This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

  • Resolved

OPA2189: Excessive noise with OPA2189

Prodigy 60 points

Replies: 10

Views: 473

Part Number: OPA2189

I wonder how you are measuring the noise spec for the OPA2189

Unfortunately, I have not been able to get near the specified numbers when using it in a differential amplifier.  I do get 30nv/rtHz or so, at 1khz and referred to input when I expect less than 9.  Non inverting inputs are tied to ground.  I tried various gains, from 4 to 21.  No difference.  Replacing the chip by an OPA2209, the only decent high voltage amp I had on hand improves the performance considerably, so it looks like it is something from the op amp itself.  I tried several different 2189 with similar results.

And, by the way, I get what I expect when I test a low voltage low noise self correcting op amp like the CS3002 with the same setup.

For your information, I am using a SRS SR1 analyzer to measure the noise.

What is wrong?  Am I missing something?  On paper, the 2189 looks like it will improve our products but the test results, as they are, do not look good.noise test.pdf

  • Hello Robert,

    Accurate Op amp noise measurement is an acquired skill and it is easy to obtain results that don't match the datasheet. Since semiconductor companies who produce low-noise amplifiers must be able to assure their product's noise performances and they have developed and apply accurate measurement techniques.

    I am providing you a link to TIPD147, a TI Precision Design, that provides information about the techniques we use here in Precision Amps Applications for Op amp noise measurements. It was written by Art Kay, who has literally written the book on Op amp noise, "Operational Amplifier Noise: Techniques and Tips for Analyzing and Reducing Noise." Here is the link to the TIPD:

    http://www.ti.com/lit/ug/tidu016/tidu016.pdf

    Do note that our product characterization team has even more sophisticated methods for noise characterization and they are used for the datasheet information. Their techniques are proprietary. 

    As a point of reference the OPA2189 Spice simulation model accurately models the noise. Here is a TINA Spice noise simulation using a single-ended version of your circuit:

    The OPA2189 is specified with a typical voltage noise density en, at 1 kHz of  5.2 nV/√Hz. The addition of resistors which contribute thermal noise as discussed in Art's TIPD and increases the overall noise. The TINA simulation indicates an input referred voltage noise spectral density of approximately 6.5 nV/√Hz at 1 kHz. the output referred noise is 11x the input referred noise; simply the input referred noise gained up by the circuit noise gain (1 + R3/R1).

    Do observe that since the OPA2189 employs chopping techniques that the Op amp does not exhibit 1/f noise. Therefore, 1/f noise does not appear in the TINA Spice noise plot shown above, or in an actual bench measurement.

    Regards, Thomas

    Precision Amplifiers Applications Engineering

  • In reply to Thomas Kuehl:

    Thank you Thomas.

    Unfortunately, the information you provided does not help very much.  I know that the spice model predicts 7nv/rthz or so for a single ended model.  Say 10nv for the differential one.  The problem is that measured noise is much higher than expected noise.

    And, while I agree that noise measurement is a difficult acquired skill, I do believe that I know what I am doing there.  I design extremely low noise sensor preamplifiers for a living and I have over 30 years experience at that game.  I do not need the basics and I learned, over the years, not to trust Spice too much at that level of performance.

    The SR1 spectrum analyzer does not need an extra preamplifier.  It can measure reliably to 5nV/rthz or so, in the audio range.

    I did the initial test on a partially assembled target circuit PCB and, to confirm, rebuilt the thing on an universal op amp EVM with the same results.  To make sure there are no circuit problems or mistakes, I replaced the OPA2189 by an OPA2209.  I measure down to 3nV or so with the 2209 while I get 30nV or more from the 2189.  Same circuit and setup.  Admittedly, there is some external noise pickup but, as the 2209 measurement shows, it does not mess the readings too much.  I would not say that my measurements are perfect but they definitely are close to reality.

    I suspect that the specified noise for the 2189 is measured at a very high gain.  May be 1000 as seem to be traditional.  Most "chopper" amplifiers I used before seem to be tested that way and do have excellent performance at high gain but tend to have a dismal noise performance at low gains.   As I am targeting a low gain amplifier this time, I did not try to measure the 2189 noise with a gain of 1000.  I suspect I would have much better results that way but that is irrelevant for the current application.

