Low analog lowpass + ADC vs High analog lowpass + ADC + low digital lowpass

Question

I'm doing EEG data collection and my amp forces use of an analog lowpass filter before the ADC. I'm fairly sure that I won't be interested in phenomena associated with frequencies of 30Hz or higher, so it strikes me that the most disk-space efficient thing to do is to analog lowpass at 30Hz (it uses a Bessel filter) and tell the ADC to sample at 60Hz. However, disk space considerations aside, I'm curious whether there is any difference between this strategy and, say, setting the lowpass to something really high like 250Hz and sampling at 500Hz, then once digitized and doing analysis, applying a digital lowpass at 30Hz. Would the double filtering induce artifacts? (i.e. Should I opt for the low analog lowpass for reasons other than disk space?)

Answer 1

I'm curious whether there is any difference between this strategy and, say, setting the lowpass to something really high like 250Hz and sampling at 500Hz, then once digitized and doing analysis, applying a digital lowpass at 30Hz

Good idea. This is how oversampling ADCs work, sampling at a high rate using a poor quality converter (even a 1-bit converter), and then using digital filtering to downsample to the desired value with better accuracy. Digital filters can have a much sharper cutoff than analog filters for essentially no cost, reducing aliasing that would otherwise get into your recording. It can also improve your signal-to-noise if the analog signal isn't already restricting this.

Would the double filtering induce artifacts?

Not unless you design the filters poorly. :)

Answer 2

A single lowpass at 30Hz with a sampling rate at 60 Hz should provide accurate data, and filtering a second time should not manipulate any components of the signal at 30 Hz and below.

However, be sure that if you are converting between sample rates that you don't need to interpolate between samples- 500 Hz to 60 Hz should be fine, since it's an 8:1 ratio.

Answer 3

I'm curious whether there is any difference between this strategy and, say, setting the lowpass to something really high like 250Hz and sampling at 500Hz, then once digitized and doing analysis, applying a digital lowpass at 30Hz

Neither is correct.

Filters have non-zero transition width, and Nyquist sampling requires sampling at a rate above (not at) twice the highest spectrum frequency to avoid aliasing.

So sampling at 500 Hz after using a cheap non-steep-transition 100 Hz low-pass filter, then using a higher quality 30 Hz digital low-pass filter, and resampling to some rate ABOVE 60 Hz might be a suitable possibility that lowers the cost (or gets your signal farther away from a ripply or phase distorting transition edge) of the analog front-end filter in exchange for supposedly even lower-cost computational power. Or set the filter somewhat below 30 Hz and sample at 60 if you don't need that top end of 30.