Improving the CMRR of Instrumentation Amplifiers

A typical instrumentation amplifier (IA) in IC form has a pretty good power-supply rejection ratio (PSRR) and common-mode rejection ratio (CMRR). However, once you put the device in your circuit, things will probably get worse. The PSRR may degrade some. The more serious problem will likely occur at the device's signal inputs. Add your own circuitry in front of the IA, and the CMRR will get worse.

This unhappy occurrence is inevitable. If the impedances aren't matched perfectly for any circuitry you add at the +input and -input terminals, the CMRR will suffer. In an actual application, you would likely add low-pass filter (LPF) networks to each of the inputs. These would take the form of a series input resistor and a shunt (to ground) capacitor.

You might even cascade R-C sections to get a two-pole (or more) passive filter. Better yet, an active filter would be good, but more on that later.

The resistors that you select (R1 and R2) can be matched at reasonable cost to 0.1 percent. That will help keep the CMRR at acceptable levels for many applications. But getting good capacitor matching (C1 and C2) is a lot tougher. You can either pay an exorbitant price or custom select and match devices yourself, and unless your labor and overhead costs are nil, this would still be expensive.

An additional complication comes from the miscellaneous stray capacitances not shown above. These would include the parasitic capacitances across R1 and across R2, across the +input and -input terminals, and from +input and -input to output. All these capacitances will be somewhat unpredictable, so you can expect them to cause CMRR degradation.

Depending on the physical board layout, there may be sufficient PC board traces to allow for capacitive coupling of stray electrical fields. If the twists and bends of the PC board traces are just right, there may be small inductive loops that will pick up stray magnetic fields. Either way, CMRR degradation occurs.

A possible solution is to integrate the resistor and capacitor network in silicon. Just creating an IC with these passive components could prove helpful. But there are still some parasitic PC board capacitances and, to a lesser extent, parasitic resistances. And there still may be enough inductance to cause trouble.

If we could combine the R-C network mentioned above with the IA, we might get closer to the best possible embodiment of the IA. And if we could add provisions to tweak the capacitor and resistor values in the low-pass networks slightly, we might be able to adjust the IA for maximum CMRR. Now, that would be a sweet device. Could that process be automated via an onboard microcontroller unit?

A variation on this would be to build the usual IA using three op-amps: one for the +input, one for the -input, and one to sum the output of each of these input op-amps. The two input op-amps could be configured as active LPFs. But again, you must match the transfer functions of each of these LPFs, or you can kiss your CMRR goodbye.

The same techniques for tweaking resistor and capacitor values could be applied to the LPFs' passive components. If there were an MCU involved, you could communicate with it and program the LPFs to the desired corner frequency.

Have you designed such a device already? Would you have a use for one?

Related posts:

26 comments on “Improving the CMRR of Instrumentation Amplifiers

  1. Scott Elder
    August 6, 2013

    Brad, It seems the best approach is to find ways to install the amplifiers at the source where the common mode is local or doesn't exist.  And as soon as one can put an amplifier at the source one can put a data converter there as well and send the information in a digital format.  Just need a way to harness the energy locally.  Seems practical as long as local isn't real hot or real cold.

  2. Michael Dunn
    August 6, 2013

    Low ESL X2Y caps can be good choices for IA LPFs, as the two halves tend to be very well matched. E.g.

  3. Brad_Albing
    August 6, 2013

    @MD – that could work – I'd use the NP0 version rather than the more common X7R version. Costs a little more and you can't get them in as high cap value as X7R of course.

  4. Brad_Albing
    August 6, 2013

    @Scott – well sure, you could do that too – if you have local access. I was just looking at another approach to solve the problem when you don't have local access.

  5. Scott Elder
    August 6, 2013

    @Brad – Why not use a single capacitor across the input so that you don't need matched capacitors?  And then it would be practical to “tune” on the IC (as you suggest) the parasitic capacitance to ground because the amount would be much smaller, hence practical, to solve a different problem (i.e. RF Noise).  Breaking the problem in half if you will.  

