Improving robustness of vector fitting to outliers in data

Introduction: Robust broadband macromodelling techniques are of crucial importance for efficient time domain and frequency domain simulation of high-speed interconnect structures. The standard vector fitting (VF) algorithm starts from an S-parameter frequency response, sampled at a discrete set of frequencies. It computes a macromodel by minimising a weighted iterative cost function in the L2 sense [1]. In real-life situations, the quality of the L2 fitting model may be degraded owing to outliers in the data. An outlier is a value in the data that deviates strongly from the other values, and is usually caused by measurement or instrumentation errors. In this Letter we propose a modified vector fitting algorithm that minimises the L1 norm of the complex fitting error instead [2]. Numerical results illustrate that the new approach is more robust with respect to outliers [3].