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.

THS4500: Fully Differential Butterworth lowpass filter

Part Number: THS4500
Other Parts Discussed in Thread: THS4561, THP210, OPA1637

We are looking for a tool that allow us to develope a fully differential 4º order low-pass butterworth filter but FilterPro doesn´t allow it.

I´ve written in the chat (CS1413091) and  I`ve been redirected to this forum.

We need a 1.8KHz bandpass with a 0.01db ripple and a 40db attenuation over 14.34kHz. Our design can be checked in the pictures. Dual power supply. VDD:+-3v3, G=1 

  • Hi Alberto,

    We need a 1.8KHz bandpass with a 0.01db ripple and a 40db attenuation over 14.34kHz.

    Do you need a lowpass filter or a bandpass filter?

    Kai

  • Hello Kai, We need Low pass filter, but we need to ensure a 40db attenuation to 14.34kHz too. I attached the bode diagram and the poles situation. Let me know if you need more info.

    Thx!

  • Well Alberto, I have the tools to respond on this - this might not be the final solution, but food for thought this fine Thanksgiving morning. 

    THS4561 4th order low F chebychev.docx

    and then the 3 TINA files, 

    each stage seperately and then together, 

    THS4561 higher Q stage MFB.TSC

    THS4561 lower Q stage MFB.TSC

    THS4561 4th order MFB.TSC

    Oh and I see you needed the 14.3kHz rolloff, if I get time, I will add tolerances to the RC and run monte-carlo that - the routine here goes to 1% R and 5% C values, but you can buy them in tighter tolerance, what is your intention? Better be pretty tight if you really want that 0.01dB, but you won't get it I would think - way too tight. 

  • Well Alberto, I have the tools to respond on this - this might not be the final solution, but food for thought this fine Thanksgiving morning. 

    THS4561 4th order low F chebychev.docx

    and then the 3 TINA files, 

    each stage seperately and then together, 

    THS4561 higher Q stage MFB.TSC

    THS4561 lower Q stage MFB.TSC

    THS4561 4th order MFB.TSC

    Oh and I see you needed the 14.3kHz rolloff, if I get time, I will add tolerances to the RC and run monte-carlo that - the routine here goes to 1% R and 5% C values, but you can buy them in tighter tolerance, what is your intention? Better be pretty tight if you really want that 0.01dB, but you won't get it I would think - way too tight. 

  • Oh and Alberto, if your real question is what public tool will get you to ok results, you might want to peruse these two articles. 

    https://www.edn.com/testing-op-amp-tools-for-their-active-filter-design-accuracy-and-dynamic-range/

    https://www.edn.com/active-filter-design-tools-shootout/

  • Hey Alberto, I had a little more time and thought I would illustrate the slight integrated noise improvement that is possible by putting the higher Q stage 1st in a multi-stage active filter design. If there is any chance of clipping ue to overshoot you would not do this and at a gain of 1 in the overall filter it is a very modest improvement - as the overall filter gain goes up, the improvement becomes more significant. 

    Also, to really see this there should be some finite single pole RC after the last stage, I put it at 100kHz here

    Anyway, here is comparison with the lower integrated noise being the higher Q stage first, about a 20% drop here. 

    Did quite a bit of work on this topic building the Intersil online filter designer, some of that is published, essentially the idea is to rolloff the higher noise peaks coming out of the higher Q stages by lower Q following stages. 

    https://www.edn.com/advanced-considerations-for-gain-and-q-sequencing-in-multistage-lowpass-active-filters/

    And here we illustrated this modifying an extant ADI 8th order example for improved integrated noise, online this has disappeared, but here is my copy, 

    3000.Testing Filter Designer for gain and Q sequencing AN1580.pdf

  • Michael, thank you so much about that extended analysis. Let me check your anotations and work on it. We had chosen a butterworth topology in order to improve the bandpass ripple, but if you think chevishev is the better option we wiil analyse it.  Have a good week! and thanks again

  • Well Alberto, couple of things before I go on to other requests, 

    1. Your 0.01dB rolloff is purely academic, no implementation can do that 

    2. I tried a few different pole location tools, the ADI tool will only accept a min -0.1dB entry as anything lower is a waste of time once you start putting component tolerances in

    3. Using the ADI tool, I get these targets to try and meet your general needs, and here I was adjusting to stay 4th order - moving out the -0.1dB point as far as I could to hold flatter at 1.8kHz, and increasing the atten at 14.34kHz until it clicked over to 5th order

    4. The ADI tool has gone to a descending Q stage strategy, I had sent my articles to the guy doing that some years back and apparently it made sense to him, 

    5. The THS4561 implementation is still showing instability out there, in fact far lower than 10deg phase margin in that 2nd stage - oh and a couple of comments on the RC solutions my flow reaches - I limit the max C to 54nF as being the max value for C0G 2% with ok cost, that might have moved up since I updated this last and, I move the standard R values around for best Q and Fo fit, letting the DC gain go off one E96 value if need be. 

    6. There is a newer FDA that we can try, This is out of the precision group where there are two versions - a super DC precise version and a less expensive audio release. THP210 and then the OPA1637. They appear to have the same TINA model, so lets drop in the THP210 model. Darn, they reversed the supplies, fixing that and taking out the phase margin improvement elements, no resonance but less stop band rejection way out, 

    7. And then looking close in, an odd non flatness - I have looked a little to see if I have a component solution error, can't find anything yet - need to move on. 

    and this file, good luck - 

    THP210 4th order MFB Butterworth shape.TSC

  • Well this is why I respond to these things, I need to exercise these tools occasionally to remember the flow - 

    The 54nF in the inputs was for an op amp, I need to cut that in 1/2 to 27nF (and my actual max C in the flow is 47nF right now) for an FDA, so here I reran the design for the THP210 device, it actually found different RC for the 2nd stage that got the nominal DC gain back to 0dB, and then I zoomed in -

    -0.006dB drop at 1.8kHz, again this nominal, Monte Carlo will show more spread, if I can get that to run - have this at 0.5% R's and 2% C's right now, it would not run for me, but crashed - I usually have to restart my computer to get better luck - too many other things open to do that now, 

    And out at 14.34kHz - about -44dB atten

    And now if I look out past that 100kHz noise limiting RC I put in, exactly -0.01dB loss at 1.8kHz, how serendipitous

    So this looks like a reasonable nominal design, file, 

    1663.THP210 4th order MFB Butterworth shape.TSC

    Now for Kai - 

    The descending Q, ascending gain thing I did at Intersil generated a lot of pushback from the RF guys who suggested I was ignorant of the Friis equation. 

    While there are many things I am ignorant of, but that is not one of them 

    The Friis equation works (put most of your gain 1st in RF chains) but is kind of spot or noise figure oriented. If you think about final integrated output noise in a mult-stage LP filter, the highest Q stages have the highest noise gain peak inside the filter. Putting that highest Q stage first will in fact be satisfying the Friis equation from an integrated noise standpoint. And the simulations bear that out. 

  • MOrning, Michael!! I can´t open the file above, can you reload it?

  • oops, I need to save out of my V11 to V9 type, 

    Here you go, 

    7416.THP210 4th order MFB Butterworth shape.TSC

  • Oh, and did shutdown everything and restart, got the Monte Carlo to run, 200 cases with 0.5% R and 2% C on Quassian distribution, About +/- 0.2dB range at 1.8k. It is easy to ask for .01dB flatness, pretty much impossible to get reliably.