Reducing Errors in Optical Data Transmission Using Trainable Machine Learning Methods

Abstract

Reducing Bit Error Ratio (BER) and improving performance of modern coherent optical communication system is a significant issue. As the distance travelled by the information signal increases, the bit error ratio will degrade. Machine learning techniques (ML) have been used in applications associated with optical communication systems. The most common machine learning techniques that have been used in applications of optical communication systems are artificial neural networks, Bayesian analysis, and support vector machines (SVMs). This thesis investigates how to improve the bit error ratio in optical data transmission using a trainable machine learning method (ML), that is, a Support Vector Machine (SVM). SVM is a successful machine learning method for pattern recognition, which outperformed the conventional threshold method based on measuring the phase value of each symbol's central sample. In order that the described system can be implemented in hardware, this thesis focuses on applications of SVM with a linear kernel due to the fact that the linear separator is easier to be built in hardware at the desired high speed required of the decoder.

In this thesis, using an SVM to reduce the bit error ratio of signals that travel over various distances has been investigated thoroughly. Especially, particular attention has been paid to using the neighbouring information of each symbol being decoded. To further improve the bit error ratio, the wavelet transforms (WT) technique has been employed to reduce the noise of distorted optical signals; however the method did not bring the sort of improvements that the proponents of wavelets led me to believe.

It has been found that the most significant improvement of bit error ratio over the current threshold method is to use a number of neighbours on either side of the symbol being decoded. This works much better than using more information from the symbol itself.