Je-Heon Lee

This paper proposes an artificial intelligence algorithm based on a mixture density network (MDN) to solve the inverse problem of estimating the excitation phases of antenna elements for array antenna radiation pattern synthesis. In array antenna radiation pattern synthesis, a nonuniqueness issue exists in which multiple phase solutions can produce an identical radiation pattern, which limits the effectiveness of conventional deep learning models in learning multiple solutions. The proposed MDN estimates the probability distribution of antenna excitation signals from a target radiation pattern, thereby effectively learning multiple feasible solutions. In addition, an iterative retraining framework based on the error between the target radiation pattern and the predicted radiation pattern is introduced to improve prediction accuracy. Simulation results for an eight‐element array antenna demonstrate that the proposed MDN achieves superior phase estimation performance for radiation patterns not included in the training dataset and arbitrarily generated radiation patterns.