant! But why did Dick Sequerra do this with a $25,000 preamp that he sold, a few years ago? Why did he NOT follow Scott's technical instructions? Any ideas, out there? Has anybody been 'paying attention'
?
Do you have any radio receiver that you can torture?
Compare what you can demodulate before the 1'st preselector, and after the last IF stage.
Got an idea of possibility to hear quietest signals in a loud noise?
Then imagine many narrow-band microphones connected to computer in parallel, each with own 2048 bit A/D converter.
Then imagine a software that highlights the difference between what it has already in the memory and what it hears here and now. It reacts on similarity, and stores the difference, using conscious attention to find a new reaction on the difference.
Such a principle of multi-cell parallel-compensating computer was used in a self learning British robot in 1990'th, before an information about it suddenly and completely disappeared.
Now imagine that you hear sounds that decay, and subconsciously all corresponding narrow band microphones follows this decay increasing sensitivity, while abruptly they hear wider than expected specter. Some people call it "Harsh sound", some people call it "Fuzzy sound", but some people call it "Too sterile sound"
Why "too sterile"? Because such a dirt is generated by amps that are sterile clean according to measurements that measure wrong things.
Last edited:
An externally hosted image should be here but it was not working when we last tested it.
"J" means that a feedback signal is perpendicular to distortions...
Last edited:
Nope, take Barrie's analysis and substitute any one of the linearized input stages that he or the folks at Tek invented (all the way back in the 60's). No GBW modulation no (input induced) PIM same "quadrature" (if you mean the input error signal is 90 degrees out of phase with the output you should just say that).
The multi-tanh technique can make an "exactly" linear gm over a specified range.
I am waiting for some substantial new technical data to discuss.
I would have at least used two and gone fully differential (RIAA) across pin 8's and 3dB noise improvement.
Scott, would this app not have an undefined absolute gain? That was the reason I used the '844 in my amp.
jd
Thanks, Scott. I could not have said it better.
I'm still waiting to hear John's reply in regard to the spectral amplitude difference between PIM and TIM. This one is an honest head-scratcher. I do suspect, by the way, that the presence of PIM may cause some asymmetry in the upper/lower IM sidebands due to cancelation on one side and augmentation on the other side due to the differences in upper/lower sideband phase relationships for TIM and PIM. This is, however, very different than John's assertion regarding the appearance of new spectral lines with PIM that would not be there with TIM alone. What say you?
Cheers,
Bob
At least, you are up to date. For everyone else, this preamp of Dick Sequerras, gave me a real challenge to compete with. I happen to respect the efforts of Dick Sequerra, even if he tends to charge a little extra. Sonically, I am sure that he created a 'winner'. Now, go back to your '2 buck Chuck' Scott. ANYONE knows that better wines are 'overpriced'.
You would need some common mode reference, which is do-able with either feedback to a null pin or splitting some RIAA resistance off to ground. I think the idea is MC only using the input gm as V to I into RIAA as load, the input is an undegenerated NPN pair so linear region is very small.
I'm not up on Volterra analysis, but there is probably an underlying theory there that precludes this. More likely in Fourier theory there is also something. For instance multiplication (modulation) as convolution in the frequency domain would not produce these artifacts.
Scott, the output will be contaminated with common-mode noise
Used open-loop there is no common mode gain from the input (first order) so some split of the
RIAA network should work. I guess the servo would not work.
Last edited:
Not that much, just thinking off the cuff here but I did try it at g = 100 more than 15 yr. ago and the basic idea works fine. It was considered unpublishable as an application.
- Status
- Not open for further replies.