Periodic Reporting for period 2 - ARIADNE (Artificial Intelligence Aided D-band Network for 5G Long Term Evolution)
Reporting period: 2021-05-01 to 2023-07-31
With the aspiration to transform the current (5G) wireless thinking from focusing on "local" network improvements (e.g. isolating the radio access level or the resources management level etc.), to realizing a longer term vision of pervasive mobile virtual services, through a network managing computing and connectivity functions in an integrated way, ARIADNE envisions to bring together a novel high frequency advanced radio architecture and an Artificial Intelligence (AI) network processing and management approach in a unified system beyond 5G concept.
The vision of ARIADNE is to investigate, theoretically analyze, design, develop, and showcase in a proof of concept demonstrator, an innovative wireless communications concept addressing networks beyond 5G, in which ultra-high spectral efficient and reliable communications in the bandwidth-rich D-band can be dynamically established and reconfigured by Machine Learning (ML)-based design and intelligent network management, in both “Line of Sight” (LOS) and “Non-Line of Sight” (NLOS) environments (Figure 1).
In this respect, the ARIADNE is working towards identifying and assessing the critical technology gaps and inventing, optimizing and demonstrating the appropriate enablers, expected to catalyze the road to beyond 5G. In particular, the ARIADNE approach is established, developed and evaluated based on its three pillars:
• PILLAR I: D-band for 100 Gbit/s reliable wireless connectivity,
• PILLAR II: Communications beyond the Shannon paradigm, by means of metasurfaces for NLOS/obstructed LOS connectivity, in order to guarantee connectivity reliability, by making the environment itself reconfigurable
• PILLAR III: Artificial Intelligence based wireless system concept, by means of ML approaches to optimize the architecture, the signal and data processing and all network management functions.
The vision of ARIADNE is to investigate, theoretically analyze, design, develop, and showcase in a proof of concept demonstrator, an innovative wireless communications concept addressing networks beyond 5G, in which ultra-high spectral efficient and reliable communications in the bandwidth-rich D-band can be dynamically established and reconfigured by Machine Learning (ML)-based design and intelligent network management, in both “Line of Sight” (LOS) and “Non-Line of Sight” (NLOS) environments (Figure 1).
In this respect, the ARIADNE is working towards identifying and assessing the critical technology gaps and inventing, optimizing and demonstrating the appropriate enablers, expected to catalyze the road to beyond 5G. In particular, the ARIADNE approach is established, developed and evaluated based on its three pillars:
• PILLAR I: D-band for 100 Gbit/s reliable wireless connectivity,
• PILLAR II: Communications beyond the Shannon paradigm, by means of metasurfaces for NLOS/obstructed LOS connectivity, in order to guarantee connectivity reliability, by making the environment itself reconfigurable
• PILLAR III: Artificial Intelligence based wireless system concept, by means of ML approaches to optimize the architecture, the signal and data processing and all network management functions.
Already during the first six months on its life time, the ARIADNE project consortium defined a reference system model as basis for future investigations, which will be carried out within the project scope. The ARIADNE system concept includes three general use cases and corresponding scenarios, allowing detailed system modelling and design as well as further planned research activities; 1 - Outdoor backhaul/fronthaul fixed topology network, 2 - RIS based NLOS connectivity, 3 - Ad-hoc connectivity in moving network topology.
After one year on its lifetime, the ARIADNE project specified its system model as basis for future investigations which will be carried out within the project scope. An analysis of the D-band directional link, including consideration of suitable channel modelling approaches, possible ways forward in performance evaluation, and preliminary studies on appropriate application of machine learning techniques in this context have been concluded during the first project 18 months. Furthermore, bases for application of Reconfigurable Intelligent Surfaces and reconfigurable antennas for D-Band have been laid down.
During the second part of the project, the channel propagation issues in D band have been investigated and concluded, which involved several measurements campaigns, establishment and application of appropriate modelling and simulation tools, resulting with channel characterization and models for further usage. Furthermore, advanced modelling and analysis of RIS aided wireless systems have been performed, to conclude the ARIADNE’s theoretical framework beyond Shannon. Also, a complete D-band radio was designed and implemented, including baseband and RF front-end units and high-gain antennas, leading to the successful execution of the Point-to-point D-band demonstrator. Further work was dedicated to explore using AI/ML techniques to optimize physical and MAC layers of RIS aided wireless systems and ensure D-band network intelligence tackling beamforming, association of access points to mobile users, resource network management, etc.
The ARIADNE project implemented the following demos:
• Point-to-point D-band demonstrator, achieving error-free communication in distances larger than 200m
• Presentation of RIS (designed and manufactured by ARIADNE) in a NLOS scenario
• Software demonstrators:
o AI/ML application for LoS-aware directional connectivity
o Deep Reinforcement Learning for B5G Wireless Communications
o Complex Event Forecasting for Handover Reduction
After one year on its lifetime, the ARIADNE project specified its system model as basis for future investigations which will be carried out within the project scope. An analysis of the D-band directional link, including consideration of suitable channel modelling approaches, possible ways forward in performance evaluation, and preliminary studies on appropriate application of machine learning techniques in this context have been concluded during the first project 18 months. Furthermore, bases for application of Reconfigurable Intelligent Surfaces and reconfigurable antennas for D-Band have been laid down.
During the second part of the project, the channel propagation issues in D band have been investigated and concluded, which involved several measurements campaigns, establishment and application of appropriate modelling and simulation tools, resulting with channel characterization and models for further usage. Furthermore, advanced modelling and analysis of RIS aided wireless systems have been performed, to conclude the ARIADNE’s theoretical framework beyond Shannon. Also, a complete D-band radio was designed and implemented, including baseband and RF front-end units and high-gain antennas, leading to the successful execution of the Point-to-point D-band demonstrator. Further work was dedicated to explore using AI/ML techniques to optimize physical and MAC layers of RIS aided wireless systems and ensure D-band network intelligence tackling beamforming, association of access points to mobile users, resource network management, etc.
