This paper aims to address the signal coupling problem of a shared-aperture dual-band dual-circularly polarized microstrip antenna. A decoupling optimization method that combined a convolutional neural network (CNN) with binary particle swarm optimization (BPSO) was proposed. The method introduced pixelated decoupling branches near the antenna feeds, constructed a surrogate model to capture the nonlinear mapping between the branch topology and the electromagnetic performance using a CNN, and adopted BPSO to perform global optimization on the binary pixel matrix, thereby alleviating the time-consuming optimization in a complex, high-dimensional parameter space. Simulation results showed that the optimized S₂₁ was reduced by an average of 20 dB over 1.607–1.620 GHz and by an average of 25 dB over 2.487–2.502 GHz, effectively improving the port isolation. These findings demonstrate that the proposed intelligent optimization strategy is effective and practically applicable for solving antenna decoupling problems.

Surface deformation of a phased array antenna (PAA) induced by external loads can degrade its radiation performance. To restore the beam of a deformed PAA, this paper proposes a new strain–excitation mapping model (SEMM) capable of rapidly calculating excitation adjustments based on measured structural strains. In the derivation of the SEMM, an analytical formula establishing the relationship between antenna excitations and the element positions and orientations for a PAA with an arbitrary surface shape is derived using the projection principle. Subsequently, the positions and orientations of the elements are expressed as functions of a limited number of strain measurements from the deformed antenna structure. An X-band PAA experimental system, equipped with a deformable mechanism and strain measurement capabilities, was developed. Two typical deformations were taken as examples to validate the proposed SEMM. Experimental results demonstrate that the SEMM can effectively recover the distorted pattern across the observation region. Compared with existing models, the proposed model achieves better sidelobe recovery. The rapid computation capability and analytical formulation of the SEMM make it highly suitable for developing an adaptive PAA that can autonomously preserve radiation beam quality under in-service deformations.