Impact of Partition Schemes in Loewner Matrix Macromodeling

The first step in Loewner matrix macromodeling is to partition the data into two parts. Previous research has shown that different partition schemes, although theoretically equivalent, may significantly affect matrix condition numbers. The impact of such ill-conditioning on the frequency and time domain responses of the macromodel, however, has been less addressed in the literature. In this paper, we clarify the importance of good conditioning for an accurate modeling in frequency and time domains, and compare different partition schemes using simulated and measured S parameters. The results demonstrate that with the bad schemes, the practically allowable bandwidth of the model will be greatly limited.