Comparative study of DSP techniques for the effective modeling and design of highly complex RF-MEMS structures

Abstract

Four DSP-based digital predictors (Prony's, covariance, forward-backward, matrix pencil), that are commonly used to enhance the time-domain modeling and design of highly complex RF MEMS structures, are evaluated in terms of computational efficiency and accuracy as a function of the model order, the decimating factor, and the size of sample train. For a benchmarking case of an RF MEMS tuner, it is found that while covariance method has the best performance in terms of accuracy matrix pencil method confers robustness to computational economies (less numerical effort) and saves more CPU time with a smaller model order that can be selected by easy and efficient criteria.