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.

OPA348: What's wrong in this double stage SK LPF?

Part Number: OPA348
Other Parts Discussed in Thread: OPA378


I am struggling with a few things designing a double stage Sallen-Key LPF and between PSpice and real circuit there are up to +/- 1 dB of variation at 10 kHz.

The circuit is:

I have downloaded the PSpice library for OPA348 from the TI website and applied, against the IDEAL OPAMP there are some differences as should be, first of all the gain and the unit GBW, the input and output impedances that are important in the Sallen-Key arrangement.

The circuit has an input buffer, still OPA348, unit gain with an RC at + input: 3k32 and 4n7.

The response curve is a modified Bessel to recover the non-linear response of the input transducer at frequencies higher than 2 kHz (i.e. the transducer has a +3 dB @ 11 kHz).

What's is weird in example that if R48 is 1k91 ohm, the simulation cannot converge, with 1k911 works. I have 3 boards, the prototype one is nearly like PSpice +/- 0,1 or 0.2 dB max along the whole response curve.

So, I have assembled other 2 boards but one has + 1 dB @ 10 kHz respect the PSpice and the other - 1 dB.  

I have checked several times the boards and appears the same, components are in the tolerance (resistors 1%, capacitors 2%) but the final behaviors are different.

Any suggestion on what I have mistaken or missed?

Thanks a lot.



  • Morning Maurizio, could you post your stage targets, Fo and Q

  • Hello,

    The modified Bessel curve has the following:

    1st stage, simple buffer with RC input 3k32/4n7, fc 10200 Hz, Q 0,707

    2nd stage, SK 2nd order fc 15418 Hz, Q 0.531

    3rd stage SK 2nd order fc 17326 Q 0.926

    Due I have to use plastic capacitors, I tried to use only the values we have in warehouse chosen between 4n7, 6n8, 12n, 22n, 47n and 100n, having so limited combination to achieve the right Q values; I have the solution which gives the values of resistors having set C1, C2 and Q; the procedure is got from the formula given for unit gain Sallen-Key at the point A.1.2 of sloa049b.

    Essentially, respect to the Bessel values, I reduced the frequency of 1st stage and used components value I have for 2nd and 3d stages modifying a little both Q and fc.

    Hope this can make your understanding clearer.

  • Hi Maurizio, 

    The unity gain Sallen-Key circuit has the capacitor in the positive feedback loop rather than the negative feedback loop. 

    TI has a resource to help design filters based on the parameters you mentioned and can build out the schematic: 

    Please let me know if you have further questions.
    Thank you!

    Best Regards,

  • That is correct Ashley, 

    I don't see SKF design requests very often, this give me a chance to see if I can get my old tool working, It appears I did, 

    1. This is adjusting slightly for the GBP

    2. Getting best fit to standard RC values

    3. Sequencing the higher Q stage 1st after in the input LP, should help integrated noise

    4. Running +/-2.5V for now, you handle the DC offset or AC coupling for single supply

    This looks pretty reasonable, 

    And the output spot noise, I could not get this to run - had quite a lot of sim problems with this, the current online model did not run at all, I used the 2003 Bill Sands Model (in the TINA V11 library) to get the AC response, but had to switch to Trapezoidal and Davis KLU solver to get that to run

    Anyway, here is the file - these RC in theory should give you lower integrated noise than the equal C flow you were using - That came out of a lot of work I was doing over 10years ago, barely remember it - but I did find the design file. Eventually, once you get a PSpice thing running (maybe, the model seems suspect) - you can compare what you have to an RC set up like this for integrated noise, 

    SKF 5th order filter.TSC

  • Thanks for your reply....

    indeed, all literature I have read told me to use higher Q as a last stage to avoid saturation of the middle stages .... anywhere, I have limited resources of capacitors then I cannot use your circuit.... but I can use it for reference as m and n factors to be used.

    Another thing is the GBW: in my final stage, the theoretical GBW shall be not less the 1.6 MHz while I have 1 MHz; could it be this the cause of so big variance in the real circuits?

    Actually, if I reduce the 47nF in the attempt to increase the impedance "seen" by the OPAMP output, I obtain that real measures are closer to the PSpice values and between boards; now, the difference between the 2 boards is less than 1 dB (before was about 2 dB between 2 boards).

    The circuit has changed to the sch below


    Then I am still asking: the OPA348 has the unit GBW of 1 MHz, could it that the cause of the difference with PSPICE and between boards? 



  • The filter workbench could give some helps but it provides solutions with values I cannot manage, and it cannot select the OPA348, that's my opamp.... 

    Having the feedback in the positive loop is worse than in negative loop? Because its effect on the output resistance of the OPAMP?

    In this case, lower values of C2 are better than higher ones, aren't they?

    Thanks for help


  • Ok, I spent some more time on this so might as well share some results - 

    1st I tried to find a closely related part that might run successfully in my V11 TINA version, the OPA378 is a similar speed but chopper - it also crashed on noise (TINA shut down) darn - I remembered I had V9 on my 2nd computer - voila, it works in V9. Before I leave the OPA378 version, here is that spot noise stage by stage where you can see the final output is cutting down the high peaking of the intermediate higher Q stage (in your case, the input real pole will protect against clipping incidentally)

    Then back to the OPA348 version in V9, it of course shows a lot more low F noise not being a chopper, can't really see any higher F detail here, 

    So lets focus in just the Vout and run integrated noise on the OPA348 version, well through 100kHz it is still going up, might want a post RC to limit the poor stop band rejection in SKF

    Now lets modify these RC back to your settings (same filter shape, quite different spot noise detail) - lot going on when you do that, you had a dominant noise in that 1st R in the real pole - think I will leave the 402ohm version in there for now to emphasize the Q sequencing thing, well yes, about 50uVrms now through 100kHz, way over the modified version at around 12uVrms

    So you may not care about this level of detail, but I also sat through many hours of filter classes with the usual ascending Q thing - that was all written in the 70's by professors (Dr. Budak and Dr. Huelsman are the ones I lean on for this)  who never themselves developed or introduced a single op amp - nor did they do much lab work on tuning these things in. Surprisingly, the actual op amp development world has moved quite a long ways past circa 1970s textbooks, but none of that has been updated. In any case, you say everything you have seen says to ascend the Q's - well if you read this, can't say that anymore - incidentally, I have been told that the ADI tool now descends the Q for this improved dynamic range. When I was doing all of this, I was trying wring a bit more "free" dynamic range out of any given design target and device selection - and it does work, but may be beyond the scope of this discussion,

  • Thanks for your helps, 

    Everything has helped me to figure out what was the problem.

    It was not related to parameters or topologies or characteristics, even if some concepts have helped to have a better design.

    All the issues were due the boards are prototypes and samples with SMT components but manually assembled. To make the filters stable, it is used the plastic capacitors and the people who welded those components were not trained enough. Then, sometime they used lower temperatures and shorter time, sometime higher temperature and or longer time.

    The capacities were affected but it is not easy to measure on board and they will be destroyed if reworked.... 

    So, after guessed something like that and set the right condition for soldering, the variance between boards has been back to normal due tolerances.

    It maybe useful a quote from Sir Conan Doyle when the impossible has been eliminated, what remains, even if improbable, shall be the true.

    Thanks again for having increased my skills on the matter.