Improved Residual Analysis in ADC Testing

Analog-to-digital converters are commonly tested by applying a pure sine wave to their inputs. Since the exact parameters of the sine wave are very difficult to precisely obtain, an indirect method is used: the measured samples are used to determine the sine wave which best fits them. The ADC is characterized then by analysis of the differences between the samples and the sine wave. This is described in the IEEE standards 1241-2000 and 1057-1994. The generally applied method for the fit is least squares (LS). If the error (the deviation of the ADC output samples from the true sine wave) is a random sequence, with independent, identically distributed zero-mean samples, the LS fit effectively averages it out. However, when the quantization errors dominate in the observations, this is not true. Strange, systematic errors may arise, which cannot be averaged out. The paper examines these errors, and makes suggestions how to reduce them.