Project Technical Concept

In this proposed research, we direct signal processing and network traffic optimization from analytical approaches to data-driven inferring, especially neural network (NN) based approaches. Convolutional and recurrent NN models will be developed in optical fibre and radio communication networks to mitigate stochastic distortions, to optimise network lightpath and to facilitate network load allocation. We will exploit heterogeneous data analysis to build models for optical fibre and radio access channels and to estimate traffic load to optimise both the deployment and operations of optical and radio networks. We will design ML algorithms and ANNs that facilitate future high-capacity optical and radio communications to enable a new age of intelligent communication networks beyond 5G. The proposal is developed to address the following open issues in elastic optical fibre systems and radio access networks:

Approaches

To address the research and innovation (R&I) objectives, DIOR pursues tight academic-industrial cooperation combined with effective integration of the partners’ expertise in radio systems, RANs, optical communication systems, OCNs, and AI-based signal processing and topology optimisation. These skills combined are essential for obtaining fundamental understanding, acquiring new knowledge, and devising optimal solutions for intelligent and robust OCNs and RANs. All R&I tasks demonstrate intersectoral aspects as they are both academically challenging and industrially relevant and will contribute to the strengthening of knowledge and competitiveness of Europe in ICT. The DIOR project is organised into five work packages (WPs) (see Table B2.1). WPs 1-4 focus on R&I objectives. WP5 implements dissemination, exploitation and public engagement activities, and project management. Each WP has a leader: TAU (WP1), UoW (WP2), SK (WP3), IT (WP4) and UoW (WP5). R&I WP leaders are responsible for: (1) coordinating intersite WP to complete the R&I tasks and deliverables on time; (2) exploiting the complementary

Technical Objectives

The corresponding research objectives to address the challenges are as follows

ML-based DPD

Develop new ML-based digital pre-distortion (DPD) methods for linearizing base-station power amplifiers, and thus to enhance the radio signal quality in massive MIMO and radio network power efficiency

Audio/Video Aided Beamforming

Develop fusion model for acoustic/video context information (user behaviour, trajectory, location) and radio beamforming design for high accurate beam alignment & management in 5G NR.

Learn-based physical radio link

Develop the ANN-based baseband radio processing without dependence of explicit channel model, and evaluate the performance of stochastic gradient descent (SDG) method

Load demand awareness modelling

Curate high-resolution physical layer and traffic demand data to achieve high-fidelity channel models.

AI-based optical channel model

Develop power-dependent channel models for optical fibre system to include stochastic impairments.

NNs for compensating nonlinear optical impairments

Develop ANN frameworks to compensate for stochastic nonlinear impairments in optical fibres and semiconductor lasers.

OCN topology optimisation

Design algorithms and signal path topologies to optimally allocate the latency and load of communication networks with intelligent flexibility.

Joint optimisation of OCNs and RANs

Design mechanisms to integrate data and load analysis to optimise the resource allocation of optical and radio access networks.