Instrumentation Amplifiers And How To Measure Miniscule Change

Instrumentation Amplifiers with Bil Herd

These days there a large number of sensors and analog circuits that are “controller friendly” meaning that their output signal is easily interfaced to the built-in Analog to Digital Convertors (ADCs) often found in today’s micro-controllers. This means that the signals typically are already amplified, often filtered, and corrected for offset and linearity. But when faced with very low level signals, or signals buried in a larger signal an Instrumentation Amplifier may be what’s needed. The qualities of an Instrumentation Amplifier include:

  • A differential amplifier with high impedance and low bias current on both inputs.
  • Low noise and low drift when amplifying very small signals.
  • The ability to reject a voltage that is present on both inputs, referred to as Common Mode Rejection Ratio (CMRR)

DGIABefore I get too far let me point you at an excellent resource; Analog Devices has made available a free book “The Designers Guide to Instrumentation Amplifiers” (PDF). This has been added to my collection and because I am from the databook generation I have printed and stapled my copy.

If we look at a standard differential amplifier made with an Operational Amplifier (OpAmp) we see some issues , namely that the inputs are not high impedance, and they are not balanced. The node at the non-inverting input is a virtual ground due to being driven by the output through Rf; the input impedance of this leg is equal to R1. On the non-inverting input the impedance is equal to R2 plus Rg as Rg leads directly to ground.Consequently the inputs are dissimilar due to the different configurations.

Differential Amplifier OpAmp
Differential Amplifier OpAmp

We can represent a more ideal Instrumentation Amplifier as shown below. Note that only high impedance inputs are used with no extraneous paths to ground. The final stage provides the differential operation of subtracting one signal from the other. Creating an Instrumentation Amplifier this way would still require the use of stable amplifiers preferably with very low input currents, in the old days I used to use Precision Monolithic’ s OP-07 amplifiers, alas PMI is long gong these days having been acquired by Analog Devices.

Instrumentation Amp with 3 Amps
Instrumentation Amp with 3 Amps

To demonstrate an Instrumentation Amplifier in use in the video, I connect a sensor used in digital scales (known as a load cell) to the instrumentation amplifier and measure the output with my Keithly voltmeter. Since my custom PCB did not come back in time I used one of my two Instrumentation Amplifier evaluation boards I have lying around, this one made by Analog Devices.

A load cell is made from strain gages attached to a metal frame and essentially we are measure very minute changes in length of the copper traces due to the stress on them – a very tiny signal! Also because the gauges act as a voltage divider across 15 volts, the very tiny signal we want is superimposed on top of the relatively huge 7.5V. The ability to strip the huge 7.5V DC signal that is common to both inputs is known as the Common Mode Rejection Ratio and is an important aspect when picking an instrumentation amplifier.

Connecting the load cell to the evaluation board set for a gain of 1000 does give us enough amplification where we can easily see the effects of applying a force to the load cell. As this is a wide band amplifier and there is no real filtering going on you can see what is essentially amplified noise present in the oscilloscope trace.

 

In a real weighing application there would be judicious use of shielding and some low pass filtering applied to create a consistent DC voltage representing the weight on the load cell.

I will talk more in the future about reducing drift and noise and other analog tips, meanwhile the Instrumentation Amp is often the key to some low level or critical signal measurements that may otherwise elude attempts to be measured by the standard A/D fare.

60 thoughts on “Instrumentation Amplifiers And How To Measure Miniscule Change

  1. Good Post as always. For anyone who is to lazy to build something like that yourself: Look für HX711 on ebay/ali. It gives you a serial interface to your loadcell. You will learn nothing about amplifiers and stuff but will get the job done.

  2. I don’t think it is correct to say that the inverting input is a “virtual ground” in your differential amplifier circuit. The voltage at the inverting input will be the same as the voltage at the noninverting input, so if V2 is not equal to zero volts then the voltage at the inverting input will not be zero volts. You could say that there is a virtual short between the inverting and noninverting inputs, but they are not necessarily at ground potential in this configuration.

    1. Ground is only held at 0V for convience. In practice a proper ground can be any voltage that is sufficiently stable and this is the exactly correct use of the term virtual ground.

