On-Board Array Self-Calibration Using Amplitude-Only Proximal-Field Sensors and Machine-Learning-Based Phase Retrieval

A technique for calibrating phased arrays using on-board proximal-field amplitude sensors is introduced. Using these sensors, the phase excitation of each element in an array is estimated using local measurements. This allows the array to self-calibrate and correct itself in the field with zero downtime. The sensors are implemented as low-complexity peak detectors, coupling a small amount of energy from the radiating elements and minimally affecting their performance. An information-theory-based sensor placement methodology is introduced. The resulting non-linear phase retrieval problem is solved using a convolutional neural network to predict the element phases. A proof-of-concept 2x4 array operating at 2.5 GHz with proximal-field sensors is built and tested. The system achieves an average RMS phase error of 5.3 degrees.