| Gordy |
Many thanks for the pointer.
G. |
|
|
| janneman |
Excellent! Thanks,
Jan Didden |
|
|
| unclejed613 |
| interesting...... i'm doing some power amp output impedance experiments. looks like i've got some reading material on the subject. output impedance is one of the least written about subjects, both for opamps and power amps. |
|
|
| john curl |
| About time you folks got updated about thermal feedback. :geezer: |
|
|
| cliffforrest |
| quote: | Originally posted by john curl
About time you folks got updated about thermal feedback. :geezer: |
Yes. I remember many years ago trying to explain to some otherwise bright people how an early JFET op-amp (LF256?) could appear to have negative gain in a standard test circuit. |
|
|
| wakibaki |
That was worth reading.
w |
|
|
| Christer |
| quote: | Originally posted by john curl
About time you folks got updated about thermal feedback. :geezer: |
Updated? Why? Have they finally eliminated the problem? :)
Edit: The topic was discussed quite extensively in this classic app note almost 35 years ago:
http://www.national.com/an/AN/AN-A.pdf
|
|
|
| john curl |
It doesn't really happen with discrete circuits in the same way, and these sorts of measurements don't show much, BUT Scott Wurcer has assured me that his parts are virtually perfect in this respect (at least, that is the impression I have gotten). Also, look carefully at Fig. 23 and the loaded transfer function of the LM725. This is what I have repeatedly questioned Scott Wurcer about regarding his AD797. Has he fixed the xover glitch as well?
Please Scott, tell me specifically if the thermal and xover problem is completely fixed? If not, how much different from the LM725 example? I am very serious, and somewhat frustrated that I might have to measure it myself, if I can't get any response. |
|
|
| jcx |
Good article but it still doesn't explain why people wanting excellent performance don't just use Walt Jung's advocated buffered multiloop circuits when they want to avoid these probelms
remove the load, remove the heat, remove the thermal feedback from the input op amp altogether
the output device can be Class A for no crossover distortion
the output device can be really fast, have low output impedance to MHz (better RF ingress immunity?)
and for those who know how to design with it the output amp can add loop gain to the overall feedback loop at audio frequencies - even further reducing input op amp distortions by reducing differential signal at its input and Vswing at its (practically unloaded) output |
|
|
| Brett |
| quote: | Originally posted by jcx
Good article but it still doesn't explain why people wanting excellent performance don't just use Walt Jung's advocated buffered multiloop circuits when they want to avoid these probelms |
I must have missed this in my travels. Got a link?
Cheers |
|
|
| syn08 |
| quote: | Originally posted by jcx
Good article but it still doesn't explain why people wanting excellent performance don't just use Walt Jung's advocated buffered multiloop circuits when they want to avoid these probelms
|
What? Like this one?
http://www.diyaudio.com/forums/show...373#post1536373
Naaahhh... Three legs good, eight legs bad :D |
|
|
| jcx |
also my thread:
http://www.diyaudio.com/forums/show...ght=#post512806
discrete buffers can be a lot simpler than that!
