OpAmp Measurement Series

Interesting approach. Nothing beats a real music signal, although something synthetic might be easier to standardize on. Maybe it's possible to analyze a library of recordings and derive a set of 'eigenvalues' representative of a particular musical style, a particular sequence / combination of tone bursts. Regardless, the beauty of digital analysis is, one would not need to create a passive phase-shift network to null the amplifier's phase shifts - it can probably be modeled with some fairly straightforward code. The phase shift profile could be extracted from a DUT with just a frequency sweep.

On the other hand, simpler tests such as THD or IMD still offer the ability to model or study specific phenomena in isolation, which may make it easier to separate benign from objectionable distortions, and gain insight into the causes.

I was thinking along the lines of a gated distortion spectra measurement, wherein the first full sine cycle immediately after the high power "heating" burst is captured and analyzed in comparison with an 'identical' sine cycle applied before the power-dissipating burst. A synchronous trigger system would be needed to mute the high-level burst from the analyzer input, and also to gate the captured signal. But, one sine cycle will surely not offer a sufficiently low noise floor for any meaningful analysis. The measurement would have to be repeated many times over (automatically), and the output signals averaged in order to bring the noise floor down sufficiently. One could also parse out the 2nd, 3rd, 4th... sine cycles after a high-power burst, in order to observe the thermal recovery tail.

As a side note, I often wonder why the weighted-harmonic version(s) of THD never caught on, even though the idea has been around since what... at least the 60's, probably earlier (was Crowhurst the first to suggest it?).

I suspect highly-asymmetrical signals are going to be the most revealing.

I recall hearing of a little box with an input and output that Keith O. Johnson made, that measured well with symmetrical waveforms, but went utterly bonkers on asymmetrical ones. There was also a simple nonlinear analog circuit I saw that detected speech based on detecting asymmetry.

I do agree that thermal distortions, long known to 'scope vertical amp designers but little discussed outside of that often highly proprietary field, are one of the little-explored frontiers in audio. Perrot believes (or his supporters at least believe) that he discovered these effects, which he dubs memory distortion(s), but as is so often the case they predate his encounter and his patents by many years. I studied them in the context of precision research instrumentation in the 70's-early 80's, and I eventually discovered that the 'scope people had done a lot of the work well before that. BTW they found effects, as things got smaller and faster, important at remarkably short time scales, down to the few-microseconds range.

The purpose of these tests is to reveal flaws, so we can fix them. My suggestion of the kitchen sink digital test as essentially a null test which has been used for years by designers, and yes phase shift at both frequency extremes reduces the null quality.

Tweaking the null with digital phase correction could help reveal lesser errors, but any tweaked absolute number needs an asterisk for total accuracy "ignoring phase response".

Listening to this null, easier when phase shift from bandpass extremes are tweaked out, will reveal any thermal tail effects, or the sundry imaginary distortions people argue are missed by conventional test equipment. I'm just looking for a definitive proof that a circuit is free of artifacts, or not. Any artifacts found with this can be isolated by using more conventional test signals, and then design any revealed artifacts out.

JR

PS: Another technique to increase the sensitivity of this null test is to loop the signal through the test path multiple times. This will exaggerate errors.

Of course this depends on digitally recording and playing back this test result precisely over multiple repeats without adding problems from that process. That is another non-trivial task. With a less that universal agreement on requirements of digital recording this technology may be marginal, but that can be proved by the same test. Linear errors from the recording process could be first order cancelled by looping the dry signal the same number of times on another track of the recorder then nulling between those two. This will not be great for characterizing noise floor.

Thanks a lot! That is nice.

Hi folks:

I've spent the last couple of hours going over the book (it's a book, not an article). Samuel, you've done excellent work; I salute you.

A question: Have you looked at the chips' behavior with multitone tests, say 19+19.5kHz, or 19+19.5+9.6kHz? Sometimes those can reveal interesting stuff.

Peace,
Paul

I'd also suggest the multi-tone in natural harmonic series,
which is why I suggested an oboe at super high sample rate.

What do the Opamps do in actual musical instrument use means a lot to me.
But in any case Sam great work. Thanks.

Well done. thank you for sharing, samuel. There are many subtle things to be learned from those plots. the FFTs showing harmonic structure especially.

an interesting addition would be to try to correlate the measured performance with subjective performance via controlled listening tests. ie, does the impressive measuring 797 actually sound better than another IC in a real application in a way that the tests would predict? and what does common mode distortion "sound like"? -- I didnt know it at the time, but Im pretty sure that my own personal testing/listening experiments that ended up focused on high frequency distortion was really all about common mode effects (IMO rising distortion at HF sounds BAD). The horrible measured performance of the 33078 bothers me. this doesn't correlate with the (well known?) subjective qualities of this part. if output loading is avoided these chips can sound very nice, but look at those plots!

mike p

Thanks for the further discussion and interest.

Have you looked at the chips' behavior with multitone tests, say 19+19.5kHz, or 19+19.5+9.6kHz?

Click to expand...

I have run some dual-tone measurements for some devices. Why this is not published has various reasons:
* multi-tone tests do not usually reveal more distortion sources than single tone measurements do (see e.g. a paper by Putzeys: IMD.pdf).
* single-tone distortion measurements allow easy interpretation of distortion residuals (either time or frequency domain), which multi-tone test do not.
* it is relatively tedious to reconfigure the System One from one test to the other. May not seem a big deal at first but at that amount of time needed (we're talking weeks here) things look different.

An interesting addition would be to try to correlate the measured performance with subjective performance via controlled listening tests. ie, does the impressive measuring 797 actually sound better than another IC in a real application in a way that the tests would predict?

