Abstract:
In this paper, machine learning-based techniques
are used to solve and analyze the modulation format recognition
problem. The combination of intelligent software and highperformance
hardware provides a large scope for innovation in
optical networking. Machine learning algorithms can use a large
amount of data available from the network monitors to learn and
make the network more robust. This is a problem in optical
communication that consists of defining the type of digital
modulation process in which an electrical signal should be sent.
A dataset to represent realistic transmission behaviors was
generated using a simulator based on a Gaussian noise model. A
multi-layer perceptron was used and tested with different
architectures to show that a high level of accuracy is achievable
with machine learning. An analysis of the input features was
made by using the select K best features method. Finally, an
attempt to visualize the data in 2-dimension was made using the
Principal Component Analysis (PCA) and t-distributed
Stochastic Neighbor Embedding (t-SNE) methods to reduce the
dimensionality of the input features and see their relationships.