cheapest would be resistive biased ef
for low parts count I'd look at the dual njm/jrc4556 with a pull down R for Class A bias if the 8 MHz gbw worked in the particular app without extra compensation components for the input op amp - wouldn't this 40 cent op amp outrage audio snobs paired in a multiloop with a AD797 or OPA627 input op amp
Walt usually shows the AD811 cfa which was good (but more probably because it was available in DIP)
more modern cfa op amp performance simply can't be touched in hobbyist level discrete designs - not too many can close feedback loops around 4 GHz transistors - but we can buy cfa op amps with -90 dB distortion driving 25 Ohms at MHz - I'm pretty sure we wouldn't have to worry about crossover distortion at audio frequencies |
|
|
| Christer |
John, in case you missed my point. I assumed your point was that IC op amps usually do have thermal coupling problems and wanted to point that out to forum member who haven't realized it. However, I found the use of the term "updated" somewhat amusing, since there is nothing to be updated about. If people don't know about this problem they have simply missed what has been well known for many decades, which is why I linked to that app note. I just made a joke about your your phrasing (with no malicious intention). :)
Then, as you point out, progress seems to have been made in decreasing the thermal coupling problem and perhaps some modern op amps don't suffer from it at all. |
|
|
| john curl |
| I work to point out such factors in electronic design. This is one that I didn't get around to discuss recently, but I read Solomon's article in 1974, when it was in the 'IEEE' transactions. I also had the test equipment made by Tektronix at the time to directly test for the effect of thermal feedback and xover distortion, and I ran lots of tests. However, my own designs did not have this problem, and I found the measurement was not useful for discrete circuits. |
|
|
| Christer |
| quote: | Originally posted by john curl
I work to point out such factors in electronic design. This is one that I didn't get around to discuss recently, but I read Solomon's article in 1974 or so when it was in the 'IEEE' transactions. |
And you should point it out, since most forum members probably aren't aware of it. As I said, I just made a benign joke about the way you phrased it.
The effect is hardly surprising since all amp stages share one small single piece of silicon. |
|
|
| unclejed613 |
| quote: | Originally posted by Christer
Updated? Why? Have they finally eliminated the problem? :)
Edit: The topic was discussed quite extensively in this classic app note almost 35 years ago:
http://www.national.com/an/AN/AN-A.pdf
|
a "must-read", definitely. good app note. i have one of the "orange and blue" NS books. even tells how to minimize thermal distortion in op amps. it seems not a lot of designers have read it... a good app note with information useful in power amp design as well.
yet there are still op amps and power amp chips with the same thermal problems, and they're relatively recent designs.....
too bad those of us working on discrete amps dont have the availability of multi-collector and multi-emitter transistors....... |
|
|
| jackinnj |
BP was kind enough to omit competitor's devices.
I can imagine the folks at NSC trolling around people's desks looking for the oldest versions of the LM741 they could find. |
|
|
| janneman |
Then again, with opamps like the LM4562 at 0.025ppm non-linearity available, can't we say we licked it for practical purposes?
Discrete circuits also have thermal non-linearities; even a few degrees temp difference inside an amp case (and often its more than that) makes the bias drift here and yonder. Admittedly, it's a lower (much) time constant, but again, are we supposed to go to discrete to overcome 0.025ppm? And can we at all?
Jan Didden |
|
|
| syn08 |
| quote: | Originally posted by janneman
Then again, with opamps like the LM4562 at 0.025ppm non-linearity available, can't we say we licked it for practical purposes?
Discrete circuits also have thermal non-linearities; even a few degrees temp difference inside an amp case (and often its more than that) makes the bias drift here and yonder. Admittedly, it's a lower (much) time constant, but again, are we supposed to go to discrete to overcome 0.025ppm? And can we at all?
Jan Didden |
Yes we can, but so can we choose to go for the ancient greek language and culture instead of e.g. french. It might be interesting and entertaining, but to expensive and ultimately useless for any practical purposes. Discretes are soon going to be only a die hard DIYer and simulation afficionados thing. |
|
|
| unclejed613 |
| pioneer, denon, yamaha, harmon kardon, and others still use discretes. granted, some of them look like an amp design from the early 70's (resistors instead of current sources, no current mirrors, darlington output transistors) but discretes are still in use on a massive industrial scale. class D makes for nice "drop-in" amps, and many manufacturers are using large chip amps (STK430-xxx), but there are still plenty of discrete amps being rolled out every year. |
|
|
| myhrrhleine |
| quote: | Originally posted by jcx
Good article but it still doesn't explain why people wanting excellent performance don't just use Walt Jung's advocated buffered multiloop circuits when they want to avoid these probelms
remove the load, remove the heat, remove the thermal feedback from the input op amp altogether
the output device can be Class A for no crossover distortion
the output device can be really fast, have low output impedance to MHz (better RF ingress immunity?)
