A Machine-learning and Compressive-sensing Inspired Approach to the Optimal Array Pattern Synthesis

We propose a new approach to the power-pattern optimal synthesis of 1-D array antennas. The technique pursues the minimization of the number of active elements for fixed radiation performances, guaranteeing the achievement of a maximally-sparse array layout and hence the minimization of the system's overall weight and cost. The proposed procedure jointly exploits three amongst the most powerful tools available today for pattern synthesis, namely the Spectral Factorization technique, the Compressive Sensing theory, and the Machine Learning approach. Moreover, while most of the Compressive-Sensing-based techniques available for the synthesis of maximally-sparse arrays pursue a nominal far-field distribution, the proposed one is able to provide a mask-constrained shaping of the power pattern, and hence exploits a considerably larger number of degrees of freedom.