Siva Sudani, Li Xu and Degang Chen
This Website provides references and reference code for 4 state-of the-art ADC testing methods for comparative study.
Full paper download:
Simulation data download:
The quantitative details of requirements and challenges to satisfy the requirements for ideal DFT based spectral test are as follows.
Input Signal: In order to test an N-bit ADC, the input signal should be at least N+3 bits pure.
Clock Signal: The contribution from jitter between input and clock signals should be less than the noise floor.
Sampling: The input signal should be coherently sampled.
Input Amplitude: The peak-to-peak voltage of the input signal should be within the ADC input range.
Data record: The number of samples in data record should be selected such that at least five periods of input signal are sampled
Reference:
Matlab code:
Using windows is one of the most widely recommended methods in both industry and academia to perform spectral test when the data is not coherently sampled.
Reference:
Matlab code:
Four Parameter Sine Fit (FPSF) method is another approach that is widely used for spectral test. It can be used when the data is non-coherently sampled or clipped,
thus, relaxing two conditions for spectral test (coherent sampling and in-range amplitude).
Reference:
Matlab code:
The FIRE method was recently proposed to provide accurate and robust spectral results,thus eliminating the need for coherent sampling.
Reference:
Matlab code:
FERARI method performs accurate and robust spectral test with simultaneous non-coherent sampling and amplitude clipping
thus relaxing the constraints on Sampling and Amplitude.
Reference:
Matlab code:
The foloowing Table provides the summary about situations when a particular method can be used. The default condition is that the data is non-coherently
sampled. The first column indicates the test setup requirement or capability. The box (T) indicates the method can be used and the red (F) indicates the
method cannot be used.
| Test Setup | Window | FPSF | FIRE | FERARI |
| Need fast test | T | F | T | T' |
| Can obtain small δ | T | T | T | T |
| Cannot control δ | F | T | T | T |
| Knows DUT Resolution | T | T | T | T |
| Does not know DUT’s Resolution | F | T | T | T |
| Need Full Spectrum Test | T* | F | T | T |
| Only Need ENOB and SNDR | T | T | T | T |
| Need Multi-tone Test | T*' | F | T^' | F |
| With Clipped Data | F | T | F | T |
*: Provided the bin containing the frequency of interest is not in the primary lobe.
^: Provided the tones are well separated and less number of tones are present in the signal.
': in Multi-tone test row indicates the methods cannot work robustly, while in FERARI column indicates the method can be used, but FIRE method is preferable as it is
way faster than FERARI.
This example shows how to select different spectral testing method to obtain spectral performance for given ADC output data.
Example