and for those who know how to design with it the output amp can add loop gain to the overall feedback loop at audio frequencies - even further reducing input op amp distortions by reducing differential signal at its input and Vswing at its (practically unloaded) output |
You might check out Bateman's capacitor distortion articles.
he plays with a few combinations and looks a thd around -120db |
|
|
| scott wurcer |
| quote: | Originally posted by john curl
It doesn't really happen with discrete circuits in the same way, and these sorts of measurements don't show much, BUT Scott Wurcer has assured me that his parts are virtually perfect in this respect (at least, that is the impression I have gotten). Also, look carefully at Fig. 23 and the loaded transfer function of the LM725. This is what I have repeatedly questioned Scott Wurcer about regarding his AD797. Has he fixed the xover glitch as well?
Please Scott, tell me specifically if the thermal and xover problem is completely fixed? If not, how much different from the LM725 example? I am very serious, and somewhat frustrated that I might have to measure it myself, if I can't get any response. |
Pease writes an article with references to the uA709 you ask about the LM725. You folks are going to have to let go of some of this 40yr old stuff and move forward. Paul Brokaw and Mitch Madique presented cross quaded input devices and outputs on a thermal center line with NO thermal feedback (fractional ppm) more than thirty years ago, it was then end of story. |
|
|
| syn08 |
| quote: | Originally posted by scott wurcer
Madique |
Maidique? |
|
|
| scott wurcer |
Sorry for my lack of precision.
A high precision monolithic super-beta operational amplifier
Maidique, M.A.
Solid-State Circuits, IEEE Journal of
Volume 7, Issue 6, Dec 1972 Page(s): 480 - 487
Summary: A single-chip (67/spl times/90 mil) integrated-circuit operational amplifier using thin-film resistors and super-gain transistors has been designed to achieve dc follower accuracies of 0.001 percent with 100-k/spl Omega/ source resistance. The circuit achieves gains of 140 dB using thermally balanced layout designs for both input and output stages, nulled drifts of 0.3 /spl mu/V//spl deg/C, and offset currents well under 1 nA. All other dc specifications including power-supply variation error (PSRR), common-mode gain error (CMRR), etc., are in the 1-10 ppm error range; and a procedure is given by which long-term drifts of less than 10 /spl mu/V/month can be assured. AC performance is comparable to general-purpose integrated-circuit operational amplifiers, i.e., f/SUB t/=300 kHz and slew rate of 1.2 V//spl mu/s at gain of ten. The circuit is externally compensated for unity gain with a single 390-pF capacitor and is fully input and output protected
EDIT - Excuse the special character snafus, but the abstract gets across the point Solomon pointed out a problem, it was easily solved, end of story. |
|
|
| syn08 |
| quote: | Originally posted by scott wurcer
Sorry for my lack of precision.
A high precision monolithic super-beta operational amplifier
Maidique, M.A.
Solid-State Circuits, IEEE Journal of
Volume 7, Issue 6, Dec 1972 Page(s): 480 - 487
|
Yep, I thought it's your co-founder. |
|
|
| john curl |
| How about the xover distortion, Scott? The LM725 seems to be OK with thermal feedback, but it still has that zero crossing bump. What about the AD797? |
|
|
| scott wurcer |
| quote: | Originally posted by syn08
Yep, I thought it's your co-founder. |
To be accurate co-founder of Nova Devices with a bunch of ex-Transitron people. Bought out by ADI around 1973. An ND709 might be one of the rarest IC's you could find. |
|
|
| syn08 |
| quote: | Originally posted by scott wurcer
To be accurate co-founder of Nova Devices with a bunch of ex-Transitron people. Bought out by ADI around 1973. An ND709 might be one of the rarest IC's you could find. |
ND was the first semiconductor fab for AD, isn't it? |
|
|
| jackinnj |
One of the opamps on my prototyping board -- the LT1013 of which I have like popcorn, and the LM4562 -- both with a 1K load.
