ADS1230 - quality of bypass cap for linearity improvement

In oversampling data converters, and especially in high gains everything in the signal path must be high quality to ensure the best performance.

In the case of the bypass cap for the ADS1230, we recommend ta poly capacitor for two reasons. First, the dielectric absorption of many ceramic caps can cause a long setting time to get to the signal's final value. Second, in most ceramic caps, there is a small (but noticeable in high gain) piezo effect. Any vibration, even coming from audible noise can be converted to an electrical signal that will be read as noise in the ADC. These effects are seen in many types of capacitors, including C0G ceramics.

Joseph Wu

Dmitry,

The quantitative is effect is hard to determine. Part of the problem is that the effect is not a non linear behavior, but it is a slow settling problem. However, it can look like a non-linearity.

For example, using an ADC in G=1, full-scale range of +/-2.5V, taking 11 equally spaced points in the range, your input would go +2.5V, 2.0V, 1.5V, ....etc. If you took these measurements at 100ms intervals, your INL measurement would be worse than if you took these at 15 second intervals because your input voltage would have time to settle. Note here that the INL value will vary with time, but also it varies with the direction of the measurement. You will get an opposite INL if you start from -2.5V instead of 2.5V.

This settling problem can also apply to the reference too. If the reference voltage moves during the INL measurement then the drift of the reference voltage will look like non-linearity in the measurement (if you have been assuming a constant reference). However, if you accurately measure the reference for each data point, this may not be a problem.

To model dielectric absorbtion, the equivalent circuit of a capacitor is a main capacitor in parallel with a resistor in series with a capacitor. In one of Bob Pease's old articles about "Capacitor Soakage", He describes a 1uF capacitor as 1uF in parallel with 1G resistor in series with a 0.006uF capacitor.

I found another article ("Advances in Class I C0G MLCC and SMD Film Capacitors" coming from the Kemet website) where the dielectric absorption is listed as a percentage of the original capacitance:

X5R/X7R/X5U    1.0 - 1.8%
C0G        0.5%
PPS        0.02 - 0.05%

While I don't know what the series resistances or the time constants would be, you can get a feel for what the relative dielectric absorbtion would be for the different materials.

I imagine that in a gain of 1, the effects are not very noticeable, especially with a even a second of wait time. However, with high gain, I know that the INL cannot be accurately measured without the PPS capacitors. Sorry, but I don't have any data to say how much the INL measurement would be disrupted (especially by varying the sample time, and the capacitor type).

Joseph Wu