DIOR: Deep Intelligent Optical and Radio Communication Networks

Projektinformationen

DIOR

ID Finanzhilfevereinbarung: 101008280

DOI

10.3030/101008280

EK-Unterschriftsdatum 30 November 2020

Startdatum 1 Dezember 2021

Enddatum 31 Mai 2027

Finanziert unter

• EXCELLENT SCIENCE - Marie Skłodowska-Curie Actions

Gesamtkosten

€ 1 803 200,00

EU-Beitrag

€ 1 633 000,00

Koordiniert durch

TAMPEREEN KORKEAKOULUSAATIO SR
Finland

Periodic Reporting for period 1 - DIOR (DIOR: Deep Intelligent Optical and Radio Communication Networks)

Berichtszeitraum: 2021-12-01 bis 2024-11-30

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

The DIOR project aims to tackle the complexity and performance limitations in optical and radio parts in telecommunications networks with the aid of machine learning. The key challenges include: 1). increasing demand for the information capacity of the existing telecommunications networks; 2). inefficient radio and optic components in the telecommunications infrastructures (e.g. fiber nonlinearities, such as self-phase modulation, cross-phase modulation, and four-wave mixing, equalization-enhanced phase noise, radio power amplifier nonlinearity of the radio transceivers, mobility-induced beam misalignment, performance degradation in long-haul optical connections); 3). inefficient network resource allocation; 4). lack of joint optimization of radio and optic resources across the access network, transportation network in the traditional signal processing and network management approaches.

The DIOR project addresses critical challenges in next-generation telecommunication infrastructure that impact various aspects of modern society, including supporting high-speed and reliable Internet; enabling AI-driven 5G and Beyond Technologies; boosting digital Inclusion and connectivity in both urban and rural areas; improving energy efficiency and sustainability by optimizing the radio and optic components and operational costs and advancing AI applications in telecommunications.

The DIOR project sets out to develop Machine Learning (ML) & AI-driven solutions for Communication Networks, focusing on the following objectives:
1. ML-based Digital Pre-Distortion (DPD): Develop ML-based methods to linearize base-station power amplifiers, improving radio signal quality and power efficiency.
2. Develop advanced AI-based methods to model optical fiber channels with power-dependent impairments, incorporating stochastic effects such as polarization mode dispersion (PMD) and amplifier spontaneous emission (ASE) noise, to enhance the accuracy of signal quality predictions.
3. Learn-Based Physical Radio Links: Replace traditional analytical models with deep learning approaches for radio signal processing.
4. Load Demand Awareness Modelling: Utilize high-resolution data to enhance network traffic prediction and optimization.
5. AI-Based Optical Channel Modelling: Create power-dependent optical fiber models that account for stochastic impairments.
6. Integration of optical wireless communications (OWC) with AI-driven optimization techniques and the increasing use of machine learning to enhance system performance in dynamic environments.
7. Topology Optimization for Optical Networks (OCNs): Design AI-driven algorithms for intelligent routing and resource allocation.
8. Joint Optimization of OCNs and radio access networks (RANs): Develop integrated AI mechanisms for hybrid optical and radio networks.

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

Our research on the radio aspect focuses on machine learning, optimization, and neural network applications in radio signals processing and network traffic modeling. On the radio channel side, we explored machine learning based channel classification in 5G, studied indoor signal propagation characterization by using vector antennas, and developed a generative model based channel power estimation for THz V2I networks. On the radio signal and network side, we examined spatial-spectral THz networks with leaky-wave antennas and designed neural network-based digital pre-distortion for power amplifiers. Additionally, we implemented the autoencoder-based model to exploit the nonlinearity of the radio signal for radio transmitter identification and network security enhancement.

For the optic and network research, we have conducted research on AI methods for optical channel modeling, signal Processing, and optic network optimization. The results include: 1). AI-based optical channel models to incorporate power-dependent impairments; 2). DNNs to compensate for nonlinear optical impairments, including fiber nonlinearities, transceiver noise, and laser phase noise; 3). mutual information estimation methods to optimize optical network capacity; 4). Bayesian-trained deep learning models for optical network topology optimization; 5). On-demand wavelength adaptation strategies for elastic optical networks (EONs) to optimize resource allocation.

For the hybrid optical-radio Networks, we designed AI-driven strategies for joint optical-radio network optimization, integrating load-aware data analysis, and developed probabilistic ML-based optimization models for network topology adaptation.

We have developed productive international collaborations involves the active research organizations from EU, UK, Japan, Taiwan, China and Australia. The research results and knowledge are disseminated via activities such as the DIOR Communication Networks Summit (London, Hybrid, 2023) and Warwick Secure and Intelligent Communications Workshop (Warwick, 2023), invited talks at IEEE ICAIT, IEEE Summer Topicals, and IEEE International Conference on Photonics.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

The expected beyond state-of-the-art techniques or results developed in DIOR include: 1). data-driven methods for radio channel characterization, classification, and estimation/prediction; 2). neural networks models for radio signal nonlinearity and digital compensation; 3). autoencoders for secured radio link and transmitter identification, 4). data-driven models for radio network traffic characterization and prediction; 5). neural network architectures for mitigating nonlinear impairments, enabling real-time compensation in optical systems; 6). Fiber-optic parametric amplifier-based solutions for all-optical format conversion; 7). end-to-end learning-based constellation shaping in optic system; 8). gradient-free optical neural network architectures; 9). novel receiver architectures and mode permutation techniques for high signal integrity; 10). end-to-end learn based low-complexity CPR method to mitigate NLPN in multi-carrier modulation systems; 11). symbol-rate and bit-rate adaptability for hybrid FSO/RF link; 12). experimental long-distance field-deployed FSO link (integrating coherent optics, optimized coding, wavelength multiplexing, and atmospheric turbulence mitigation through optical pre-amplification) 14). Experimental hybrid FSO and millimeter-wave system without pre-alignment (computer vision algorithms and other automated alignment techniques).

The results of the DIOR project are set to make a socio-economic impact and a broader societal impact in the following areas: 1). Lowering operational costs for telecom providers through AI-driven network optimization; 2). The application of AI-optimized optical fiber and radio signal models and nonlinear impairment compensation techniques can significantly enhance spectral and energy efficiency of the radio and optic data transmission, contributing to sustainable telecommunication practices; 3). Strengthening global research collaboration among institutions in Europe, Asia, and Australia; 4). Providing early-stage researchers with hands-on experience in cutting-edge AI applications and extending their career prospects; 5). Raising awareness of AI’s role in communication networks through public engagement initiatives. 6). Providing insights for regulatory bodies and standardization efforts in AI-integrated network optimization.

dior-logo-color.png