That depends on a a few properties of the signal chain, for example:

1. What exactly does the conversion include? Is it a simple scaling operation (linear) or does it include nonlinearities (e.g. gain+offset or even a higher order calibration)?

If the latter is the case, then doing the conversion early in the signal chain will mean that distortion caused by the nonlinear calibration operation will be reduced by the following filters.

2. What's the magnitude difference between the physical units and the ADC units? If the conversion means that you're throwing away several LSB of the ADC output, it might be better to do it after the filtering.

3. Numerical issues. What are the allowed input ranges of your filters to avoid overflow, and would the conversion put the signal outside these ranges?


I had one project that deals with two different types of signal - one that is calibrated using gain only, and it's scaled to physical units at the end of the signal chain. The other one is calibrated with a nonlinear function, and it's scaled to physical units in the middle of the signal chain.