Acyclic Connected Graph Optimization for Generative Measurement Based Models

Measured S-parameters are extremely useful in characterizing a system, but are cumbersome to use in many contexts. When used in a simulation as a blackbox macromodel, artifacts from launch structures can obscure the details of the measured system and cannot be separated. This paper presents a genetic algorithm to automatically create surrogate models of measured S-parameters using an acyclic graph of transmission lines. The algorithm is further accelerated by smart evolution based on local fitness, and seeding of the initial generation based on simple deterministic heuristics. Surrogate models generated in this manner are easily used in simulators, enable what-if analysis, and identify impedance changes and stubs which help identify their sources and can be easily mapped to the underlying physical topology.