Synthesis of Mask-Constrained Pattern-Reconfigurable Nonuniformly Spaced Linear Arrays Using Artificial Neural Networks

In this article, a novel multibranch encoder-decoder-based artificial neural network (ANN) framework is proposed for pattern-reconfigurable nonuniformly spaced linear array synthesis. In the proposed framework, different encoder-decoder branches are devoted to synthesizing different radiation patterns to satisfy the desired radiation patterns. In each encoder-decoder branch, the encoder, and the decoder serve as the array synthesizer and analyzer, respectively. By minimizing a suitably-defined loss function with respect to multiple radiation patterns, element amplitudes, and positions, not only multiple mask-constrained radiation patterns can be successfully achieved, but also the common element amplitudes and positions with minimum inter-element spacing control can be obtained. Thus, different radiation patterns can be switched by phase-only control. In addition, by using different training samples, the proposed method can consider the ideal synthesis of the array factor, as well as the actual synthesis of antenna arrays to consider the mutual coupling effects in the synthesis process. Simulation results are provided to validate the capability, efficiency, and attractiveness of the proposed method.

Automated summary

In this article, a novel multibranch encoder-decoder-based artificial neural network (ANN) framework is proposed for synthesizing different radiation patterns. By using different encoder-decoder branches to synthesize different radiation patterns, multiple mask-constrained radiation patterns can be achieved, and common element amplitudes and positions with minimum inter-element spacing control can be obtained, allowing for switching between different radiation patterns through phase-only control. Furthermore, the proposed method considers both ideal synthesis and mutual coupling effects.