LM4562 Distortion Measurements

[Samuel Groner]

Hi

Just found this: http://www.dself.dsl.pipex.com/ampins/webbop/LM4562.htm

Heck, common-mode distortion again...

I wonder if this distortion would have shown up if R1 = 100 and R2 = 10k in the schematic shown on page 23 of the datasheet (LM4562.pdf)? Or does the presence of R1 in any case reduce the significant impedance seen by the inputs to its value? [Edit: after a second look it seems as if R1 were bootstrapped and does not reduce impedance--right?]

Samuel

 

[dale116dot7]

That's a problem shared with many op-amps. There is one almost universal solution.... William's Rule. Always invert. The source of this rule is Jim Williams of Linear Tech, and it was published in one of his articles while he was describing the development of an ultra-low distortion oscillator.

-Dale

 

[JohnRoberts]

I would love to see the front end topology for that puppy. I suspect there is a lot of good magic going on.

One thing that that doesn't quite compute, is that Nat Semi needs to crank up the noise gain quite high to see distortion above the noise floor.

So I would speculate the first link is mostly reading signal generator residual or noise. The CMRR of that opamp is quite good (better than some others). Perhaps the good CMRR number looks bad in comparison to remarkable overall non-linearity.

I would be curious to see if there was any noise component associated with those higher impedance tests.

This is perhaps another cautionary note about using any opamp non-inverting as a follower or at very low gain, while the <.005% at 50kHz is not likely to be hugely audible it could be avoided by being more selective with topology.

JR

 

[bcarso]

[quote author="JohnRoberts"]
So I would speculate the first link is mostly reading signal generator residual or noise.
JR[/quote]

Doug acknowledges that, I believe, although he doesn't use the word "generator" per se, for the ultra-low areas. The remark is made for example that the best curve on the first graph in shunt feedback is indeed identical with analyzer (i.e., meaning almost surely overall Ap including generator and notch filter). If it were an external generator I'm sure Self would note such.

So many opamp manufacturers fail to say anything about common-mode distortion that it's almost a Big Lie. Of course it presents an opportunity for some to become known for characterizing and publishing such and doing it right. [EDIT: But I do see that National has provided a lot of CMRR data with frequency, which is a start.]

Even simple common-mode range is often missed---I couldn't find any data on National's power amp, the LM4766, just yesterday, and for confidence am going to have to either characterize them myself or contact someone at the company.

But yes, I agree with Dale and company: if you don't need high input Z go inverting. That story by Williams about the low distortion oscillator, reproduced in one of his collections, is quite instructive.

 

[JohnRoberts]

Even the national CMRR data didn't say much about how they tested it which could perhaps advise a circuit designer about how to best manage it when you can't avoid. If they do say, I didn't work hard enough to find it.

I am not sure whether this is a new problem, an old problem, or just a rocky bottom now revealed by a lower water level. My guess is mostly the latter.

JR

 

[Samuel Groner]

That's a problem shared with many op-amps. There is one almost universal solution... William's Rule. Always invert.

Sure, that's no secret. It's not clear to me why you attribute this to JW though--if I'm not mistaken others have publish this before Linear even existed.

I would love to see the front end topology for that puppy. I suspect there is a lot of good magic going on.

I suspect there is more magic in the output stage. The AD797 (which circuitry is well disclosed and discussed in the literature) reaches (or perhaps even surpasses) the LM4562's input performance with a simple differential pair plus bootstrapped cascode. Making the output stage good while maintaining low quiescent current seems to be more demanding.

Perhaps the good CMRR number looks bad in comparison to remarkable overall non-linearity.

But I do see that National has provided a lot of CMRR data with frequency, which is a start.

High CMRR seems not to be sufficient--it's propably about common-mode path linearity as well. There are opamps which have pretty low CMRR and nonetheless rather good distortion in low-gain noninverting configurations. The AD811 e.g. got just 70 dB CMRR at 10 kHz, but distortion is < 110 dB at a gain of +2 (measured at a relatively low level of 2 Vpp though).

So many opamp manufacturers fail to say anything about common-mode distortion that it's almost a Big Lie. Of course it presents an opportunity for some to become known for characterizing and publishing such and doing it right.

Indeed. Two lines about it would have been two lines. Hopefully the get enough calls about the opamp not reaching its specifications.

The source of this rule is Jim Williams of Linear Tech, and it was published in one of his articles while he was describing the development of an ultra-low distortion oscillator.

That story by Williams about the low distortion oscillator, reproduced in one of his collections, is quite instructive.

So that's just AN43 or something more fancy?

Samuel

 

[JohnRoberts]

[quote author="Samuel Groner"]
High CMRR seems not to be sufficient--it's propably about common-mode path linearity as well. There are opamps which have pretty low CMRR and nonetheless rather good distortion in low-gain noninverting configurations. The AD811 e.g. got just 70 dB CMRR at 10 kHz, but distortion is < 110 dB at a gain of +2 (measured at a relatively low level of 2 Vpp though).

Samuel[/quote]

My question about how the CMRR was measured relates to several different non-ideal factors that could be at play. The distortion identified by Self may not show up as a difference in voltage transfer function between +/- input terminals but may be a nonlinear input impedance which causes a nonlinear input term due to circuit impedance imbalance that then gets amplified. This is consistent with the increased distortion at higher source impedance, and might even null at some combination of circuit impedance, but his reported dip at 2.2k:687 doesn't fit as nicely as if they nulled at equal source impedance.

Just like I prefer to avoid capacitors when not absolutely needed, the same goes for low gain non-inverting stages (Series feedback?) .

JR

PS: I'm not sure what you mean by "Path linearity", so sorry if we're talking about the same thing.

 

[SonsOfThunder]

>>>The effect of 600 Ohm loading is very slight.

I would say in this case "negligible". You could just as easily get similar differences comparing two samples with no load or the same load.

>>>The THD plot is indistinguishable from the analyser output when it has no external load.

Yes, improper terminology IMO, as the "output" is the generator not the analyzer proper. The whole box is so commonly (and really incorrectly) referred to as "analyzer". Proper terminology? "Test Instrument" :grin:

 

[Samuel Groner]

I'm not sure what you mean by "common mode path linearity", so sorry if we're talking about the same thing.

A CMRR plot just tells you how much the fundamental gets rejected but it doesn't tell you how much it gets distorted.

Samuel

 

[bcarso]

[quote author="Samuel Groner"]So that's just AN43 or something more fancy?
Samuel[/quote]

Don't know it as an App note "but I read the book" :grin: