Abstract
A new approach to the real-time synthesis of antenna array factor and weight calculation is introduced, utilizing a conditional Generative Adversarial Network (cGAN). To our knowledge, we are the first to employ a GAN for array antenna pattern synthesis. The neural network takes a predefined simple array pattern as input, which includes only the main lobes and null angles, and certain constraints like SLL control. The output of the network is the actual antenna array factor. We demonstrate that by performing a fast Fourier transform on the network's output, the corresponding weights can be obtained instantly since the patterns on which the network is trained are derived from an iterative FFT algorithm explained extensively in the main body of the text. The proposed method eliminates the need for computationally expensive and time-consuming iterative methods. Furthermore, unlike other AI-based array antenna pattern synthesis methods, our approach requires minimal data, even for large array antennas. It will be demonstrated that for a planar array, with only 1000 array factor featuring a single main lobe and null at random angles, the trained neural network is capable of generating patterns with any desired number of main lobes or nulls, at any arbitrary angle.