MetaAntennaNet: Meta-Learned Radiation Pattern Optimization for Reconfigurable Smart Antennas

MetaAntennaNet introduces a meta-learning-driven framework for radiation pattern optimization in smart reconfigurable antennas that can quickly adapt to changing wireless communication environments with a minimum of retraining. The proposed approach utilizes a model-agnostic meta-learning strategy to learn a generalized initialization that can be very quickly fine-tuned for different antenna configurations and propagation scenarios. By directly mapping environmental and system constraints to optimal antenna control parameters, MetaAntennaNet drastically reduces the computational load of conventional iterative optimization methods. Numerical results show that this method leads to a main-lobe directivity improvement of 27.4% and a side-lobe level reduction of 31.8% as compared to the gradient-based baseline techniques and achieves convergence within five adaptation steps. These findings confirm the potential of meta-learning to enable on-the-fly radiation pattern reconfiguration in future smart antenna technology.