  6. Brad_Albing
    August 6, 2013

    @Scott – Hmm… OK, that'll work for the passive version. I'll stand by my thoughts tho' for the active filter cktry built around each of the two input op-amps.

  7. Vishal Prajapati
    August 7, 2013

    X7R and X2Y capacitors are new thing for me. I have heard it for the first time. I have new thing to explore. Thanks for sharing.

  8. samicksha
    August 7, 2013

    Sounds oka, but hope we are not mitigating the fact that CMRR often depends on signal frequency.

  9. Netcrawl
    August 7, 2013

    @Brad that was great, I do have some question, how about if input V1 and V2 are the same, does it means that the output voltage is also zero?  

  10. BradWood
    August 7, 2013

    If the overall use of the IA output is in a sampled-data system that will tolerate some deadtime and glitch energy, a strategy would be to do input switching and a synchronous detection, to, as it were, chop out the inevitable differences in input filters.

    Short of that, the topologies for multipole filters include ones in which at least some of the capacitors are shared between the two inputs.  This helps a lot.

    At low frequencies, the errors arising from a mismatch in impedances to common can be alleviated by simply having higher normal-mode impedances at the IA inputs. For higher frequencies in the presence of high common-mode noise, a well-balanced common-mode choke can be helpful.


    Brad Wood

  11. RedDerek
    August 7, 2013

    X2Y was started for EMC noise limitation in motors. They have expanded their application. They can be used for lightning protection as well.

    Is the purpose of adding the capacitance on the amp input to force equalization of the input capacitance and not depend on the typical parasitic input capaticance? Then the X2Y would help swamp out the parasitics with a matching cap.

    Drawback would be a slower response due to the input RC filtering.

  12. TomAtMuse
    August 7, 2013

    I was unaware of X2Y caps; thank you, Michael, for bringing them to my attention.  Their use with IAs is discussed here:



  13. Netcrawl
    August 7, 2013

    @TomAtMuse thanks for a great link, understanding  X2Y technology and different connection options would definitely help you achieve succesful lab evaluation and optimal production design, recently there's some new development going on here, development such as single component solutions for noise suppresion.

    here some good discussions:




  14. Davidled
    August 7, 2013

     CMMR might be improved in the instrumentation. Typically, measurement instrumentation need to be maintained by calibration every a certain years that manual indicated. Unless calibration is checked as like oil change, CMMR might be degraded.  Good Calibration provides a good CMMR.

  15. Brad_Albing
    August 7, 2013

    @Netcrawl – yes – if you apply the same voltage to the +Vin and the -Vin, the output will be zero.

  16. Brad_Albing
    August 7, 2013

    @VP – Keep in mind that the X7R is a particular formulation of the ceramic material for the capacitors. It has certain characteristics regarding how the capacitance varies with ambient temperature, applied voltage, and how it ages Compare that to other types or compositions like NP0, Y5V, and Z5U.

    The X2Y defines a configuration of 2 capacitors in one package. The compostion could be any of the common types that I mentioned above. The key is that the 2 sections are very similar in terms of capacitance variations with temperature – sort of a matched set of capacitors. So that's handy for applications like what we talked about in the blog.

  17. ZekeR0
    August 8, 2013

    @Brad, “The compostion could be any of the common types that I mentioned above.”

    This is true if the inputs are at the same voltage as each other. However, since this is an IA, you can't expect that. For this purpose you'd prefer a capacitor chemistry with very little voltage coefficient—less than 3% variation over the full voltage range to capitalize on the X2Y structure's excellent zero-bias matching.

  18. Vishal Prajapati
    August 8, 2013

    Thank you very much BA and RD. Thanks for sharing. It is great to know such a new things. Great insites sir.

  19. Brad_Albing
    August 8, 2013

    @ZekeR_ 'Tis a valid point. Off the top of my head, I'm not sure what the voltage coefficient is for X7R. We would hope tho' that in an X2Y [dual] capacitor, what ever the V-coef. is, it would track between the 2 sections.

