Orbital angular momentum mode demodulation with neural network-assisted coherent nanophotonic circuits

Vortex beam carrying orbital angular momentum (OAM) shows promising potential in enlarging communication capacity and improving modulation ability for its mode orthogonality. Although various OAM multiplexing schemes have been proposed to boost channel capacity density, few studies have focused on OAM shift keying (OAM-SK) communication because of the lack of rapid and low-latency mode demodulation technologies. In this study, we propose a neural-network-assisted coherent nanophotonic circuit (CNC) strategy for demodulating OAM modes and investigate its application in OAM-SK communication. The CNC comprises 380 Mach–Zehnder interferometers (MZIs) and achieves a unitary matrix transformation by simultaneously modulating the amplitude and phase of optical pulses via multi-light interference, where a neural network-assisted design algorithm is used to obtain the phase parameters in the CNC that satisfy the mapping relationship between feature information and OAM modes. Since the satisfied mapping relationship is performed in the optical domain, OAM modes can be demodulated at the speed of light. We show that this CNC can effectively identify the OAM modes ranging from -10 to +10, and the identification accuracy exceeds 90% under moderate turbulence intensity. As a proof of concept, we simulated a hexadecimal OAM-SK communication link, and the OAM signals under turbulent perturbations were successfully demodulated with an accuracy of 99.33%.