Modulation-Format-Transparent Polarization Tracking Using a Neural Network

We propose a novel technique for modulation-format-transparent polarization tracking and demultiplexing in a coherent, polarization multiplexed fiber-optic communication system. The technique employs Stokes space analysis of the received signal, which allows us to uniquely identify the polarization state independently of the modulation format. A neural network is used for acquisition and tracking of the polarization state and the obtained information is subsequently used to update an adaptive filter that performs the actual demultiplexing.