    I was hoping that the 2189 was able to perform better at low gains but, except if you can lead me towards something I missed, I am beginning to doubt it.

    I am actually interested in improving the noise performance around 0.01 to 1 Hz, from a potentially high resistance source.  I am currently using OPA2140 in that job and am looking for possible improvements, while still using an integrated solution.

    Could you find out how the data sheet noise figure was determined?  That will help me decide if the issue is worth pursuing.

    Thanks again

    Robert

  • In reply to Robert Bazinet:

    Hey Robert,

    Nothing looks too crazy here, you might try taking off those 220pF output loads, I ran a loop phase margin with those and seems fine, but easy to try.

    One other thing I run into occasionally is the diff amp you show might have a CM loop oscillation that the chopper is mixing down. This whole chopper mixing thing is not described too often, but is brought up in this article - it would be a long shot, but you might have an interferer close to the chopper frequency that is being mixed down to where you are making your measurements. 

    https://www.planetanalog.com/author.asp?section_id=3393&doc_id=564847

    For the CM loop issue, which again is a long shot here as the OPA2819 is unity gain stable, I showed a few resolutions in section  9.1.5 in this datasheet.

    I am assuming your noise measurements are taking the diff output to single ended in some fashion?

    http://www.ti.com/lit/ds/symlink/opa838.pdf

    Michael Steffes

  • In reply to Michael Steffes:

    thanks Michael

    You are assuming right:  I am feeding the differential output from the amplifier in the differential input of the spectrum analyzer.  I used both HiZ and 600 ohms terminations.  No problem either way.

    I did try without the 220p caps.  No difference.  However some caps are necessary if I push the gain to 100 or so.  The amplifier will oscillate without capacitors.

    I doubt about common mode issues.  I did not test single ended as such but I did split the center 1k gain resistor and grounded it, to use 2 single ended amplifiers:  exactly the same noise performance.

    Pickup is improbable in that setup but still a possibility.  The test board is battery powered but was quite close to the test station with the spectrum analyzer.  I will use a longer cable and move the test board into the multilayer magnetic shield, just to make sure that the instruments are not radiating.  I usually do not bother with that shield when testing at that level as experience has shown it not to help that much.

    Bye

    Robert

  • In reply to Robert Bazinet:

    Your methods seem good, especially your comment that the OPA2140 measurements match PDS expectations - one of the things new in the OPA2189 is that chopper frequency, I think it would be nice if there was a summary listing across the TI precision parts of what that chopper frequency is, changes by device part #. And it does have some variability part to part as well. You might form the output CM voltage with two 1kohms to a center point and measure that noise over frequency - you might be seeing a CM rejection issue in the SRS if that is really high. 

    Michael Steffes

  • In reply to Michael Steffes:

    I did try the center tap approach:  I used 2 grounded 499 ohms instead of the 1K center resistor.  No obvious difference.

    I suspect the 2189 works like most "choppers" I used before.  Very low noise at very high gains and much higher noise at lower gain.  I do successfully use CS3002 choppers in a very low noise preamp with a gain of 10000 but I am aware that those op amps are not nice at all in a low gain application.  From the data sheet, I had some hope of the 2189 but it looks like we are better to stay with the 2140.

    I will wait to see if Thomas can indicate how the noise was initially specified and if there as been any noise measurements at low gain.  I hope he will prove me wrong but at that point, I am almost convinced that the data sheet is honest only for very high gains.  In which case, better forget this part for the current application.

    Robert

  • In reply to Robert Bazinet:

    Thanks Robert,

    I actually did not know that Cirrus part existed, I thought they mainly did Apple audio parts - occasionally they have tried to branch out, but the ROI is low outside their Apple business. I did just look on cirrus site, they do have an op amp category where this CS3002 is their only op amp?? have 4 audio amplifiers as well, so not a huge presence in this area. 

    https://statics.cirrus.com/pubs/proDatasheet/CS3002_F10.pdf

    I guess I am struggling with your higher noise at low gain comments - input noise is input noise, should not be changing with gain setting? perhaps there is something here I have not seen before -have also extracted and modeled numerous new high speed op amp noise terms - but no chopper experience there. 