 |
|
|
| lumanauw |
Hi, jackinnj,
do you have such a graph for OP27? |
|
|
| jcx |
| quote: | Originally posted by myhrrhleine
You might check out Bateman's capacitor distortion articles.
he plays with a few combinations and looks a thd around -120db |
My recollection is that Bateman used a "active feedback" Reddy style composite amplifier in his oscillator, these are entirely dependant on op amp matching for any improvement - I think Soliman had articles on active op amp compensation in Wireless World in the 80's
the topology can work to some degree as Bateman shows, more ususally this topology is chosen for reduced phase errors in active filters
I'd claim using "active compensation" topologies for output load distortion reduction is less reliable - for best results you'd have to be sure that you loaded the feedback path op amp exactly the same as the forward path op amp
op amp parameter matching in these circuits usally is assummed when using dual op amps, but finding data on the degree of parameter matching of the 2 op amps in a monolithic dual is rare
for more fun you could try to predict thermal crosstalk effects when using both sides of a dual in a single loop
Jung style multiloops should give more robust reduction of output loading distortion reduction
as I mentioned in my thread on composite op amp circuits I've built high loop gain multiloop circuits and couldn't see 1 KHz 2nd order IMD product in a setup with ~ -160 dB resolution limit |
|
|
| jackinnj |
Sorry, no OP27 -- i have the jig set up for duals. For some reason I don't have many ADI duals.
I do have a bunch of OPA2134's from TI. |
|
|
| scott wurcer |
| quote: | Originally posted by syn08
ND was the first semiconductor fab for AD, isn't it? |
Yes, though we had some pans in the fire at Micropower. I don't remember if Nova got seed money from ADI. My high school chemistry teacher's son was Nova's first "knob tweaker" in the fab (I found out years later that they were related).
Jackinnj - that second plot is what I expect these days. |
|
|
| jackinnj |
| quote: | Originally posted by scott wurcer
Jackinnj - that second plot is what I expect these days. |
My wavetek sine/triangle generator doesn't have precise control of the DC offset -- |
|
|
| john curl |
| Scott, didn't Bateman use your oscillator design without giving you any credit? |
|
|
| scott wurcer |
| Yes, I'm still interested to know if anyone has gotten that LTC "no" distortion oscillator to work. An oscillator with a distortion trim and claim of no measurable distortion at the same time seems contradictory. |
|
|
| jcx |
Looks like I must have misremembered, need to find my "Capacitor Sound" folder
I still have a strong recollection of seeing the Reddy/Soliman acitve feedback circuit recommended for low distortion amplification - where? |
|
|
| jackinnj |
| quote: | Originally posted by scott wurcer
Yes, I'm still interested to know if anyone has gotten that LTC "no" distortion oscillator to work. |
Oscillators to ham radio guys are like 6 cylinder Ford engines. You just can't stop tinkering with them:
You mean this?
The fundamental isn't entirely notched out by my Tektronix THD analyzer.