    But the general point of the whole discussion here is to find a way to build an IA with as much of this cktry integrated into it as possible. That would for sure minimize these sorts of problems.

  20. Brad_Albing
    August 8, 2013

    @BradWood – Yep, your suggestions can and would help. I was suggesting my course of action for situations where you needed the absolute best possible performance – i.e., extremely low CMRR – for the very high precision data acquisition systems. For those, even a little impedance mismatch at the inputs will create some annoying errors.

  21. Kenneth43
    August 10, 2013

    Put the cut off frequency of the input RC-filters well over the desired bandwidth of the wanted signal. Then filter the single channel to the desired bandwidth after the instrumentation amplifier. The commonmode signal will generate signal due to missmatch through the IA just around the cutoff frequency of the input filter. The only penalty is that the bandwidth of the IA must be higher than with input filtering to the desired bandwidth.

  22. Guru of Grounding
    August 10, 2013

    Seems to me that a big point has been missed in this discussion. In reality, CMRR is a system measurement – there must be a specified driving source connected to the differential inputs in order to make such a measurement. We are accustomed to seeing CMRR figures for diff-amps under ideal lab conditions (generally with inputs shorted to each other) where the influence of the real-world signal source's pair of common-mode source impedances can seriously degrade the “ideal” CMRR figure. As I've harped for years to the professional audio community, real-world CMRR is often far, far less than the figure touted on data sheets. I explain it briefly starting on page 22 at As shown there, the way to reduce the influence of CM source impedances is to raise the CM input impedances of the diff-amp. Infinity is, of course, ideal but in audio work you can't just bring an IA's input pins out to an XLR “balanced input” connector and let them float! Many audio balanced inputs have CM input impedances under 10 k-ohms. When it's this low, even a fraction of an ohm of impedance imbalance in the driving source can quickly degrade 90 dB of CMRR down to 60 dB. The main reason diff-amps that use input transformers offer such high CMRR, regardless of source imbalances up to several hundred ohms, is that their CM input impedances are over 50 M-ohms at 60 Hz. I have a patent on a method of bootstrapping the bias-supplying resistors at the input of an IA to raise the CM input impedances to similar levels. An IC using the patent is available from THAT Corp. as the 1200-series InGenius(r) input stage. One has to be very careful adding capacitors for CM filtering of RF, since they too lower CM input impedance. But bootstrapping can again be used to make these capacitors effectively remain small at audio frequencies but become large at RF frequencies – minimizing sensitivity of CMRR to CM source impedance imbalances. This is shown in the applications section of the 1200-series data sheet. This issue is important enough in audio equipment that I suggested a change in the way IEC standards test audio equipment for CMRR. They admitted that their old test bore no correlation to real-world results … and adopted my new test in 2000 (see IEC Standard 60268-3).  — Bill Whitlock, president & chief engineer, Jensen Transformers, Inc., AES Life Fellow

  23. Brad_Albing
    August 11, 2013

    @Bill – you're right of course – we often overlook the system aspect of our designs and just try to design an IA with near-perfect specs. The real world throws a monkey-wrench into the works, so your ideas are worth a closer look.

  24. SunitaT
    August 20, 2013

    The current-mode instrumentation amplifier based on second-generation current conveyors (CCII) offers many benefits over conventional instrumentation amplifier architectures. It does not require any matched components to reach high CMRR and its bandwidth is not gain-bandwidth product limited. However, enlightening the CMRR of this amplifier is not as direct as with voltage-mode topologies. Therefore, the matrix representation of the current conveyor is extended to include all linear non idealities and based on this an enhanced current-conveyor macro model is needed

  25. Brad_Albing
    August 23, 2013

    @samicksha – You always need to keep that in mind. Sometimes the CMRR spec is good at low frequencies but deteriorates and higher frequencies. Sometimes is tose pesky parasitic reactances….

  26. Pingback: Purpose of capacitors connected to both inputs of Op-Amp – GrindSkills

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.