      1. To eleborate real quick, An ideal ground is a source of (approximately) infinite free charge. By it’s nature, that charge will equalize to 0V and since the charge supply is inifinite, it will equalize any connected floating node. With an electronically created ground, you can hold the voltage to any point you want because you have a device to source the charge that equalizes anything to that voltage. Typical grounds are the common mode voltage of your signal, the half rail point, or 0V. The Opamp produces a virtual ground by enforcing it to the voltage on the second input, but not strictly referencing it to the circuit ground. This is done internally via a current mirror that forces equal current splitting between two amplifiers with a common active load. Lastly, it is important to remember that in electronics everything is about relative potentials, not absolutes, so two 0V grounds can wind up being totally different voltages when you connect them.

        1. Well said everyone, when I talk about ground in terms of impedance I don’t necessarily mean 0V, I mean to a point of low impedance, which then determines the input impedance. My point was trying to learn what affects the impedance without branching into discussions of superposition and Thevenin equivalence., In the video I also mention needing to bias the inputs and there I said ground but really didn’t mean a path to 0v necessarily.

          As an old analog guy I will sometimes say something that sounds counterintuitive but am trying to build an intuitive approach that may not be clear by just reading textbooks. If before now you did not recognize the inverting input in this this configuration as being a virtual source (ground or otherwise) then I have partial succeeded,, hopefully your eyes spot these junctions even before the brain has time to digest .

          1. BTW, just to chime in a bit more, a virtual ground has limitations so we use the term in context, such as the context of understanding the equivalent impedance. If you were to hang a large capacitor on the virtual ground node mentioned it most likely would oscillate proving its not really a 0V ground after all.

          2. It is more like a virtual short bought on by a closed feedback loop with the value same as the voltage divider connected to V2. Yes, it has a low impedance, but you cannot connect an AC short to it as you are assuming that it is DC only or not carry any signal or common mode noise components for a long cable (or both !). If your V2 came from the strain gauge or a balanced audio, it is part of the signal!

            Just treat it as a virtual short for the AC/DC analysis assuming the opamp is operating within its parameters and can close the feedback loop. i.e. no input/output saturation/within common mode/within voltage swing/load capability/

          3. Calling the Non inv. input a virtual ground in this circuit is completely wrong and misleading.

            The Impedance @ V2 is the same: Rg/(R2+Rg), for both common mode and differential input voltages, but the impedance for V1 is different for common mode and for differential input voltages in the single opamp circuit.

            Because the voltage of this (here not so) “differential ground” changes with the value of V2 there is also a feedback from V2 to V1.

            Wikipedia has it right:
            In electronics, a virtual ground (or virtual earth) is a node of a circuit that is maintained at a steady reference potential, without being connected directly to the reference potential.

            The Voltage at the non inverting input in this circuit is not a “steady reference potential”, it changes with V2. Only if V2 is a constant voltage (potmeter, etc) it could be called a “virtual ground” but if the inamp is used om a wheatstone bridge with for example 4 strain gauges, then V2 and therefore also the “virtual ground” changes with the input signal.

          4. The inverting pin of the opamp IS maintained at a steady state potential due to the feedback of the amp (not counting the difference between V1 and V2 which is not part of calculating the impedance on this leg).

            A change in the potential or resulting current into the inverting input does not result in a change of the potential or total current at the node right at the amp (due to feedback) , it is a “grounded ” reference for purposes of calculating the impedance of the input of the amplifier. If it makes more sense to you, you can refer to a low impedance node of unknown potential for purposes of understanding why this input does not have a high impedance.

          5. Check out http://www.ti.com/lit/ml/sloa075/sloa075.pdf or google “op amp circuits superposition virtual ground”

            Frankly the argument about the single definition of the word ground as meaning 0v (measured to what?) distracts from the point of determining the drawbacks of this circuit compared to the alternatives. If I chose to draw the ground symbol on the +5v source instead of the 0v source nothing would actually change, but saying that ground means only 0v would now have to adapt their explanations to a drawing convention.

            (Excerpt from TI app note)
            3.3 The Inverting Op Amp
            The noninverting input of the inverting op amp circuit is grounded. One assumption made
            is that the input error voltage is zero, so the feedback keeps inverting the input of the op
            amp at a virtual ground (not actual ground but acting like ground). The current flow in the
            input leads is assumed to be zero, hence the current flowing through RG equals the current
            flowing through RF. Using Kirchoff’s law, we write Equation 3–4; and the minus sign
            is inserted because this is the inverting input. Algebraic manipulation gives Equation 3–5.

