THS4551: Does the simulation file provided from the datasheet on the THS4551 have an error?

Hey John, 

I have been trying to reconstruct what was going with that circuit (also shows up on the first page). Nominally, as drawn it is not a 500kHz MFB? In fact that post RC at the ADC inputs is only a 360kHz F-3dB.

I looked through my development files and that 330ohm is there (as it needs to be to make and MFB filter). The attached file might have gotten changed at some point later - but the original circuit is not right either?

If you tell me what you are trying to do, I can work up the RC's pretty easily.  

Some comments on the extra elements here, this was an earlier file I found, 

1. The response to the FDA output pins is about 400kHz F-3dB, 

2. That 1nF across the inputs is intended to help improve phases margin without impacting the filter shape much, seems high to me now, but removing it does not change the filter shape. 

3. That 3.2ohm in series with the feedback caps are the internal trace R to the input side of the package

4. Those 5ohms inside the loop were intended to help phase margin, I found later that the feedback caps should go outside those 5ohms for better results. 

5. That 2nH in series with the final cap is self resonance, the next set of simulation was to model the switching going on insde the ADS and look at settling times. 

Michael, thanks for the reply!

I'm glad to see that you are also noticing some inconsistencies because I was starting do doubt my understanding of this circuit.

I'm trying to create a design using the ADS127L01 ADC with the output data rate (Fodr) at the ADC's maximum rate of 500 kHz. The ADC is a delta sigma and it has a sampling clock around 16 MHz, with a built-in digital brickwall lowpass filter at 0.4 * Fodr.

One of the key parameters for my system is AC amplitude flatness, so I was going to try to extend the 3 dB cutoff of the MFB + RC filters further out as compared to the sample circuit. I was hoping to have less than, say, 5% rolloff near 200 kHz. At the moment it is a little over 8 % rolloff at 200 kHz, but I also need a bit of headroom for my other signal conditioning circuits before the THS4551. I understand that this will decrease the effectiveness of the antialiasing filter. The input to the THS4551 would be single ended, with GND as the signal reference voltage.

Would taking the existing circuit and applying frequency scaling to the MFB filter components and RC be an appropriate strategy? I saw a TI app note sboa114--I think it is one that you actually wrote!--and it had excellent information for the design of a single-ended MFB filter, but I'm still getting myself up to speed with fully differential amplifiers.

Well John, normally folks go to Butterworth first for flatness, but you can actually say target a 0.1dB Chebchev at 200kHz and it will be very flat through that frequency - I would do this 3rd order with the MFB as the 2nd order stage and the last RC as the single pole stage. I do worry about using a chebychev as the next issue we were dealing with was the sampling edge settling interacting with the source impedance looking back towards the filter - I would suspect the cheychev might have more ringing than a butterworth, 

1. what input impedance do you want (real low will load stages, real high will create dominant R noise)

2. 5% down at 200kHz is -1dB

3. Using filterpro, that would require a 250kHz 3rd order Butterworth target - seem ok?

Yes that app note was the start for me, have gone a lot further since then in building the Intersil online active filter tool - that original app note had errors in the cubic coefficients I fixed later -need those to adjust RC for GBP. 

Hey John, 

Have not heard anything on this - I am off working on other things, but if you answered I could pretty quickly gin up a set of RC values, 

Hi,

Thanks for following up. I wanted to carefully consider what you wrote before responding.

A small ripple is acceptable; the filter doesn't necessarily have to be Butterworth as long as the ripple (i.e., maximum deviation) is small enough. 0.1 dB would certainly be sufficient. -- EDIT -- I was looking at the step response of some of the filters, and I'm wondering if that is another reason to avoid Chebyshev as it seems to have a lot of ripple. Since I will be looking at both time and frequency domain signals, maybe Bessel or Butterworth is the best choice.

1. With regards to input impedance, 10 kOhm would be ideal for the amplifier driving the filter for distortion purposes. I understand this will raise the noise. I think that there is enough noise headroom for that. Anything under 100 nV / root-Hz should be sufficient.

3. Yes

On a side note, I've tried to find an installer for filterpro, but it seems like TI is making it harder and harder to find. I know they are trying to restrict their tools to online-only, but the online tool doesn't seem to have as many features yet as filterpro did. Online-only also makes it hard to save calculations for documentation purposes.

Hey John, et al

So I did actually publish the corrected cubic coefficients not long ago. Steve needed content on Planet Analog and he had found some odd article about this where they were using an amplifier structure without an output stage -tying the filter elements directly into the comp node? Not very useful, so I went ahead and published some of the work I had done at Intersil - 

https://www.planetanalog.com/include-the-op-amp-gain-bandwidth-product-in-the-rauch-low-pass-active-filter-performance-equations/#

There is a pair of more recent articles continuing these things where I first discuss why GBP is so confusing, 

https://www.planetanalog.com/why-is-amplifier-gbp-so-confusing-insight-12/

And then a rational (as opposed to just guessing like FilterPro does) approach to setting required GBP

https://www.planetanalog.com/use-true-gain-bandwidth-product-to-estimate-required-margin-in-active-filters-insight-13/