Reliability enhancement of EM-based tuning of microwave components using regularized operating band scanning

Numerical optimization is ubiquitous in contemporary microwave passive component design. Reliable quantification of electromagnetic (EM) cross-coupling, substrate losses, or environmental effects (e.g., connectors, housing) is impossible using analytical or equivalent network models. Consequently, the designs rendered using such tools must be further tuned at the EM simulation level. Although local algorithms are predominantly employed for this purpose, they are likely to fail if the starting point is of insufficient quality. On the other hand, global search techniques are associated with tremendous computational expenses. This study suggests an innovative technique for immunizing local algorithms against poor starting points, enabling their quasi-global search capability. The presented methodology reformulates the underlying optimization task by evaluating the circuit performance concerning the target operating bands scanned within the predefined ranges and by adding penalty factors proportional to the distance between the original and the scanned targets. The penalized objective is computed as the minimum of the above combination over the entire scanning range. As a result, the design specifications become attainable through local optimization over considerably broader frequency spectra at the expense of only a slight increase in the algorithm running costs. The presented approach is comprehensively verified using several microstrip components and the sensitivity-based routine as the search algorithm. The results unanimously showcase the ability of the procedure to allocate the optimum even under troublesome scenarios (poor initial points), which are unmanageable by conventional algorithms.