Self-coherent multicore fiber based systems assisted by neural networks (SELFIE)
Fundação para a Ciência e Tecnologia
Funding reference: 2023.16564.ICDT
Institution:
Abstract
SELFIE project proposes and demonstrates the integration of multicore fibre (MCF) technology, low-complexity direct-detection (DD) self-coherent receivers, and machine learning algorithms based on neural networks as a powerful solution to significantly increase the system capacity in short-medium reach optical networks where cost is of primary concern.
This is assured by a multidisciplinary team with expertise worldwide recognized in the research areas addressed in the project.
This project is coordinated by IT with international collaborations from:
Objectives
The main objective of SELFIE is to design an integrated self-coherent MCF system that represents a novel solution that can be employed in next generation intra/inter data centres or access networks, and cope with the huge capacity demand experienced by these networks. To achieve this goal, the SELFIE project focuses on:
I
The development of direct-detection (DD) receivers based on self-coherent techniques as the Kramers Kronig (KK) algorithm, to improve the detection linearity, and the Stokes vector (SV) receiver, to exploit other degrees of freedom of the optical signal, to cope with high data rate signals transmission (=200 Gb/s) while avoiding/mitigating transmission impairments as chromatic dispersion or signal-signal beat interference.
II
The system performance optimization by testing DD-compatible transmission techniques, as virtual carrier-assisted DD-OFDM, or more complex quadrature amplitude modulation and single sideband modulation, and to evaluate their robustness to the random variation of the ICXT along time.
III
To design a new MCF with higher core count than current standards and moderate to high crosstalk levels, -30 dB/km @ 1550 nm, still acceptable for short-reach applications.
IV
The proposal of machine learning to improve the robustness of the network performance to the random variation of the ICXT along time. In this context, three scenarios with different requirements on the ML algorithms are addressed:
a. Point-to-point DD MCF systems
b. DD MCF networks with |skew|×bit rate<<1 or |skew|×bit rate>>1. This is the more complex situation as the data signals of the interfering core are not available at the receiver side because signals can be added or dropped from the network at any node. Thus, unsupervised learning is required to perform a blind tracking of the random variation of the ICXT along time.
Key Innovation Results (KIRs)
KIR 1
Design of advanced self-coherent transceivers to be used in MCF short-reach systems;
KIR 2
Design of a new MCF with higher core count;
KIR 3
ML techniques for end-to-end performance optimization of MCF networks;
KIR 4
Software to simulate ICXT-impaired self-coherent MCF short-reach systems;
KIR 5
Prototype of the self-coherent short-reach MCF system employing machine learning for end-to-end optimization.
Membros equipa Investigação
Tasks
TASK 1
Design of advanced
Start: 03/11/2025
End: 31/10/2028
TASK 2
Specification and application scenarios definition (IT-IUL: JR)
Start: 03/11/2025
End: 31/01/2026
TASK 3
Development of a software simulator (IT-IUL: AC)
Start: 01/01/2026
End: 30/04/2028
TASK 4
Design of self-coherent transceivers and high core count MCF (IT-Av: AIA)
Start: 01/01/2026
End: 28/02/2028
TASK 5
Design of neural networks for end-to-end optimization (IT-IUL: TA)
Start: 01/05/2026
End: 28/02/2028
TASK 6
Off-line integrated network prototype (IT-IUL: TA)
Start: 01/07/2027
End: 30/06/2028
TASK 7
Real-time integrated network prototype (IT-Av: PM)
Start: 01/03/2028
End: 30/10/2028
TASK 8
Open house demonstrator (IT-IUL: TA)
Start: 01/09/2028
End: 31/10/2028
TASK 9
Dissemination and exploitation of SELFIE results (IT-IUL: TA)
Start: 03/11/2025
End: 31/10/2028