 |
|
|
| scott wurcer |
| And you didn't even have to use Vishay resistors. |
|
|
| john curl |
| Who said that Vishay ,'bulk metal' resistors are best for audio? NOT me. I tend to use Vishay (now) Roederstein that cost me 0.05-0.10 dollars each. They don't have any measurable distortion either, but their tempco is not as good as 'bulk metal'. Cheap shots, even as a joke, are misleading to others, Scott. |
|
|
| john curl |
| Jackinnj, can you measure an AD797 or an LT1028? It would answer a number of questions that I can't get otherwise. |
|
|
| scott wurcer |
| An interesting definition of cheap shot. I guess distortion follows the model of homeopathy the less of it but still existant increases the detrimental effect. |
|
|
| Christer |
| quote: | Originally posted by scott wurcer
Paul Brokaw and Mitch Madique presented cross quaded input devices and outputs on a thermal center line with NO thermal feedback (fractional ppm) more than thirty years ago, it was then end of story. |
Yes, but the interesting question is, are these methods sufficient (that seems implied in your answer) and are they standard methods used in (almost) all opamps since long, or used only in som high-performance ones? |
|
|
| unclejed613 |
| reminds me of "the law of decreasing returns". after a certain point, halving the distortion residual has exponential effects on the cost. the key is finding the optimum breakaway point and being satisfied with it. |
|
|
| jcx |
My AD797 simplified circuit sim
http://www.diyaudio.com/forums/show...757#post1525757
modified to match the conditions of Scott's AES paper and compared with his pic (from the previous page of that thread)
some scale differences but remarkable congruence of waveform features temps me to believe the sim can “explain” the bumps at zero (output current) crossing
Av 1000 amp, input to gnd, +/-10 mA triangle ramp pulling on the output (output V =green trace)
top with full 50 pF internal comp limiting high frequency gain
bottom with Scott's “distortion canceling” Cn (increased Vres to see any features)
1st we can dismiss John's characterization of this as crossover distortion, simply run the output stage Class A by changing the output current ramp to +/- 1mA which keeps within the push-pull Class A operating range of the 500 uA output Q bias (to me using the term "crossover distortion" unmodified implies a static nonlinearity)
all John has to do is buffer the output to get this level of linearity, keeping the AD797 in its Class A operating range
again the top trace is without Scott's recommended Cn, lower with Cn
2nd we can see that at 16 Khz Scott's test is pretty fast, so what does the sim show at 100x slower?"
note the V scale is 8 mV full scale, compare to over 100 mVpp error waveform amplitude in the 1 plot without Cn
a little slope change is evident without the Cn cap in the top sim, the bottom looks as linear as could be desired
I think these sims are enough to finger the dynamic charging requirement of full output transistor commutation when running the output devices Class AB as the real “crossover hat” culprit in Scott's AES pics
perhaps external Class A bias would also achive higher high frequency linearity by avoiding the output Q commutation - I would rather use a few more components to make a buffer in the feedback loop after Jung and not have to consider the skewed bias conditions inside the chip and also totally ignore thermal feedback |
|
|
| john curl |
| Thanks for the analysis. Just what I wanted. |
|
|
| scott wurcer |
| Excellent jcx, thanks from me too. |
|
|
| scott wurcer |
| quote: | Originally posted by Christer
Yes, but the interesting question is, are these methods sufficient (that seems implied in your answer) and are they standard methods used in (almost) all opamps since long, or used only in som high-performance ones? |
I don't mean to appear high-handed. The "gull-wing" effect shown in Solomon's paper is eliminated by these methods. There is certainly a shift in bias point with temperature and there might be some subtle second and third order effects. Cross-quading of input devices and thermally centering the output devices is a pretty easy thing to do and most designers are aware of this. There are many amplifiers around with ruler flat gain curves into 1k loads. |
|
|
| john curl |
| Scott, my discrete Levinson JC-2 modules looked like a straight line back in 1974, with the same test. I am glad that the problem is now often fixed, but you must agree that it was not often done very well, until years after Solomon's paper, back in the early 70's. |
|
|
| Christer |
I get a feeling that some people may perhaps have misunderstood me, which is then probably my own fault. I never meant to make any claim about the state of affairs today and appreciate whatever Scott and others can tell about it.
