Using switched capacitor filter as a variable cut-off anti-aliasing filter

You can do that, but it's expensive, and the better parts have been disappearing lately, e.g. the LMF60 and LMF100. Also they're noisy as can be, and you have to filter out their clock feedthrough, so they aren't useful over a really wide range.

The usual method is to use a fixed filter and sampling rate, then filter in software. That's cheap and flexible, and as a bonus it reduces the LSB/sqrt(12) sampling noise.

Cheers

Phil Hobbs

Switched-capacitor filters alias, and have other switching artifacts. I wouldn't say "don't use", but rather "use with caution".

Can you even get them any more? They kind of lost most of their market niche when digital stuff became small and cheap.

If you have any influence over the sampling end of the system, and if you really do need to have a sampling rate that is ultimately variable, consider sampling at a high rate with a fixed, simple anti-alias filter (if needed at all), then filtering and decimating further in digital-land.

Unless you can really justify the need for anti-aliasing, read this (and then you will!):

I've found switchcap filters to be noisy. And they alias everything available, including power supply noise. And they make output spikes at the clock frequency. So you need real lowpass filters at the input and at the output. But they do work in some apps.

Do you need continuous frequency variability? You can make an active filter and switch resistors with some analog multiplexers.

Can you oversample and process digitally? Then you'd only need one fixed filter.

Den onsdag den 1. juni 2016 kl. 17.09.01 UTC+2 skrev djt294:

quency cut off range. The highest frequency will most like be a few kiloher tz (3K-5K). The first thing that came to mind was a switched capacitor netw ork. It seems to fit the application. My other options seem a bit more comp licated (switching in different values of loop components). I have not desi gned using switched capacitor filters and was wondering if anyone has exper ience in using switched capacitor filters as anti-aliasing filters and the issues that came up in its implementation. Any other ideas regarding a vari able frequency anti aliasing filter implementation topologies would be grea tly appreciated.

why not sample much faster and do it in software?

-Lasse

How does it work to do an active filter and change gains with DACs? I know it's POSSIBLE, I just don't know if it's a good idea.

I think everyone who's commented on this so far has mentioned sampling fast and filtering in digital-land. There's a clue there.

A state-variable filter works fine with dacs to program the integrators.

You could use digital pots, in a sallen-key or some such simpler filter, to program the frequency.

Hmm. And if you did it right you'd only need to change the frequency- setting gains -- the damping ratios could be fixed with resistors, at least until you started bumping into the op-amp phase shifts.

You'll bump into the R-2R (or whatever network they use internally) rolloff around the same point (100s kHz), but for a filter in the low kHz, that's perfectly fine. It's a good match.

Honestly, I think I'd even do it in digital... even an AVR can crank enough samples and MACs to do it. Continuously variable IIR filter coefficients might be an, interesting computational exercise, or a fairly easily synthesized/rescaled FIR filter can be used at the expense of more MACs. Otherwise, just get a bog standard ARM, or a proper DSP even.

Tim

The other goofy analog thing you can do is use an analog switch as a duty-cycle-controlled resistor, instead of a DAC.

After all, it *is* a lowpass filter.

It'd get you back to worrying about aliasing, and switching noise.

But yea, I see your point.

In low volume yes. If the controller does bit banging you can use the filter as S&H and get rid of the spikes:

Skeptical. High pole non-Butterworth will foul up phase at the upper end and distort the signal.

For multiplexing you need a filter for every channel. Otherwise switching channels will be very slow because you have to wait for the filter to settle.

MfG JRD

If you're worried about phase, the proper filter model is Bessel instead of Butterworth.

It wasn't an anti-alias filter but had to make a variable high-pass with near-0 and flat passband and >70db attenuation an octave down - was for measuring speaker distortion, put a tone into the speaker then measure the 2nd harmonic and up. What I did was use a LT1068 with a 100-1 clock/center ratio then to get a variable clock used a 74HC4059 divide-by-N counter followed by a flip-flop to square it up, driven by a 8mhz clock from the PIC that was controlling it. Also had a MCP23S17 port expander to drive the 16 inputs needed to set the '4059 "jam" inputs. Frequency set resolution wasn't all that great above 1khz (at 10khz next step was 13.3khz) but worked great for my app. Although I was just feeding a x10 amp and peak detector into a 10 bit AD, didn't notice any noise, read 0 with no input signal - about 80db down, not bad. With a 100-1 range it's easy to filter clock noise with just a 6db low pass.

Nice thing about the LT1068 is LT has a nifty app where you tell it what you want and it calculates all the values, worked perfectly although I did actively drive the bias since it was a high-pass, the 1uF bypass they suggested didn't begin to cut it.

Although.. as others have mentioned, beware of switched cap filters, I probably got lucky and it wasn't for audio just measuring. Since OP mentioned 5khz max suspect that app isn't audio either. Others mentioned digipots but be careful there - most have horrible absolute tolerance and will require calibrating if used in a filter.

- Terry

although there will be howls of "yuck", this is an old school approach: use a transconductance amp to build variable filters. According to the data sheet these are useful over several decades of frequency, although your more demanding requirements may knock that down some.

LM13700s are fine for few-kilohertz stuff. Long ago I built a super wide ra nge PLL using a VCO with an exponential V->F curve and OTAs for the loop fi lter. The OTA's bias came from a F-V converter, so the loop bandwidth was a constant fraction of the operating frequency and the damping was constant.

Cheers

Phil Hobbs

he only needs 5-6 k

Sure. I was sort of agreeing with you, except that I'm not sure what the sk irts will look like. Getting small shape factors (BW@-60dB/BW@-3dB) in an a ctive filter requires that the gain hold up reasonably well some distance i nto the stopband, iirc, and the LM13700 craps out at 500 kHz typical, even at max bias (again iirc).

I very rarely do anything fancy with active filters, so I'm not as sure as usual. ;)

Cheers

Phil Hobbs

Hi all,

I mised the OP, so I'm not in the right place in this thread, sorry.

Using SC-filters allways keep in mind that SC-structures allways are sampling devices that needs analog anti-aliasing filters before! Sure these AA-filters could be more easy because the highter sampling rate used to clock the SC-filers, but forgetting these may create very surprising results.

Marte

There is a switched resistor design at:

The details of how it works are in Section 6.4 of the instruction manual which can be downloaded by following the links from: