Robust Passive Macro-Model Generation With Local Compensation

This paper presents a new passivity enforcement technique for linear time-invariant multiport systems generated from tabulated measured or simulated data. Traditional methods based on iterative eigenvalue/singular value perturbation do not guarantee convergence, and the error introduced is sometimes large due to the lack of explicit error control. The key to the new algorithm is to correct passivity violations locally with moderate increase of system size. Since all violations are fixed locally, the impact on system transfer function outside the passivity violating frequency range is minimized. Thus, the convergence issue is avoided and the accuracy degradation due to passivity enforcement can also be minimized. The proposed method is very efficient, as optimization procedures are not required. Experimental results demonstrate its performance.