Click to expand...

Sure it would be interesting but a "typical application" has so many degrees of freedom that it simply does not look reasonable to find a standardised test; in fact I'm rather more interested in separating distortion contributions and not painting overall pictures. In any case doing statistically relevant double-blind listening test is asking for PhD research time and not a free-on-the-web paper... And an "informal listening session" is definitely below the standards I set for my work.

The horrible measured performance of the MC33078 bothers me.

Click to expand...

I intend to re-run these measurements with a "pre-biased" output stage (i.e. a CCS to one of the rails); looking forward to see the improvement (hopefully).

Samuel

Thanks again for your huge effort here, and there is merit in keeping it simple.

I found two-tone IMD pretty useful back when I was designing phono preamps on '70s-'80s. Simple THD was rolled off by RIAA EQ, while my home made 19-20kHz IMD test generated in band products that received tens of dB more closed loop gain than the higher THD products.

This is obviously an application specific test and I agree 19-20kHz two tone IMD is roughly stressing 40kHz linearity.

A co-worker of mine presented a paper to AES justifying a specific series of 3 tone testing. These are all ways of looking at linearity but IMO simple THD can understate the actual audible contribution in limited GBW circuits.

These are all subtleties. There is no magic single test that reveals everything as you know.

JR

Gerald Stanley favors a IMD multitone test with specific numerological (well, o.k., my euphemism for number-theoretic) properties. I think one of the authors' names is Thiele, but not I believe the same as the speaker parameter guy. And maybe a different spelling too.

I'll try to dig up the references.

I've just started to digest Samuel's paper and all I can say is that I am blown away by his generosity in sharing a paper that undoubtedly was a real labour of love.

One of the coolest things I've found on the 'web in recent weeks - I expect to refer to it for much time to come.

A sincere thank you, Samuel :guinness:

BTW - You've been more than generous already, Samuel, but out of curiosity, did you think of including PSRR figures?

Justin

[quote author="bcarso"]Gerald Stanley favors a IMD multitone test with specific numerological (well, o.k., my euphemism for number-theoretic) properties. I think one of the authors' names is Thiele, but not I believe the same as the speaker parameter guy. And maybe a different spelling too.

I'll try to dig up the references.[/quote]
My multi-tone buddy is Jon Risch, still at Peavey

http://www.geocities.com/jonrisch/PhiSpectral1.htm

I'm not pimping for his or any other oddball test.. they all require understanding and interpretation to be of any value beyond less is better.

FWIW masking of distortion products was not a concern with my crude two-tone IMD. :wink:

JR

PS: Again apologies to Sam for hijacking his thread, and no you don't need to add these tests. You can usually figure out all you need to know from conventional tests, as long as you have a good data set to compare and contrast.

The IM stuff can be found as references here:

http://www.eselab.si/doc/Chapter13_3.pdf

It is the same Theile (and Small as well did an article for AES).

I'm not sure from what book the eselab material is taken---they appear to only have that particular chapter on their site.

Did you think of including PSRR figures?

Click to expand...

It's not entirely clear to me what you mean--just standard AC/DC data copied from the datasheet or distortion measurements with the setup of figure 2.6?

Samuel

[quote author="Samuel Groner"]

Did you think of including PSRR figures?

Click to expand...

It's not entirely clear to me what you mean--just standard AC/DC data copied from the datasheet or distortion measurements with the setup of figure 2.6?

Samuel[/quote]

In the discrete Vreg thread, you suggested that few op-amps offer in excess of 60dB PSRR in the audio band. Considering the OEMs such as National typically quote figures of 100-120dB PSRR, I enquired as to whether you felt it was worth including real world PSRR figures, i.e. those that apply to the audio band.

Either way, you've published a useful resource for many people into audio :thumb:


Justin

Considering the OEMs such as National typically quote figures of 100-120 dB PSRR, I enquired as to whether you felt it was worth including real world PSRR figures, i.e. those that apply to the audio band.

Click to expand...

The 120 dB figures are at DC; essentially all datasheets however show typical curves for AC up to the MHz region.

Samuel

Samuel, did you send this to Walt J yet?

No, not so far.

Samuel

[quote author="bcarso & Samuel Groner"]
Samuel, did you send this to Walt J yet?

No, not so far.

Samuel[/quote]

Well, someone else did, and I find it fascinating reading. What a sense of dejavu this brings. Some preliminary observations:

1) Having received an AES fellowship for some similar IC op amp dist. studies from the 70's, using manual plots from a Sound Tech., I can say that I do hope you get this published in the AES, or somewhere where it sees broad and lasting exposure. I tip my hat to you on this huge effort, Sam. This is good stuff, and important stuff.

2) These families of distortion plots can give a clever circuit designer lots of info as to what to do (or *not* to do) inside. Like for example, use JFETs, but watch out for CM distortion due to nonlinear C, a killer mechanism with high-Z drive.

3) Sam, you might ask around on some of the points. Like "What's an LT1115?". A: essentially an LT1028 that has AC emphasized specs (not precision DC, which are more expensive to test).

4) For perspective, I would have liked to see more discrete circuit test examples, say the 990 circuit.

But all in all, a laudable effort. Don't be heavily influenced by comments about you didn't test my favorite IC, or did you do measurement X, and correlate it to what you show here? What you have done stands well just as it is.

Walt Jung

Thanks very much for your interest and encouragement, Walt. 4) is sceduled for a future revision of the document--currently awaiting the replies for my sample requests. 1) is in my mind as well, I have not settled for a specific possibility yet though.

Samuel