Machine learning opportunities for integrated polarization sensing and communication in optical fibers

As the bedrock of the Internet, optical fibers are ubiquitously deployed and historically dedicated to ensuring robust data transmission. Leveraging their extensive installation, recent endeavors have focused on utilizing these telecommunication fibers also for environmental sensing, exploiting their inherent sensitivity to various environmental disturbances. In this paper, we consider integrated sensing and communication (ISAC) systems that combine data transmission and sensing functionalities, by monitoring the state of polarization to detect environmental changes. In particular, we investigate various machine learning techniques to enhance the performance and capabilities of such polarization-based ISAC systems. Gradient-based techniques such as adaptive zero-forcing equalization are examined for their potential to enhance sensing accuracy at the expense of communication performance, with strategies discussed for mitigating this trade-off. Additionally, the paper reviews novel machine-learning-based approaches for blind channel estimation using variational autoencoders, aimed at improving channel estimates compared to traditional adaptive equalization methods. We also discuss the problem of distributed polarization sensing and review a recent physics-based learning approach for Jones matrix factorization, potentially enabling spatial resolution of sensed events. Lastly, we discuss the potential of leveraging dual-functional autoencoders to optimize ISAC transmitters and the corresponding transmit waveforms. Our paper underscores the potential of telecom fibers for joint data transmission and environmental sensing, facilitated by advancements in digital signal processing and machine learning.