Amalgamation of Machine Learning Techniques with Optical Systems: A Futuristic Approach

With the availability of huge bandwidth backbone networks in terms of optical fiber links, there is a surge in data at the user end. It has been boosted further by the deployment of 5G services and IoT. Hence to incorporate this big data, demand for flexible and scalable networks has risen thereby increasing complexity. So, there is a need of new techniques that can monitor their performances and accordingly adjust the parameters as per the real time requirements and reduce the cost/bit/sec. In this direction, we have reviewed various machine learning algorithms that have the potential to be incorporated with optical systems that can further optimize the efficiency across many dimensions. Hence, the complex network can be self-reconfigured by managing failure and estimating Quality of Transmission (QoT) using machine learning algorithms. Further, input to machine learning algorithms is the dataset. In the present scenario, dataset to this scientific domain of optical communication and networks is not abundantly available and hence can either be generated through simulations, experimentation or can be synthetically provided by advanced machine learning algorithms. The dataset can, thus, be taken from field trials and testbeds, lab trials and testbeds, open-source platforms and from some government funded networks etc. As per the type of data (image data, sequential data or augmented data), machine learning algorithms are classified. In this paper, we have generated data for optical mesh network connecting four Indian cities (Delhi, Mumbai, Bangalore and Kolkata) using different transceivers on a public and open-source platform – GNpy model – a QoT estimation tool and optimized the data rates and modulation format for distance and generalized signal to noise ratio (GSNR) values. Data similar to this can further be generated for different networks and can be applied as input to machine learning algorithms in some standard format and ML algorithms can be developed in optical domain.