Study On Evaluation Method For Measurement Uncertain Of ADC

ADC plays an important role in digital instruments; and its performances,especially its dynamical ones, directly affect the properties of the instruments.One of the most important methods to evaluate the performances of an ADC is to analyze its response characteristics via the modern error analysis; while the best modern error analysis is the evaluation of uncertainty.The uncertainty evaluation of dynamic properties on ADC is an obstacle confronted in testing and metrological field.Try to solve the problem above, the paper caries out study as following:(1)Analyze the principle of the measurement uncertainty,especially the effective and application field of different measurement uncertainty evaluation methods available; and find that these methods are not applicable for evaluation the uncertainty of ADC.(2)Construct a dynamic testing platform for high-speed ADC; and then compare different ADC testing methods, including the static testing method based on histogram and dynamic testing method based on the FFT; finally chose FFT as testing method for the high-speed ADC, whose five dynamic performance parameters, including NF, SINAD, ENOB,THD and SFDR, could be acquired.(3)Since the performances of ADC have great influence on properties of an instrument, the noise signal is applied to improve the the ADC conversion accuracy and the anti-jamming; and the best dynamic performance parameters of the ADC are also obtained as the sources of measurement uncertainty evaluation.(4)Since the measurement uncertainty methods proposed by GUM are not fit for evaluation of the dynamical properties of ADC, an artificial neural network based algorithm is presented, which is an multi-input single-output system, and is realized via MATLAB. The simulation results indicate the approach is applicable.( 5) To verify the effectiveness of the artificial neural network basedalgorithm,the evaluation of measurement uncertainty on ADI’s AD6645-105 is carried out, whose results are compared with that of the type A and type B methods proposed by GUM. The conclusion is that the neural network algorithm is more efficient and more accurate than that of GUM methods.