How to make a sharp anti-alias LP-filter?

If I read your post right your highest frequency of interest is 110kHz, and you are trying to get 16 bits worth of rejection at Nyquist. But if your highest signal frequency is 110kHz then then you shouldn't really have to worry about signals between that frequency and the first alias frequency, or 390kHz. Is there something going on in your environment outside of sampling that makes 250kHz special?

A passive LC filter won't gain you any attenuation over an active filter with a like number of poles, and at those frequencies it probably won't be smaller. Even if you can put your 98dB point at 390kHz you're still talking about 10 poles for a Butterworth filter, and probably not too many fewer for a Chebychev.

Is there any way that you can push up the sampling rate? You should be able to get 12-bit ADCs that will do in excess of 1MS/sec; if you sampled faster then you could have a filter/decimate step in your digital hardware. The overall cost with the faster sampling rate may be lower than fiddling with such a sharp filter.

Forgot to mention -- there is a 5-pole continuous (_not_ switched capacitor) active filter out there. I just can't remember who make it.

Sorry, I meant a 16-bit SAR, not a 12-bit, sorry.

And you are right about 390 kHz, I didn't think of that. So I need

-98 dB at 390 kHz ideally.

I must check if it's possible to select a different ADC with higher sampling frequency.

The 16-bit part makes it harder to find a higher sampling rate; there are parts out there but they're expensive.

While you're pondering attenuation vs. frequency you may also want to ask if you need 98dB of alias rejection at the high end of your frequency range -- you may find that you need high accuracy at DC but that various other forces conspire to screw up your accuracy at high frequencies anyway, allowing you to relax the spec.

And check the LTC1563 -- it's a monolithic 4th-order filter who's frequency you can set with one resistor; with another resistor you can turn it from a Butterworth into a Chebychev. With two or three of those you should be able to build your filter pretty quickly.

You might want to look at a combination of active and passive filters.

Tam

Sounds like an active Chebyshev to me ... given the limited information.

A sixth-order elliptical will do it, with 1dB or less passband ripple. We used to do things like that with op amps, but beware that you need to use a pretty fast op amp to get good performance. Presumably, you want distortion products down 100dB from full scale, too, or you're wasting effort on the AAF part. There are some freeware and trialware filter design programs out there which could help you explore both active and passive (LC) filters. The free filter program from aade.com is very decent for LC filter designs.

But for a 110kHz bandwidth and 16 bits, could you just use a delta-sigma converter and make your antialias life much easier? I certainly understand that there are times when a delta-sigma isn't appropriate, but maybe it's at least worth asking.

Cheers, Tom

I think you are thinking of Linear. The make a filter that can be shifted between 500KHz and 1MHz cut off.

That's quite steep. You may want to consider sampling faster, with less bits, and do some of the filtering and downsampling digitally.

Kind regards,

Iwo

I found a free filter design program at Linear called FilterCAD. It suggested a 1596-7 filter IC. It is a "Linear Phase, DC Accurate,Tunable, 10th Order Lowpass Filter" :-) You set the cut of frequency with a resistor.

That is in the class called "brick wall filters" almost always implemented digitally. The high attenuation is in and of itself is difficult to realize practically. In the real world you would want to do about 24 dB of this filter in the analog world and do the rest digitally.