The ARIADNE project implemented the following demos:
• Point-to-point D-band demonstrator, achieving error-free communication in distances larger than 200m
• Presentation of RIS (designed and manufactured by ARIADNE) in a NLOS scenario
• Software demonstrators:
o AI/ML application for LoS-aware directional connectivity
o Deep Reinforcement Learning for B5G Wireless Communications
o Complex Event Forecasting for Handover Reduction
Important highlights of these advances beyond the state-of-the-art with respect to the D-band wireless connectivity pillar (I) include
• ARIADNE use cases and system requirements specifications,
• D-band channel measurements and characterization,
• The design and manufacturing of A/D and D/A converters board for RF front-end,
• The development and implementation of an XPIC algorithm in the baseband processing unit,
• The design and implementation of a broadband transceiver chipset at D-band.
• The error-free communication achieved over the D-band link at a distance larger than 200m,
• The development of an LC reflect-array lab demonstrator at W-band,
• The design of metasurface topologies for multi-frequency anomalous reflectors and beam splitters at D-band,
• The design of low-complexity beam-tracking algorithm for D-band wireless systems.
Furthermore, important highlights of advances beyond the state-of-the-art with respect to the Communications beyond the Shannon pillar (II) include
• capacity-achieving scheme for RIS-assisted communications
• scaling laws of RIS-assisted communications in the near-field and far-field regimes,
• analytical studies in order to model RIS as a MIMO system and/or as a radiating sheet,
• analytical assessment of the beamforming efficiency in RIS aided links,
• optimal RIS placement and orientation scheme,
• analytical evaluation of the impact of beam misalignment, blockage, rain interference, and hardware imperfections,
• prediction-based tracking algorithm proposed,
• quantitative analysis of network interference generated by RIS,
• design, implementation and demonstration of a D-band link into a shadow region via reflection in anomalously reflecting metasurfaces.
Finally, important highlights of advances beyond the state of the art with respect to the Artificial Intelligence based wireless system concept pillar (III) include:
• a low-complexity algorithm design based on the stochastic block gradient descent method for optimum beamforming in the presence of random blockage,
• an Intelligent Spectrum Learning algorithm design exploiting Convolutional Neural networks,
• application of deep unfolding neural network for channel estimation in RIS-aided systems,
• optimum beam selection in multi-user multiple-input-single-output (MISO) downlink scenarios using distributed Deep Reinforcement Learning (DRL) methods,
• the design and evaluation of a protocol for minimizing Initial Access delay in D-band wireless systems and RIS-empowered D-band wireless systems,
• formulation and solution of the cell-user assignment problem using the proposed AI/ML hybrid framework combining Metaheuristics and Machine Learning algorithms, and
• cell-user assignment problem exploiting Complex Event Forecasting (CEF) leading to handover reduction.
In respect to techno-economic analysis performed in the project, business models and roles in the future communications systems, in particular the RIS aided, have been explored, where an assessment of a particular backhauling case has been done.
• ARIADNE use cases and system requirements specifications,
• D-band channel measurements and characterization,
• The design and manufacturing of A/D and D/A converters board for RF front-end,
• The development and implementation of an XPIC algorithm in the baseband processing unit,
• The design and implementation of a broadband transceiver chipset at D-band.
• The error-free communication achieved over the D-band link at a distance larger than 200m,
• The development of an LC reflect-array lab demonstrator at W-band,
• The design of metasurface topologies for multi-frequency anomalous reflectors and beam splitters at D-band,
• The design of low-complexity beam-tracking algorithm for D-band wireless systems.
Furthermore, important highlights of advances beyond the state-of-the-art with respect to the Communications beyond the Shannon pillar (II) include
• capacity-achieving scheme for RIS-assisted communications
• scaling laws of RIS-assisted communications in the near-field and far-field regimes,
• analytical studies in order to model RIS as a MIMO system and/or as a radiating sheet,
• analytical assessment of the beamforming efficiency in RIS aided links,
• optimal RIS placement and orientation scheme,
• analytical evaluation of the impact of beam misalignment, blockage, rain interference, and hardware imperfections,
• prediction-based tracking algorithm proposed,
• quantitative analysis of network interference generated by RIS,
• design, implementation and demonstration of a D-band link into a shadow region via reflection in anomalously reflecting metasurfaces.
Finally, important highlights of advances beyond the state of the art with respect to the Artificial Intelligence based wireless system concept pillar (III) include:
• a low-complexity algorithm design based on the stochastic block gradient descent method for optimum beamforming in the presence of random blockage,
• an Intelligent Spectrum Learning algorithm design exploiting Convolutional Neural networks,
• application of deep unfolding neural network for channel estimation in RIS-aided systems,
• optimum beam selection in multi-user multiple-input-single-output (MISO) downlink scenarios using distributed Deep Reinforcement Learning (DRL) methods,
• the design and evaluation of a protocol for minimizing Initial Access delay in D-band wireless systems and RIS-empowered D-band wireless systems,
• formulation and solution of the cell-user assignment problem using the proposed AI/ML hybrid framework combining Metaheuristics and Machine Learning algorithms, and
• cell-user assignment problem exploiting Complex Event Forecasting (CEF) leading to handover reduction.
In respect to techno-economic analysis performed in the project, business models and roles in the future communications systems, in particular the RIS aided, have been explored, where an assessment of a particular backhauling case has been done.