ADC Front end Filtering

Use your sound card in spectrum analyzer mode to 'see' what's there. all is free.

Then you might see that a simple/single LP is NOT going to be sufficient.

You can use octave, free Matlab clone, to completely simulate what you're attempting to do to verify you like what you'll get.

It's usually best to understand what you're doing. This may help:

It really boils down to just one thing: is the signal with filtering more or less representative than the signal without filtering. Sometimes the answer is that filtering improves things, sometimes the answer is that filtering makes things worse. The above paper should help you decide.

You need *some* sort of low-pass / bandpass filter in order to avoid having your measurements confused by aliasing, if there are any frequencies >= 1/2 of the sample output rate present in the signal you are measuring.

The filter may be implemented via analog means, or via digital means when using an oversampling converter (e.g. delta-sigma), or by a combination of the two

Do you have any information as to the actual frequency content of the current that you'll be sampling via the shunt?

The MAX11060 you mention has an on-board low-pass digital filter (starts rolling off at about 1/4 of the sampling frequency) so in order to get the 500 Hz bandwidth you want, you're probably looking at

They do say (on page 32) that the digital filter has no rejection of noise located near the harmonics of the 3.072 MHz sampling clock... and its common-mode rejection extends up to several hundred kHz but apparently not higher than that.

So, it might not be a bad idea to add a stage of gentle analog filtering - just a differential RC? - with a -3 dB point set well above your frequency range of interest. This would help ensure that any RF noise present on the line you're monitoring doesn't saturate the converter or generate spurious results.

If you're configuring the converter for (e.g.) 4 kSamples/second, and you want to decimate this down to 1k (for your nominal 500 Hz frequency range) you'd need an additional digital filter, implemented in software, to avoid aliasing during this decimation.

it's a delta sigma ADC it samples at several MHz and digitally lowpass filter and decimate the signal to the output rate

you can add an input filter but the filter is already there, just digital

-Lasse

Well, yes and no. It is quite common in control loops to have no anti- aliasing filters on the ADCs. If the externally-generated noise is low, then the cost of cleanly filtering out any possible aliasing signals is significantly increased phase shift that severely limits the attainable bandwidth of the loop.

When delay or phase shift matters and there's not a lot of high-frequency noise, it's often better to leave off the anti-aliasing filters and the corruption that they contribute to the signal.

Thanks for the link, I'm drinking it in.

sampling rate frequency is usually needed for ADCs?

it. My highest frequency of interest is 500Hz.

input of the part?

The part you are looking at has an internal digital filter. Since it sample s at 3.072MHz, it might be worth the trouble to put enough low-pass filteri ng in front of it to reduce any 3.072MHz content to a pretty low level to m inimise aliasing. Any leakage at exactly the sampling frequency could show up as a DC offset, and anything around 3MHz would show up as lowish frequen cy noise.

Tim Westcott assures us that many people don't bother, but that strikes me as risky. Any filter does introduce some phase delay, but if you only want to go up to 500Hz, a reasonable anti-aliasing filter isn't going to introdu ce enough phase delay to matter, over and above the phase delay already bui lt into the digital filter inside the ADC.

And don't forget a solid ground plane beneath the analog front end. jb

Ground planes are not a panacea. Often in high impedance systems driven guard rings and shields replace ground planes. Often nothing is used due to capacitance issues.

?-)

Some PCB materials are quite hygroscopic, thus varying the capacitance due to humidity. For this reason, you do not want a ground plane below a free running oscillator frequency determining components.

Of course, in a balanced circuit, much of the capacitance change due to humidity is canceled out.