On the use of the unscented transform for uncertainty estimation in DSP based measurements

Estimation of the output expectation and standard uncertainty in indirect measurements based on digital signal-processing (DSP) algorithms is dealt with. DSP measurement algorithms can, in fact, be considered as complex models, which operate on sampled data acting as direct measurement results. The related uncertainty sources can mainly be associated with the analog conditioning and analog-to-digital conversion the input signal undergoes. Due to the inherent non-analytical formulation of DSP measurement algorithms, the use of IEC-ISO recommendation for standard uncertainty estimation is usually unfeasible. The paper aims at showing the efficacy and reliability of an original approach, mandated to uncertainty estimation and based on the unscented transform, when applied to DSP measurement algorithms. A number of numerical tests are conducted on simulated measurement data. Remarkable concurrence between obtained estimates and those granted by Monte Carlo simulations confirms the efficacy of the proposed approach.