I'm currently working on a design with an adc3424 ADC with 125 MHz sampling frequency. This is my first design with an ADC and I assumed picking an anti-aliasing filter (AAF) would be a trivial task. Linearity or passband ripple is not a critical factor. How come it seems to be a rare thing with AAFs with corner frequencies above 10^5 Hz? Something like 60 MHz would make sense here.

On digikey, there are not many active AAFs. Other filters with higher corner frequencies do not seem to be for anti-aliasing. Could anyone direct me to where I can find what I'm looking for? Thanks!

  • \$\begingroup\$ You're thinking Nyquist frequency so 125 MHz sampling rate results in a 62.5 MHz Anti-Aliasing filter, right ? But that 62.5 MHz is a very hard limit so that filter would need to have quite a steep roll-off. In practice, instead of letting the sampling rate determine the filter, it is often better and easier to let your input signal set the filter. So what is the BW of your signal ? Also note that at the end of the datasheet of the ADC there are some application diagrams. Note that only for the low input frequency design there is some low-pass filtering: 39 nH and 22 pF. \$\endgroup\$ – Bimpelrekkie Sep 5 '17 at 6:27
  • \$\begingroup\$ So depending on your input signal you might not even need an AA filter and if you do then a simple LC 2nd order filter (made from discrete components) might do the job. \$\endgroup\$ – Bimpelrekkie Sep 5 '17 at 6:28
  • \$\begingroup\$ I will study what the actual BW is and go from there. Thanks alot! \$\endgroup\$ – toxUP Sep 5 '17 at 7:22
  • 1
    \$\begingroup\$ You don't pick one, you design one (or steal a design, or scale from another frequency) And it won't be an active filter, but a passive L-C filter. \$\endgroup\$ – Brian Drummond Sep 5 '17 at 9:56
  • \$\begingroup\$ Will do! I'm exploring Scilab for this now. Why LC and no simply RC? \$\endgroup\$ – toxUP Sep 6 '17 at 1:03

Oversampling is necessary to increase the quality of the waveform and signal-to-noise ratio; so with 125 MHz I wouldn't go beyond 20-30 MHz of bandwidth, maybe the 40 you mention, not the 60 or sharp 62.5... especially if the anti-aliasing filter has 1 pole only.

Choices: passive or active filter.

Let's assume that the ADC is driven by an OpAmp buffer, possibly doing level shifting and offset adjustment. Its output impedance at several MHz increases to let's say a fraction of ohm to a few ohm if it is already a good OA (150 MHz GBWP minimum).

Passive filter: it is stable it is simple, and alas has only 1 pole; if you do not expect big noise above the chosen cutoff frequency, then it is ok for the purpose of anti-aliasing. 47 ohm and 100 pF give 33 MHz. This circuit has linear phase over a portion of its bandwidth, up to about 20% of cutoff frequency. The change of OA output impedance is "absorbed and masked" by the 47 ohm.

You may increase the order as suggested by @Dan Mills and @Bimpelrekkie: take into account filter tuning for resonances, maybe adding some larger resistors in parallel to inductors.

Active filter: there are good architectures (Sallen-Key, Multiple Feedback, etc.) to build active filters; . Once the architecture is decided you may calculate components with high accuracy once the parameters of the desired transfer function have been decided: Bessel, Butterworth, ... including Equiripple (that I like). However, it is never clear what is the effect of limited gain of the OpAmp and parasitics. I posted a question (OpAmp performance for Sallen-Key filter); see @peufeu answer, very useful. Link to Analog Device design tables Depending on chosen filter, keeping a conveniently nominal gain of e.g. 1-2, you need a lot of margin on the open loop OA gain: with 30 MHz of cutoff, probably you have to go to 150-300 MHz of GBWP, and this necessitates RF techniques, risk of oscillations and ringing ... all things that are worsening the quality of what your ADC can do. I usually use FET input OpAmps because they do not create problems of offset with OA input bias/offset current and large resistor values, that are changing when selecting different cutoff values (e.g. AD8065, AD8039 less stable with temp.)

  • \$\begingroup\$ So when I plot your 1 pole 47ohm 100pF filter into ltspice, I obtain the following plot. Why isn't the phase linear? \$\endgroup\$ – toxUP Sep 8 '17 at 6:47
  • \$\begingroup\$ @toxUP Phase is sufficiently linear up to a max frequency: I was abusing wording while writing "linear by construction", I amend it. Please, see page 26-27 of this file montana.edu/aolson/ee503/EELE503_filters.pdf: in the example a 1 MHz cutoff RC is usable up to 200 kHz. Linear phase is ensured by Bessel and Equiripple active filter architectures (analog.com/media/en/training-seminars/design-handbooks/…): in this Analog Devices pub there are the tables of coefficients I normally use. \$\endgroup\$ – andrea Sep 8 '17 at 8:21
  • \$\begingroup\$ Thanks for a thorough reply! The phase just seems quite non-linear close to DC. \$\endgroup\$ – toxUP Sep 8 '17 at 15:19

Usually for an RF application like that you would just design a LC filter network having whatever corner frequency and stop band you decided you need.

Plenty of software to do this, or hit the "Handbook of filter synthesis", remember to consider the finite Q of real components and the parasitic C of the inductors, but something like 5th order at 40MHz or so 0.5db ripple Chebyshev should be in the zone I would have thought, and should be buildable with C0G caps and SMT RF inductors.


I know it is a old question, but it comes up fairly high when searching for anti-aliasing filters in the RF region, so I thought it would be a good choice.

I use a combination of the Elsie Software from Tonne Software and AppNote 837 from Analog Devices to design my LC filters.

While the student version of Elsie is restricted, it still allows you up to 7th order filters;And when you combine that with the layout from AN-837(Showing how to split the caps for balance), you can easily create the filter you need.

I've tried the various online filter design programs, but I have yet to find one that has all the bells and whistles that the free version of Elsie has.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.