Abstract:
This paper discusses the effect of residual components on type A uncertainties determined by multiple linear regression of quadrature homodyne output signals. Primary accelerometer calibrations employing the sine-approximation method allow the determination of the fit uncertainty through the covariance matrix. Frequency components that are not included in the regression model lead to increased uncertainties. The uncertainty obtained from the regression is therefore not exclusively due to random data variance at the calibration frequency, but due to deterministic error of the model. Analysis of the regression residues and a correction method is proposed to obtain more realistic uncertainty values. Some results obtained from simulated data sets are presented.
