FONTE has produced a number of research results at the edge of modern communication technology. FONTE engineered cost-effective deep learning solutions for improving optical communication links. Among the most important results we would like to mention:
• Achieved a world record transmission rate of 1.48 Tbps over a typical inter-data-centre link through developing and implementing a deep-neural-network-based de-noiser.
• Improved transmission quality by 5× in a real-world operational 600 km link by developing and implementing a tailored complex-value-based neural network.
• Increased the capacity by 4% at no additional cost in a 2400 km link by learning the optimal link configuration via an autoencoder-based machine learning algorithm.
• Trimmed by 2× training complexity of time-series filtering algorithms and by 6× the required dataset size via developing a data augmentation procedure.
• We developed a novel approach to incorporating loss into the nonlinear Fourier transform based systems exploiting the dispersion decreasing fibres. The proposed solution substantially suppresses the degrading effects of fibre loss.
Together with industrial partner Nokia Bell Labs we proposed and developed a new receiver based on optoelectronic machine learning for intensity-modulated and direct detection systems. The optical pre-processing stage slices the received signal spectrum in small sub-bands with passive optical filters and each is detected by a photodetector with a digital post-processing based on reservoir computing. The method led to a patent filed by the Nokia Bell Labs.
• We proposed and numerically demonstrated the potential of the receiver for 32-GBd OOK signal transmission, showing an increase in reach from 10 km to 40 km, corresponding to 400%, compared with digital-only techniques.
• We validated these findings in an experimental setting at the industry partner facilities. The proposed receiver together with ML-based equalizers enabled to increase the transmission reach from 10 km to more than 80 km, i.e. a more than 800% increase in reach.
To alleviate nonlinear channel impairments in long-haul fibre-optic communications, we developed a neural network equalizer consisting of a convolutional neural network encoder and a unidirectional many-to-one vanilla recurrent neural network operating in tandem.
• We demonstrated that for 64 GBd DP-16-QAM optical transmission over 14× 80 km SSMF, the suggested CRNN-based equalizer outperforms state-of-the-art bidirectional recurrent-based approaches while having >50% lower computational complexity compared to them.
• We also demonstrated that, when phase noise is included, a mere neural network-based nonlinearity mitigation at the end of the linear equalization chain outperforms the joint nonlinearity and polarisation mode dispersion MD mitigation solution after chromatic dispersion compensation in terms of performance and complexity.
We developed a neural network (NN)-based digital pre-distortion (DPD) technique for a high baud rate coherent transmitter that compensates hardware distortions and showed significant improvements compared to other popular nonlinear DPD techniques.
By using this NN-based DPD
• We demonstrated record net data rate of 1.61 Tb/s on single wavelength over 80 km of fiber.
• We achieved a net C-band data rate of 54.5 Tb/s over 48 km of field-deployed fiber.
The 4 PhD students in FONTE submitted 1 patent with Nokia, published 1 book chapter, 6 journal papers, 16 conference papers and several software packages & codes. The consortium organised 22 scientific training events attended by over 1150 external researchers, 4 Transferable Skills Workshops, 5 Project Annual Meetings and completed 25 public engagement activities, reaching all sectors of society.
