Presentation description

Antenna arrays are foundational to modern communication systems, including 5G networks, satellite links, and high-speed Wi-Fi. These arrays generate directional radiation patterns based on the excitation, typically the amplitude and/or phase, applied to each antenna element. To maximize signal quality and minimize interference, engineers design excitation configurations that produce radiation patterns with specific characteristics, such as narrow mainlobes, low sidelobe levels (SLL), and beam steering toward a target direction.
While calculating a radiation pattern from a known excitation is straightforward using EM theory, solving the inverse problem of producing a realistic excitation from a desired pattern, is far more complex. This challenge is amplified by mutual coupling, a physical phenomenon where energy radiated by one element influences nearby elements. These interactions distort the expected radiation output and make analytical solutions computationally expensive. Additionally, the inverse problem introduces excitations which are non-unique: many different excitation vectors yielding similar patterns, meaning no single correct solution may exist, though some of those solutions could be physically impossible.
Our research addresses these problems using a neural network model trained to predict element excitations from a desired radiation pattern for an 8×8 planar antenna array. We generate a large dataset of physically accurate pattern-excitation pairs using High Frequency Structure Simulator (HFSS), a full-wave EM simulation tool that inherently accounts for coupling effects. Using this method, the model learns the physical behavior of the array, generalizing toward solutions that represent physically realizable excitations. This allows the network to consistently produce valid excitation vectors, even when multiple possibilities exist. Our approach eliminates the need for costly iterative methods at runtime, offering a fast, data-driven solution for real-time array design in beamforming, adaptive networks, and next-generation wireless systems.