Reporting ADC Noise Free Resolution from a Defined DC Record

A high-resolution ADC can produce a long digital word while several low-order bits fluctuate at a constant input. For slow measurement applications, that fluctuation may matter more than a sine-wave performance figure. Report the observed DC noise using a defined record and distinguish RMS variation from the full peak-to-peak spread.

Define the input and operating mode

Use the manufacturer’s recommended input-short or stable-input arrangement while preserving valid common-mode conditions. Record gain, reference, input span, output rate, digital filter, and any postprocessing.

Shorting differential inputs together does not necessarily mean connecting both to ground. The selected ADC may require a different common-mode voltage for valid operation. Confirm the arrangement before interpreting a noisy result.

Calculate both noise measures

The Analog Devices discussion of effective and noise-free resolution distinguishes RMS-based resolution from a peak-to-peak measure. Noise-free resolution is calculated from the base-two logarithm of input span divided by peak-to-peak input-referred noise.

Convert code variation to input units using the actual range and gain. For a hypothetical 2 V span and 20 microvolt peak-to-peak spread, the ratio is 100,000, corresponding to about 16.6 bits. This is a result for the stated observation and conditions.

Keep record length visible

Record the number of samples and elapsed time. Observed extrema can expand as a longer record captures rare noise excursions or drift. A short quiet interval and an overnight record should not be compared without acknowledging that difference.

Plot the time sequence as well as the histogram, retaining timestamps for any discontinuities or missing samples. A broad histogram caused by slow thermal movement calls for a different investigation from stationary random variation.

Include the real signal chain

After establishing the converter baseline, repeat with the actual source or sensor connected under a stable condition. Additional reference, amplifier, sensor, and interference contributions can reduce usable resolution.

Compare modes at equivalent useful bandwidth where possible. A much slower filter can improve the displayed number while changing measurement latency and the ability to follow the intended signal.

Do not convert an assumed RMS-to-peak ratio into a guaranteed maximum excursion. If using a statistical estimate instead of measured extrema, state the distribution and coverage assumption explicitly.

Save the input arrangement, time record, histogram, RMS noise, peak-to-peak spread, and calculation. Describe the result as observed DC resolution under defined conditions. The nominal converter bit count and an AC ENOB measurement answer different questions and should not be substituted for evidence about stable low-frequency readings.