Design of low pass filters for data acquisition systems

[uoficowboy]

Hi - I tend to work on a lot of data acquisition systems that need low pass filters in front of analog inputs. Typically this looks like a one or two pole active filter going into an ADC.

My questions are as follows:

How do you determine how many poles you need? I mean, I understand that more poles will give you greater roll off (-20db/decade/pole, I believe). But is there a good way to determine what your actual needs are?

How do you determine the cutoff frequency? If you are sampling your ADC at 1KHz, does that mean you want your filter to have a 3db frequency of 500Hz (I believe that is a proper application of the Nyquist–Shannon sampling theorem?)

Lastly, how do you determine what you want your Q to be? Or, equivalently, how do you determine what you want your damping factor, seeta, to be? I mean, I know what different Q/seetas look like, but what makes the most sense for this application?

I suspect these are pretty open ended questions and probably fairly application specific - but any light you all can shine on them would be very much appreciated!

 

[flatulent]

First, in practical systems you sample higher than Nyquist. Then you examine the levels of the input signal in the frequency range that would alias down by the sampling process into the desired signal range. Then you determine the attenuation. You generally want these unwanted, out of band signals to be reduced to less than 1 LSB or lower. The cutoff frequency is determined by how much attenuation you can take at the highest desired input signal frequency. Then you look at the different frequency responses (or do some calculations) to find the number of poles. This is an iterative process.

 

[LvW]

Perhaps this link can help a little:

**broken link removed**

Good luck
LvW

 

[LvW]

You also can try AN699 from Microchip.