    2. V2 itself could be part of the signal say a differential signal e.g. Balanced Audio which is not a DC reference any more. For a long cable pair from the far end, the source of V2 might have been a ground reference for a single ended signal now is no longer a simple DC offset. It carries the common mode voltage which can include AC noise from coupling + an AC/DC offset.

      If that reference point itself is an AC signal, might cause more confusion by calling it a (virtual) ground.

  3. I have been working with instrumentation amplifiers a little while.Normally I just google everything and I find what I need I’m not very social. I have a question that I have not yet found the answer for if anyone would be kind enough to answer. I am trying to build a bluetooth eeg.I’ll spare you the details unless someone finds it interesting. My problem is with my modifications to a basic eeg.I’m not sure if it will work but I’m still experimenting. At the input instead of having multiple amplifiers I’m trying to multiplex the inputs using an analog multiplexer. So far I have a dual 4:1 multiplexer going to an ad8221 instrumentation amplifier. I don’t know if the problem is in my breadboard or the circuit design but when I touch the input to the multiplexer the output changes regardless of which input is selected. I’m ordering some adg409 multiplexers to experiment with but I’m not sure they will help.So to summerize my question.

    Does anyone have an example of an instrumentation amplifier with multiplexed inputs?

    1. The output of the multiplexer or the amplifier is changing when you load your inputs? Given the cost of nice instrumentation amplifiers, it is easy to see why you would want to do this, but you might want to be careful as EEG is a bit of a pain to measure anyways since it can be so easily swamped out by EMG or background electrical noise and multiplexing before amplifying will only make things worse. You might also want to look at one of the newish EEG chips from TI. You can get a really nice 8 channel IA intended for use with EEG (or EMG or whatever, it is really just a nice low freq set of amplifiers) that outputs SPI that you could almost directly feed into your bluetooth for around $50.

    2. Could you not use one of the many ECG chips such as the ADAS1000 (Analog Front End and ADC in one) and save yourself some trouble?

      That said, the open switches in the multiplexer are essentially capacitors … presenting a voltage to those capacitors (by touching the input) will disturb the output, both the magnitude and duration of that disturbance depend on the source resistance of the input with the closed switch.

      1. I am a big believer in not re-inventing the wheel and I would be more likely to use a chip specifically engineered to deal with the real life issues of something like EEG, its a good bet that they spent hundreds of hours of engineering time on their product.

  4. Yeah you’re probably right because of the sensitivity of the amplifier but there are a lot of variables. the quality of the multiplexer for example. so I would have to have more than one sentence to throw away my plan. If you could please provide a more detailed explanation.

    Also Bil has done some great videos and I noticed the Robotech poster in the background.

    1. I’m not suggesting you throw away your design. In fact the aforementioned wonderful ADS1299 has a multiplexer block before amplification. What I was after is that we need more information on your particular setup and the problem you are facing before we could realistically advise you.

    2. woooo….. Robotech! I originally came across this in Tokyo where it was called Macross. A bad day was when my manually set VCR didn’t record the Robotech episode off of the Philly UHF channel correctly back in the 80’s.

  5. I’m a masters student in mechanical engineering and am working on a strain gauge based load sensor. I really wish this had come out before I designed my own board based around the INA 129. I would like to see what you do in part two as far as adding some RF noise suppression. Thanks for the video and providing some knowledge to those that are not experienced in the field.

    1. Take a look at the schematic for the eval and you will see the RF filters http://www.analog.com/media/cn/technical-documentation/evaluation-documentation/498571974AD8222_EVALZ.pdf . Also check out the handbook I linked to, they probably mention it there. Best way is to keep RF out in teh first place through shielding, though shields can become energized if not careful (or shield jackets conduct current creating their own RF /EMC fields.)

      If putting into an A/D converter the output should be filtered to at least 1/2 the sampling frequency of the A/D (nyquist).

  6. Great video and article Bill. I am currently working on a project that uses a couple strain gages and an instrumentation amp. I have been somewhat successful at reducing the noise with low pass filters, but I have been stumped at getting rid of the voltage drift. I am mechanical engineering student that has only recently found the Analog Devices book you referenced, so I am not exactly a wiz with analog circuits. Even so I was pretty impressed with what a couple strain gages, a wheatstone bridge, and an IN-amp could do, even with what I cobbled together. I learned a lot in the process and it peaked my interest in learning more about filtering and instrumentation amplifiers. I am looking forward to another post on reducing drift and noise in these circuits, thanks!

