Study of Training Patterns for Employing Deep Neural Networks in Optical Communication Systems
Chun-Yen Chuang, Lichun Liu, Chia-Chien Wei, Junjie Liu, Lindor Henrickson, Chih-Lin Wang, Young-Kai Chen, Jyehong Chen · 2018
Our investigation demonstrates that training patterns generated by Mersenne Twister pseudo-random number generator have potential to overcome the pattern effect. The resulting trained networks are capable of compensating the actual optical communication system response rather than merely predicting training signals.