The reason for my original post was that John told people to get updated about thermal feedback. I took for granted that his intention was to point out to people that IC op amps do have thermal issues. Perhaps it was me missing some subtle nuance of the english language, but I found the word "updated" a bit strange. Whether thermal problems still exist or not, they did exist long ago, which is why I linked to that old NS app note. If they still exist I don't think there is anything to be updated about, since it has already been known for so long. There is no news here, hence no need for updating. Many have surely missed the issue entirely, on the other hand. Anyway, my comment to John was just meant as a joke, that if there is a need to get updated something must have happend, and since thermal feedback used to be a problem it must then be gone. Perhaps I just stumbled into a semantic swamp here, and if so, sorry for the confusion it may have caused. I often tend to think in slightly different and more abstract ways than most people, which is sometimes useful and sometimes confusing to others. |
|
|
| Wavebourn |
:xeye: :xeye: :xeye:
Now my favorite LM4562 will be unobtanium. :( |
|
|
| jackinnj |
| quote: | Originally posted by Wavebourn
Now my favorite LM4562 will be unobtanium. :( |
I think that the plan is to relabel them :)
I think that you should also at the LT1115. |
|
|
| unclejed613 |
unobtainium? you can make many things with it, ideal op amps, perfect current and voltage sources, MOSFETS with Rds(on) of 0 ohms, bipolar output devices with zero Cbc,...the list is endless, literally..... :D
sorry, couldn't resist.....
it's too bad when devices that are extremely useful go away. there used to be a CMOS device that was a white noise source, and the package was a DIP-8. it was easy to add a pinking filter. a complete white/pink/red noise source could be contained in a box of a few cubic inches. alas the device has long been "unobtainium". there is probably something new that can be used, but i haven't seen it yet. i know it can be done with a DSP chip, since many new surround receivers have pink noise sources available for system setup and auto-EQ.
sometimes manufacturers come out with new improved versions of devices and are kind enough to make them pin-compatible with the old device, so they can be used as drop-in replacements. be careful of these, because sometimes what were originally unused or N/C pins may now have an input or output tied to them. i worked on some equipment that had Dallas real time clock modules, and after the equipment they were in had been in use for 10 years, the internal lithium batteries were going dead. i tried to find these modules, and they had been discontinued. but i soon found that Dallas had a pin-compatible replacement. i bought a few of the new modules, and they didn't work. i looked at data sheets for both old and new modules and found they were almost identical, with two exceptions, Dallas had added an /inh input and another input on what used to be unused pins. the /inh was tied to ground at the socket. after making a note to cut off the /inh pin, and cutting the pin off on my test units, i tried them again and they worked. |
|
|
| Conrad Hoffman |
| Yeah, those National DIP-8 noise sources (MM5437N) were really handy back when I was testing speakers in the pre-computer days. I still have a couple stashed away somewhere. OTOH, you can make a much wider band and more random noise source with the right common zener or small signal transistor. Just reverse bias the base-emitter junction so it zeners and amplify the resulting noise with a JFET follower. |
|
|
| jackinnj |
| quote: | Originally posted by unclejed613
sometimes manufacturers come out with new improved versions of devices and are kind enough to make them pin-compatible with the old device, so they can be used as drop-in replacements. be careful of these, because sometimes what were originally unused or N/C pins may now have an input or output tied to them. i worked on some equipment that had Dallas real time clock modules, and after the equipment they were in had been in use for 10 years, the internal lithium batteries were going dead. i tried to find these modules, and they had been discontinued. but i soon found that Dallas had a pin-compatible replacement. i bought a few of the new modules, and they didn't work. i looked at data sheets for both old and new modules and found they were almost identical, with two exceptions, Dallas had added an /inh input and another input on what used to be unused pins. the /inh was tied to ground at the socket. after making a note to cut off the /inh pin, and cutting the pin off on my test units, i tried them again and they worked. |
Maxim is infamous for this. The company's gpm is g-eye-normous but even this lowly experimenter has seen stuff inexplicably disappear from their repetoire. The company's corporate ethic is such that the SEC and NASD won't let them keep their stock listed on the big-boy exchanges.
Some of the nice comlinear chips disappeared.
And who can forget those great Motorola floating regulators???
Getting back to the point -- I replicated BP's experimental method with my Tek 2465 (which is the scope he used) and my TDS3012 -- while I am enamored of the 2465, the TDS3102B makes it easier to dim the noise at uV levels with its averaging function.
You should really look at the Linear Tech LT1115. |
|
|
|