    1. Google for “chopper stabilized amplifier” and also “synchronous detector”
      tl;dr version: if you are build a very high gain amplifier, you are also amplifying the DC offset of your amplifier. The DC offset is subjected to temperature or other source that are not completely corrected in the opamp.

      So the old trick of doing it is instead chopped up the DC signal into DC, amplifier it, AC couple the signal to remove the offset and carefully convert that highly amplified signal back to DC. The DC offset of that large gain is a DC component, so it is blocked by the AC coupling. You only have to worry about the DC offset at the final stage which is is small compare to the signal. Look for “chopper stabilized” opamp.

      For the fancy stuff, the AC to DC conversion can be done with a synchronous detector. There is a bit of math involved, basically anything that is not part of that switching frequency is averaged out over a longer time frame (aka low pass filter) leaving you the signal – very good noise rejection.

      There are now neat digital technology that keep track of DC drift in opamp.

      1. This technique of a carefully filtered baseline seems to be omnipresent when dealing with small signals. For example the ultrasonics I have done would not work at any distance if the baseline/ground reference wasn’t an “intelligent” one. Plus you can inject corrections like we used to have a square or cube root error vs. time that I would inject into the baseline and null out the effects.

        With that said, give me a high-speed A/D and a semi-capable DSP and I can emulate all of that and more on the fly. :)

  7. Good article.
    Not trying to detract from Bils points, however it’s worth mentioning.

    We do a LOT of strain gauge monitoring work and historically we’ve used instrument amplifiers at highish gains (ie >250 V/V),

    We are now moving to 24bit delta sigma converters reading the gauges directly (with some passive anti alias filtering). As these high resolution converters become faster and cheaper, this becomes a more viable approach.
    As a bonus, noise immunity improves significantly…

    1. Technology marches on. :) If you saw the video I wave a 30 year old PCB where we built a self-correcting amp which is now just a chip or two. Our gain and zero drift correction was all done in 6502 floating point which we had a fast library we developed.

      I did our first food class scale and had to withstand steam washdown…. and decaying chicken fat which will eat through copper wire in a week. We could probably have a whole discussion on strain gage sites and implementations: I worked with DuPont in the 80’s to help identify conformal type coatings that would not create strain errors (especially at cold temperatures!) and not “pull” on the gage from the coating side (do they still use vaseline as a low friction barrier?)

  8. This is the exact experiment I did when I was in college, mind you mechanical engineering. We had to wire up a 741 to complete the task and use labview to analyse the data from the strain gage. I miss those days, things were so much easier

  9. The timing on this is great. I want to rebuild the control panel for my Cross Trainer so that it can interface with the Bluetooth Health Monitoring app in my Android Phone and one of the features of the existing control panel is Heart Rate monitoring via the hand grips. So, I have been looking into Instrumentation Amplifiers. Can’t wait for the next instalment.

  10. Dear Bil, it is apparent that you posses a great amount of knowledge about the good old ways of doing things that should in many ways be preserved for the current generation. I’ve seen some of your videos here on hackaday, a site that I frequent almost daily.
    Unfortunately I have to say that this time it didn’t teach me anything. You kept confusing the types of amplifiers the whole time. Like every single time you had to correct yourself. Why not edit that out? Why not have a rehearsed script for your videos? I just wish you had someone to help you with turning your knowledge into a somewhat usable video. It is obvious that your strengths lie in the know how and not at all in the presentation. Admittedly making a good video is time consuming and has a steep learning curve.
    It’s just a shame to see the contents scrambled into something that takes more concentration just to follow than to understand.

    Regards zero

    1. Bil, please don’t edit out the human being part of the videos. If I wanted nothing but the facts I’d pick up a textbook. I appreciate your videos and the occasional flubs. Keep up the excellent work.

      1. Thanks. If I were speaking to a group of my peers we would all be screwing up while talking in predictable ways, but we would all look past that to try and understand the point.

        Also if I overedit that becomes a distraction and NOBODY wants to listen to script! Better that everyone know I am a burnout but one that is burnt out from having been there. :) Damm, I just heard 200 brain cells dies…. getting old has a price.

  11. It’s spelled “converter”.

    It’s pronounced at it looks: “wheat stone”.

