Hybrid Newton/Gauss-Newton algorithm for time-domain analysis ofA/D converters

摘要

This brief describes a maximum-likelihood estimator for a sum ofnharmonically related sinewaves for characterizing analog-to-digitaln(A/D) converters. The frequency is estimated by means of the Newtonnmethod. For the iterations where singularity problems occur, thenalgorithm switches to the Gauss-Newton method combined with the singularnvalue decomposition. The initial frequency accuracy required by thisnalgorithm is determined and the algorithm is modified in order tonconverge to the global minimum, whatever the initial frequencynuncertainty. Thus, one can use the frequencies of the signal generatorsnto provide the starting value. Once the frequency is determined, thenlinear parameters of the model are estimated by solving a linear set ofnequations. It is shown through simulation and experimental results thatnthe proposed algorithm is fast, accurate, and robust, with respect tonthe initial guess of the frequency