    Michael Steffes

  • In reply to Michael Steffes:

    Cirrus/Crystal do no have an extensive line of op amps.  Nevertheless, I have been using the CS3002 since it has been introduced.  The best product at the time.

    I know my needs are unusual.  The application is measuring natural electric fields.  I am after nV to uV range signals.  Unfortunately, there is a lot of other "garbage" surimposed that prevents increasing the gains.  Difficult to use a gain of say 1000 when there is also .5V of 60Hz at the inputs.  So the dynamic range and noise level of the preamplifier becomes somewhat critical.

  • In reply to Robert Bazinet:

    Hello Robert,

    I had a discussion with one of our Precision Amplifiers Test Engineers, and one of our Product Characterization Engineers in an attempt to gain a more complete understanding how the OPA2189 noise is measured and characterized. I have a better understanding of what is involved, but can only provide certain information because of TI's proprietary test methods.

    The OPA2189 utilizes a chopper architecture as you are aware. When the input switches associated with the chopping action switch the input bias current IB is modulated by charge transfer occurring within the Op amp's input circuitry. The amount of time when the charge transfer occurs is very, very short compared to the overall timing cycle. However, during that short time the input current increases from the nominal ±70 pA, to a current peak that may momentarily be upwards of a microamp. That peak IB current when averaged with the lower nominal IB input current averages out to the IB numbers listed in the OPA2189 Electrical Characteristics table.

    This chopped, or modulated IB is provided to the Op amp inverting input via the gain set resistors. And since the current flows though the resistances it is converted to voltage that appears in conjunction with the natural voltage noise component. The higher the resistances the higher its contribution will be to the voltage noise.

    Since in Op amp noise measurements it is beneficial to run the DUT in a gain as described in TIPD147 gain resistors are obviously required. In TIPD147 the DUT gain is set for a noise gain of 10 V/V, but R1 is 10 Ohms and Rf is 90 Ohms. That should lower the voltage noise contributed by the IB noise conversion.

    Unfortunately, I cannot share the test circuits that our Characterization Engineer has developed to measure the voltage noise. But what I can tell you is that each of the two chopper Op amp inputs need to see equal and balanced impedance across the usable frequency range. This can be more difficult to achieve than one might expect.

    An alternative to using a chopper Op amp is to use a low-noise non-chopper such as the e-trim™ OPA2192. The dc precision approaches that of the OPA2189, and its ac performances meet or exceed those of the OPA2189. Since it uses a more conventional, non-chopper architecture it will exhibit 1/f noise which may be an issue for your application.

    In case you want to see what the OPA2192 has to offer, here is the link to the datasheet:

    https://www.ti.com/lit/ds/symlink/opa2192.pdf

    Regards, Thomas

    Precision Amplifiers Applications Engineering 

  • In reply to Thomas Kuehl:

    Thank you Thomas, this is much appreciated.

    For the fun of it, I will try measuring the 2189 noise at a gain of 10 with matched 10 ohms or so input resistors.  I am curious to see if I can get close to the published specs.

    However, this cannot be done in the real application.  I could use lower value resistors for gain setting but the source signal impedance is variable, at least 500 ohms but more typically 1 or 2 k.  Even higher in some circumstances.  And this comes from nature, nothing I can do to reduce it.

    This being the case, it does seem that the 2189 is not such a good choice for that application.  I will likely continue using the OPA2140.  Unfortunately, as you mention the 1/f noise is a bit of a problem, mainly around .1Hz - 1Hz, where the measured signal is the weakest.  A "chopper" amp would seem to have an advantage in this range but I never found one that actually works.  Overall, from the tests we done, the 2140 still gives the better overall performance.

    Of course, the very best is a discrete implementation but it is inconvenient, both physically large and expensive, as good matched transistors are difficult to get these days.

    Finally thanks for the reference to the OPA2192.  I did not consider that model and will definitely look at the data sheet and try it out if it looks susceptible to improve the overall performance.

    Again, thank you very much for your help and let's consider this case closed.

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.