Phase calibration for optical phased arrays (OPAs) is a key process to compensate for the phase deviation and retrieve the initial working state. Conventional calibration approaches based on iterative optimization algorithms are tedious and time-consuming. The essential difficulty of such a problem is to inversely solve for the phase error distribution among OPA elements from the far-field pattern of an OPA. Deep-learning-based technology might offer an alternative approach without explicitly knowing the inverse solution. However, we find that the phase ambiguities, including conjugate ambiguity and periodic ambiguity, severely deter the accuracy and efficacy of deep-learning-based calibration. Device-physics-based analysis reveals the causes of the phase ambiguities, which can be resolved by creating a tailored artificial neural network with phase-masked far-field patterns in a conjugate pair and constructing a periodic continuity-preserving loss function. Through the ambiguity-resolved neural network, we can extract phase error distribution in an OPA and calibrate the device in a rapid, noniterative manner from the measured far-field patterns. The proposed approach is experimentally verified. Pure main-beam profiles with > 12    dB sidelobe suppression ratios are observed. This approach can help overcome a crucial bottleneck for the further advance of OPAs in a variety of applications such as lidar.

No abstract

The crucial factors governing the behavior of waveguide superlattice-based optical phased arrays are investigated. The superlattice-induced extra lobes can be minimized with sophisticated supercells while the inter-waveguide crosstalk still needs to be judiciously balanced.

Using neural networks, we experimentally extract phase errors of an optical phased array from its far-field pattern non-iteratively for calibration. Periodic and conjugate phase ambiguities are resolved, helping achieve high sidelobe suppression ratio.