    “Virtual Ground” has a very specific meaning with regard to OpAmps. What you have labeled virtual ground, isn’t one. Saying it’s understood from context doesn’t cut it.

    1. Spelling may occasionally get missed.

      I have an accent.

      I disagree. It is virtual ground for the purposes I describe. My goal is to make the concepts available in an intuitive way, hopefully helping someone to automatically “see” these nodes when scanning a new design, I will always leave myself open for picking at the details, I will always misspeak when doing a diatribe but my main goal is to allow someone to get a concept. At my age I am willing to accept the criticism, I have argued with truly brilliant people some of them famous or infamous,and I am often wrong in some people’s eyes. If one person gets the fact that that node is nailed down and therefore the input impedance is not high then I am content.

      Thanks for your feedback.

  12. it’s rare to have the people who have the knowledge to actually be able to communicate it (and be willing to…) so thankyou Bill. Just had a lecture today where this exact circuit came up in this electrical subject ‘intro to biomedical eng’. It was briefly discussed and circuit solved… videos like this are priceless for people like me who want all the tidbits, from people who have been in the industry for decades, and are sharing their knowledge for free… let the man pronounce wheat stone however he damn well please

  13. Quick note: When an EE says “ground” he/she may not mean 0V, in fact I tend to say “zero volts” when I mean the actual potential as measured. One way we use the term ground is to depict an infinite current source. “Input impedance” of a circuit would be the impedance related to this “immovable” current source. This is an important point. Think of it as meaning “grounded to something that doesn’t move”.

    VCC is also ground the way I was taught to calculate impedance. It may have a different potential but a circuit with a 10K resistor to 0V source and a circuit with a 10k resistor to +5V source both have an impedance of 10k ohms.

    Even when saying ground is 0V I would still ask as compared and measured to what other potential. I might even ask which way current flows. :)

    1. In other languages the term for “ground” is “mass” (masse,masa…) which pictures well what you describe, the terms “earth-ing” and “ground-ing” lead to confusion about the 0V, specially on an airplane ;-)

  14. Dang it! What is with you guys who wildly wave things at cameras! Sparkfun Syndrome (and most of YouTube). Being able to speak reasonably well and fast doesn’t make you a good instructor. This would be a big fail on any instructor eval – for a bunch of reasons.

        1. Always interested in learning more, I am just a guy that liked electronics enough as a kid to land some killer EE jobs and I am trying to share my enthusiasm in video form now that I am a curmudgeon. Please email me at bilherd@hackaday.com with your contact info. Thanks! Bil

  15. Here is another: http://users.ece.gatech.edu/mleach/ece3050/sp04/OpAmps01.pdf (4 sentance)

    BTW these are great reads.

    1.5.7 The Instrumentation Amplifier
    The diff amp circuit of Fig. 1.20 is known as an instrumentation amplifier. In some applications, it is called
    an active transformer. To solve for vO, we use superposition of the inputs vI1 and vI2. With vI2 = 0, the
    v− terminal of op amp 2 is at virtual ground and op amp 1 operates as a non-inverting amplifier. By Eq.
    (1.10), its output voltage is given by
    vO1 =

    1 + RF 1
    R1

    vI1 (1.42)
    Because there is a virtual short circuit between the v+ and v− inputs of op amp 1, the voltage at the lower
    node of R1 is vI1. It follows that op amp 2 operates as an inverting amplifier. By Eq. (1.3), its output
    voltage is given by
    vO2 = −
    RF1
    R1
    vI1 (1.43)
    Op amp 3 operates as a true diff amp. By Eq. (1.30), its output voltage is given by
    vO =
    RF2
    R2
    (vO1 − vO2) = RF2
    R2

    1 + 2RF1
    R1

    vI1 (1.44)
    Similarly, for vI1 = 0, vO is given by
    vO = −
    RF2
    R2

    1 + 2RF1
    R1

    vI2 (1.45)
    By superposition, the total output voltage is
    vO =
    RF2
    R2

    1 + 2RF 1
    R1

    (vI1 − vI2) (1.46)
    This is the voltage output of a true diff amp. The input resistance to each input of the amplifier is infinite.
    The ou

    1. That PDF looks really familiar (but most of the app-notes, like the great Linear Apps book, and the old bound books from Analog look the same). Leach doesn’t say if he wrote it and there is no attribution in the text.

Leave a Reply to Bil HerdCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

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