Transfer-Learning-Assisted Multielement Calibration for Active Phased Antenna Arrays

A transfer-learning-based method for accelerating power-only calibration of phased array antennas by combining the conventional array theory with deep learning is presented in this communication. The existing power-only calibration methods either require a significant number of measurement cycles or have restrictive phase shifter resolution requirements. The proposed array calibration method uses a surrogate model to calibrate all the array elements in one pass without restricting phase resolution requirements. We developed a novel feature extraction scheme (FES) that picks out the most important power features resulting in reduced measurement cycles. The burden of data acquisition for model training is further reduced by relational knowledge transfer learning. The surrogate model acquires its general calibration capability from massive theoretical data, which is easily collected by the radiation multiplication theorem, and captures the detailed nonideal response from a small number of simulations. The proposed methodology has been demonstrated and tested on several arrays for validation. The effectiveness and performance of the method have been verified, and hence, it can serve as a complementary tool to accelerate the calibration process of phased antenna arrays.