Research Round-up: October 2020

Here’s what TRANSNET researchers published during October, including papers on optical amplifiers, constellation shaping and digital signal processing.

Journal Paper
Multi-band programmable gain Raman amplifier (doi.org/10.1109/JLT.2020.3033768)
Journal of Lightwave Technology

In this paper, a multi–band programmable gain amplifier using only Raman effects and machine learning is experimentally demonstrated for the first time. The amplifier offers a novel approach to improve the capacity of future optical communication systems. This work is a collaboration across the Technical University of Denmark, Aston University and Politecnico di Torino.

Journal Paper
Coupled Transceivers-Fiber Nonlinearity Compensation Based on Machine Learning for Probabilistic Shaping System (doi.org/10.1109/JLT.2020.3029336)
Journal of Lightwave Technology

In this paper, the combined benefit of transceiver-fibre artificial neural network-based nonlinearity compensation for probabilistic constellation shaping is experimentally demonstrated for the first time. It builds on previous work which only considered transceiver nonlinearity whilst studying the effectiveness of machine learning in probabilistic shaping systems. The research is a collaboration between Aston University and Dublin City University.

Journal Paper
Space of Functions Computed by Deep-Layered Machines (doi.org/10.1103/PhysRevLett.125.168301)
Physical Review Letters

This paper presents an analytical framework to examine the distribution of Boolean functions represented by random deep-layered machines, including deep neural networks and Boolean circuits. The methods developed provide a deeper understanding of the theory, performance and application of machine learning architectures. The work is a collaboration between the London Institute for Mathematical Sciences and Aston University.

Journal Paper 
Geometric Shaping of 2-Dimensional Constellations in the Presence of Laser Phase Noise (doi.org/10.1109/JLT.2020.3031017)
Journal of Lightwave Technology

This paper provides a fresh perspective on the design of geometric shaping modulation formats for optical fibre communication systems to reduce the implementation complexity. The proposed strategy demonstrates an enhanced signal-to-noie ratio tolerance in high phase noise regimes when compared to equivalent methods – the next step is experimental verification. 

Conference Paper
Time-Domain Learned Digital Back-Propagation (doi.org/10.1109/SiPS50750.2020.9195253)
2020 IEEE International Workshop on Signal Processing Systems

In this paper, learned time-domain digital back-propagation is experimentally demonstrated for the first time. The research concentrates on wideband multi-channel long-haul optical transmission systems and shows the potential of digital signal processing techniques with learned parameters for improving the performance of optical fibre transmission systems. Colleagues from the Huawei Chengdu Research Institute also contributed to this work.

Conference Paper 
Experimental Investigation of Deep Learning for Digital Signal Processing in Short Reach Optical Fiber Communications doi.org/10.1109/SiPS50750.2020.9195215)

This paper investigates the experimental performance enhancement of auto-encoders based on a recurrent neural network for communication over dispersive nonlinear channels. The research highlights deep learning as a viable digital signal processing solution for low-cost optical communications to increase reach or enhance the data rate at shorter distances. This work also involved colleagues from the Karlsruhe Institute of Technology and Nokia Bell Labs.


 For a full list of published TRANSNET research, including journal and conference papers, please take a look at our publications page.