Member #153141

Member Since: September 26, 2010

Country: United States

  • Just an FYI for anyone wanting this chip for accurate time-domain A/D. I have tried this and learnt the following the hard way:

    Figure 3 on Pg 6 of the datasheet -> it’s digital antialias filter rolls off too late. In practice I’ve found inexplicable errors up to half the sensitivity of the device if any input is in the 0.45 to 0.5 fs range. Theoretically it shouldn’t downmix aliasing signals here onto 0 Hz, but it does. I can even get it to hold a constant (0 Hz) offset by feeding it a strong sinewave at 0.5 fs.

    If you want to use this to make accurate time-domain measurements from sensors, then it very badly needs a Bessel analog anti-alias filter just before its input, to ensure there isn’t any noise in that deadly 0.45 - 0.5 fs band. There are some ok switched-cap filter chips for this - a break out would be nice. Integrating both so that the switched cap filter runs at the same clock as this chip would be better.

    Otherwise, you can use a low-order filter at far less than 0.5 fs, so that the roll-off by then is decent, and then implement a multirate filter on the data afterwards. This can get you better accuracy, since switched-cap filters aren’t all that good for low noise measurements. The expense is only getting a sample ~ 1kHz or so, But this still gives a hell of a much cleaner signal, which can then be used in feedback loop control of some high power actuator without making everything go haywire.

    If you don’t put the anti-alias filter in, there’s nothing you can do afterwards to get rid of the noise - it ends up aliased all across the passband, and makes the “24-bit” precision a joke.

    You can still get away without an anti-alias filter, if you use, say, a completely faraday-shielded sensor. (like a vibration sensor or microphone) and never let any radio interference at all onto the wire. But as soon as you arn’t shielding up to very high frequencies, your noise will come back.

    I’d really like to see a breakout board with I2C interface, and integrated antialias filtering. The filter can easily be variable so as to match the sample rate if it’s a switched-cap type. 14 bits would be more than enough (accuracy limited by the switched-cap filter). A sigma delta converter with an overall bessel response and a digital multi-rate filter would be very very nice for getting very clear time-domain samples. Would make sonar easy. Problem is no-one designs chips like this with flat pass-band group delay in mind.

No public wish lists :(