A novel generative stochastic model for high-speed interconnection links

In this paper we introduce a new modeling approach to create a generative model for stochastic link responses. The proposed scheme starts from a limited set of simulated or measured ‘training samples’, which are first represented by a rational model using vector fitting with common poles. Next, the generative model is built, leveraging the residues' stochastic distribution, via a principal component analysis and kernel density estimation. Then, in a post-processing phase, non-passive samples are discarded. The novel method is applied to a commercial connector footprint, a multi-conductor transmission line, and a complete link composed of the cascade connection of the former components.