A Dataset-Driven RBFNN Framework for Real-Time Pattern Recovery in Deformed Wideband Arrays for Airborne and Spaceborne IoT Applications

In airborne and spaceborne Internet of Things (IoT) applications, deformation of antenna sensor arrays is a prevalent challenge in platform-integrated antenna systems. This paper proposes a data-driven radial basis function neural network (RBFNN) framework for real-time pattern recovery while simultaneously achieving low sidelobe level (SLL), in wideband tightly coupled antenna arrays (TCDAs) under bending and torsional deformations. The mathematical representation of radiation pattern of deformed arrays is established based on the element pattern under periodic boundaries. The pattern synthesis of maximizing directivity satisfying user-defined peak SLL constraint under specified deformations is transformed into a convex optimization problem and efficiently solved. The resulting solutions of amplitude and phase excitations are incorporated into the sample dataset for subsequent neural network training. Subsequently, two dedicated, data-driven RBFNNs are established and trained to predict the amplitude and phase excitations of array elements based on the input deformation states and scanning angles. A 1×16 TCDA operating in the 8.5GHz-12GHz band over ±45° scan range is employed to verify the effectiveness of the proposed method in beam pointing, SLL suppression, and gain loss when the array is deformed. The numerical calculations, full-wave simulations and experimental measurements collectively demonstrate that the proposed method can achieve millisecond-level prediction time, accurate beam pointing, SLL suppression and gain improvement relative to the uncompensated deformed arrays.