Checking FFT Coherence Before Diagnosing ADC Distortion

An ADC spectrum with energy spread around a test tone does not automatically show a defective converter or noisy PCB layout. The captured record may end at a different phase from where it began, causing spectral leakage in the FFT. Check the acquisition and analysis method before changing the analog circuit.

Define the captured record

Record the actual sampling frequency, sample count, input frequency, input amplitude, and clock relationships. Analog Devices’ coherent sampling explanation describes arranging an integer number of input cycles within the acquisition record. This is a relationship between the signal and sampling process, not merely a choice of FFT display settings.

For a record of N samples acquired at sample rate fs, the bin spacing is fs divided by N. As an illustrative example, 4096 samples at 1 MHz produce a bin spacing of approximately 244.14 Hz. A tone chosen for 41 cycles in that record is approximately 10.0098 kHz. The actual source and clock accuracy still determine how closely the acquisition meets that relationship.

Use windowing deliberately

If coherent acquisition is unavailable, select and document a suitable window. Windowing changes the spectral shape and requires appropriate amplitude and noise accounting. Do not compare results from different windows as though each FFT bin represented an identical measurement.

Define which bins are assigned to the fundamental, harmonics, noise, and excluded regions. An analysis script that counts leaked fundamental energy as noise can report poor performance even when the hardware is unchanged. Conversely, excluding too many bins can conceal relevant behavior.

Verify the setup with controlled changes

1. Confirm the stored samples and their numerical interpretation before calculating spectra.
2. Repeat with a deliberately adjusted input frequency to observe how leakage changes.
3. Check that the signal generator and filtering are cleaner than the quantity being evaluated.
4. Review clipping, common-mode conditions, and input settling independently.
5. Preserve the analysis settings and script version with the raw record.

A spur that remains at a consistent physical frequency deserves a different investigation from energy that follows the FFT boundary mismatch. Neither pattern alone identifies the complete cause, but controlled changes help choose the next measurement.

The final result should state whether coherent sampling or a window was used and how the reported dynamic metric was calculated. Keep acquisition artifacts separate from demonstrated circuit limitations. This prevents an unnecessary layout revision and makes later comparisons between converter settings, board revisions, and test laboratories substantially more meaningful.