Common-Mode Distortion In OpAmps

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

Samuel Groner

Well-known member
Joined
Aug 19, 2004
Messages
2,940
Location
Zürich, Switzerland
Hi

I wondered about the phenomenon "common-mode distortion" a long time and did some research now. Not everything clear yet, so let my ask a few things.

First to the mechanism:
* in noninverting mode, there is a common-mode voltage at the inputs present which is proportional to the input voltage and inverse proportional to the configured gain. Due to finite CMRR a little of this voltage adds up to the differential mode signal; as the CM path is highly nonlinear, a distorted version of the converted signal appears at the output. Is the way I put this correct?

Now:
* Web_Ch6_final_I.pdf, page 6.52/6.53, mentiones that balancing the source impedances for inverting/noninverting input reduces CM distortion. What is the reason for this? Does CMRR improve with balanced source impedances? (I do not have access to ref. 8 in the above PDF)
* www.dself.dsl.pipex.com/ampins/webbop/2134.htm, bottom of page, writes that choosing higher supply rails can reduce CM distortion. This is probably related to the fact that we are further away from exceeding the CM input range with more supply volts. However, with 10 dB gain and 5 Vrms out, I would have thought that we are far away from exceeding CM range anyway. Any ideas?
* 990.pdf, page 6, shows a way to cancel input bias currents; does this method affect CM distortion? I does surely reduce CM input impedance.

Thanks for you replies!

Samuel
 
Sam,

This is my major interest now, I can't give you any specific answers, but
yes, it is a major flaw in a lot of amp designs, and needs to be carefully assesed in any serious opamp implementation.

Bias current compensation, as a theory doesn't help or hurt, but it's implementation does. The LT-1037 for instance sounds like shit, but has impressive as hell specs. Mostly due, I think to it's compensation circuitry.
 
It's a lengthy subject. One word to the wise---pay attention to the modulation of the capacitances, and as well the modulation of the input device dissipations, especially if the latter changes aren't essentially equal for a given pair of devices. It's not enough for them to merely be adjacent on a substrate, although it helps.
 
You should also minimize the impedances. What they have proven is that, after selecting 1K8||1K8 feedback (no apparent reason), 900 is an optimum for the other input. Since we ultimately drive 650 ohms, we might have considered 1K||1K, 500 ohms is now the optimum dummy resistor.

> probably related to the fact that we are further away from exceeding the CM input range with more supply volts.

Read the pages again. In this case they cite capacitance to the substrate, the body of the chip, which (used to be) the negative pin. Junction capacitance is roughly inverse of cube-root. So at a fixed signal level, a higher supply voltage gives a smaller relative change capacitance, but also a smaller absolute capacitance.

While capacitance is a big factor for small-medium signals, with FETs and high excursions you face the fact that a JFET's plate resistance may be 300K at 15V, 270K at 10V, 250K at 5V, and then 100K at 4V and 10K at 3V. If you don't stay well above several volts (pinchoff voltage) the JFET works more like a soft triode than a hard pentode. It may still meet its minimum specified gain and bandwidth, but the change of gain as the channel comes out of saturation is audible distortion. That will not be cured by equal or low input impedances, only by staying away from low FET bias. And ten again, you can cascode an FET so it is always on the edge of its triode region with little variation; and some hot FETs never really get a hard pentode shape.

> 990.pdf, page 6, shows a way to cancel input bias currents; does this method affect CM distortion?

There are some super-clever, and really awful, bias cancellation schemes out there. This one is benign. It does reduce impedance and changes gain, but less than 1% worst-case. It injects supply noise but never enough to matter. It is not auto-correcting, you have to trim it for every LM394 you buy. Its main virtue here is lower DC current in T1, for minimum noise and maximum inductance.

I think each case has to be considered on its own demerits.

Note that we have been using opamps for 60 years, this type of effect is clearly noted in the 1970s, but has suddenly become a "problem". Partly because older amps had worse problems. Partly because BJTs tend to suffer less, and tend to encourage low/equal resistances more than FETs. Low-performance FETs such as on TL071 don't have such big capacitance (or variation of capacitance) as the improved FETs on newer amps. So as an old grump, I'm inclined to put this "problem" low on the list.
 
[quote author="PRR"]..snip...Low-performance FETs such as on TL071 don't have such big capacitance (or variation of capacitance) as the improved FETs on newer amps. So as an old grump, I'm inclined to put this "problem" low on the list.[/quote]

Nonetheless, per a previously related old war story from one of my first "new-guy-proves-himself" experiences at Harm*n, my boss scoffed when I attributed some out-of-spec distortion in a TL072 Sallen-Key lowpass to capacitance modulation in the voltage follower. They were switching the feedback and shunt capacitors in and out of the 3 pole low pass, which had three series 18k R's to the n.i. input.

A compensating parallel R-C in the feedback from output to inverting input provided adequate correction to bring things in beneath the max (~0.1% THD) spec., while not spoiling things when the caps were in circuit. This was an aftermarket automotive amp and the swiched lowpass was for the use as a subwoofer amp. A better way to have gone would have just switched from the direct signal to the fixed lowpass signal, but that was a higher impact change on an already-way-late program.
 
[quote author="PRR"]Note that we have been using opamps for 60 years, this type of effect is clearly noted in the 1970s, but has suddenly become a "problem". Partly because older amps had worse problems. Partly because BJTs tend to suffer less, and tend to encourage low/equal resistances more than FETs. Low-performance FETs such as on TL071 don't have such big capacitance (or variation of capacitance) as the improved FETs on newer amps. So as an old grump, I'm inclined to put this "problem" low on the list.[/quote]

Actually, some of the worst common-mode distortion I've measured was from TL07x's, fed from higher impedances. Amps like the OPA604 had considerably less in the tests I ran.

Back in the late 1970s a friend designed an otherwise-nice phono preamp with a 30kHz lowpass Sallen-Key filter, realized with a TL07x (I think it was an 074, but I could be mistaken). It sounded awful: glassy, spitty, screechy. It wasn't oscillating, as I verified with a 'scope; the problem was common-mode distortion. When I bypassed the low-pass filter the preamp sounded quite nice.

Peace,
Paul
 
Thanks for the answers - so if I understud them right, this phenomenon is not directly a result of limited CMRR and CM range, but rather one of modulated capacity (and other parameters) of the input devices?

www.angelfire.com/ab3/mjramp/cm.html offers some data on this, BTW.

The capacity modulation could be reduce by adding some linear capacity (i.e. from base to collector); probably not a good idea for stability though...

Samuel
 
[quote author="Samuel Groner"]Thanks for the answers - so if I understud them right, this phenomenon is not directly a result of limited CMRR and CM range, but rather one of modulated capacity (and other parameters) of the input devices?

www.angelfire.com/ab3/mjramp/cm.html offers some data on this, BTW.

The capacity modulation could be reduce by adding some linear capacity (i.e. from base to collector); probably not a good idea for stability though...

Samuel[/quote]

As well, PRR's points are well-taken: the "plate" resistances as a function of "plate-grid" voltage are also involved. Low pinchoff voltage FETs help, and if they also have high drain resistance even better. Unfortunately the latter tends to be incompatible with high transconductance. Bipolars are nice for achieving a high collector resistance once the collector clears the base.

Adding external C won't help. You will have to account for it somewhere, and to the extent that you do you will be back to seeing the voltage-variable effects of device C's. Bootstrapping does help, but introduces some high frequency anomalies that can give heartburn.
 

Latest posts

Back
Top