Simulated Scattering Parameter Dataset for a Randomly Generated Non-Parameterized Microstrip Filter Between 1 and 10 GHz

High-frequency electromagnetic component design is often a complex and computationally expensive task. Analytical techniques are uncommon and frequently yield only approximate solutions. Design processes often necessitate expensive full-wave electromagnetic solvers and iterative optimization to satisfy all requirements. Although machine learning has expanded the capabilities of electromagnetic design, its adoption remains limited by the scarcity of large, well-structured, and publicly accessible datasets required for training and validation. This dataset was compiled to provide an accessible entry point into machine learning and data analysis without needing access to a full-wave electromagnetic solver or extensive electromagnetic knowledge. It was created by simulating 21,000 variations of a microstrip filter using CST Studio Suite between 1 and 10 GHz to obtain the two-port S-Parameters. Each patch is represented by a 32×32 binary image indicating the presence or absence of conductive material and is derived from one of eight geometric parameter sets. The dataset is accompanied by visualizations illustrating its scope and the influence of geometric variation on electromagnetic response.