In the video Driving SAR ADCs: Distortion caused by large Rfilt, I discuss how ADC distortion can be degraded when larger values for Rfilt are used.
Let's look at an example, using the AD7980 ADC and a few values for Rfilt. The relevant equations reviewed in the video are:
Vvar is a very small voltage - small enough that we can ignore it when deriving our equation for ivar. (We can assume that Vvar is so small, that the voltage across Cvar = Vin.) But, this very small number becomes a very big problem when we're trying to get the best performance out of a 16+ bit ADC, where an LSB can be in terms of uV.
Using the AD7980 as an example - Cvar is approximately 1pF. This is not a value that you can find in the datasheet, but it's a parameter we've measured to build a distortion model for the AD7980 in the Precision ADC Driver Tool. Calculating THD at 10kHz with a Rfilt = 200:
Vvar = Vin * 2πf * Cvar * Rfilt
= Vin * 2Π * 1e-12 * 10e3 * 200
= Vin * 12.6e-6
THD = 20*log10(Vvar / Vin)
= 20*log10(Vin * 12.6e-6 / Vin)
Something to keep in mind: There are a few things I'm glossing over with these equations, and they're not intended to be used to calculate distortion from a specified non-linear capacitance value with absolute certainty - especially since you won't find that parameter specified in the datasheet. But, they do illustrate a trend that will hold - If you're observing THD caused by Rfilt interacting with Cvar, you'll see the THD degrade by 20dB when you increase Rfilt by 10x. And, you'll see the THD degrade by 20dB when you increase the input frequency by 10x.
The plot below was built using Cvar = 1pF, Rfilt = 20, 200, and 2k, and the equations above. The horizontal line represents the specified THD of the AD7980 at low frequencies, and serves as the "floor" for estimated THD - in otherwords, smaller Rfilt improves the THD performance, but it limited the other distortion contributions of the ADC. (See this .xls file for data and calculations used to build plot)
We can also look at an example using the AD4000, which has a lower value for Cvar (~0.5pF), and is therefore more tolerant of higher Rfilt values:
Keep in mind that there are several sources of distortion (driver, ADC, error from kickback not being settled) - this is only one source of distortion, and can likely be disregarded if it is not dominating the overall distortion of the signal chain.