Intent-driven NaTive AI architecturE supporting Compute-Network abstraction and Sensing at the Deep Edge

Project Information

6G-INTENSE

Grant agreement ID: 101139266

DOI

10.3030/101139266

EC signature date 1 December 2023

Start date 1 January 2024

End date 31 December 2026

Funded under

• Digital, Industry and Space

Total cost

€ 4 248 308,75

EU contribution

€ 3 999 819,75

Coordinated by

ATHINA-EREVNITIKO KENTRO KAINOTOMIAS STIS TECHNOLOGIES TIS PLIROFORIAS, TON EPIKOINONION KAI TIS GNOSIS
Greece

Periodic Reporting for period 1 - 6G-INTENSE (Intent-driven NaTive AI architecturE supporting Compute-Network abstraction and Sensing at the Deep Edge)

Reporting period: 2024-01-01 to 2025-06-30

Summary of the context and overall objectives of the project

6G-INTENSE proposes a new System Architecture for 6G, to deliver “6G as a Smart Service Execution platform”, fully in line with the vision of sustainable infrastructure sharing to reduce space and energy costs, and encouraging collaboration among all members of the value chain under a unified Network-Compute fabric. Key contribution is a novel automation architecture with a Native AI toolkit facilitating intent declaration, negotiation and decision automation across autonomous domains, termed as Distributed Intent-driven Management and Orchestration (DIMO). Moreover, Sensing is adopted as a key enabler, helping navigate the complexities and lack of reliability of the Deep Edge. To tackle this challenges, 6G-INTENSE has set a clear roadmap that is defined by delivering the following innovative tools:
(1) A Distributed, Intent-driven Management & Orchestration plane (DIMO) for autonomously reconciled intent declarations in multi-stakeholder ecosystems,
(2) An intent-based Abstraction Framework for the 6G Network-Compute Fabric,
(3) A scalable Compute Interconnection solution based on SD-WAN for self-organized Service Mesh deployments,
(4) The first concrete AI Native Toolkit implementation, offering intent decomposition, actuation and reconciliation towards fully Autonomous Domains,
(5) A federated AI plane for multi-stakeholder 6G ecosystems, via knowledge optimization, synthesis and intent propagation mechanisms,
while, at the same time,
(6) supporting privacy-preserving Joint Communication & Sensing for resilience and dependability at the Deep Edge.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

6G-INTENSE activities are deployed in mainly 4 technical Work Packages (WP2-5) which are supported by a Work Package dedicated to Management of the project (WP1) and another one dedicated to the Impact Maximisation and Outreach (WP6).
WP5 has not started its activities yet but the other three technical WPs (WP2, WP3, and WP4) report progress achievements according to the following.

WP2- Architecture design and Requirements analysis
- The technical requirements of the architecture and Proof of Concept scenarios for the project have been elaborated. The KPIs and their framework which will be evaluated throughout the pilot phase were defined. A market analysis was also performed highlighting the potential of combining Native AI with 6G Networks. A preliminary business model for “6G as a Smart Execution Platform” has been introduced.
- The functional and non-functional requirements and specifications for each of the pillars of the DIMO framework (service management and orchestration, resource management and orchestration, and Native AI) have been identified and described.
- The overall architecture of 6G-INTENSE, covering the t-DMO, DMO, and NCF operational layers and a detailed description of the integration of NATIVE AI components, including the Intent management process and the closed-control loop workflow has been presented.

WP3 – Native AI toolkit for Zero Touch operations
- Development of 3 stage approach to intent translation:
1) Retriever for few-shot learning
2) Established Validation process
3) User feedback utilisation
- First versions of trained LLMs for standards compliant intent generation measuring interoperability and accuracy.
- Development of the LLM Chatbot, guiding user driven intent creation for the t-DMO.
- Development of a context-aware AI framework for translation into network service provide compliant language.
- Development of intent benchmarking tool for initial and iterative testing ensuring models remain reliable, accurate, and compliant.
- Established intent deployment processes facilitating mapping of business service requests into structured representations following TMF standard followed by the decomposition of user requests into service deployments and resource intents.

WP4 - Network-Compute Fabric for 6G networksthe
- The development of the Cross-Domain Integration component of NCF has started by implementing the communication module for SD-WAN LMO’s interfaces, which remains to be tested.
- A Resource Abstraction and Exposure framework has been implemented to aggregate resources from infrastructure providers of different technologies and domains
- The blueprint of the algorithmic framework for orchestration and management of composite sensing services, as well as of joint communication and sensing functions in the edge and deep edge domains has been concluded.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

6G-INTENSE innovations towards its objective are:
[1] Novel automation architecture with a Native AI toolkit facilitating Intent declaration, negotiation, and decision automation across autonomous domains, termed as Distributed Intent-driven Management and Orchestration (DIMO)
[2] Separation of service and resource management using the concept of Domain Management and Orchestrator (DMO), Network Compute Fabric (NCF), and Local Management Orchestrator (LMO)
[3] Leverage emerging sensing capabilities to increase the dependability of the Deep Edge
[4] Build a tenant management layer to guarantee a scalable and hierarchical zero-touch service management (ZSM) layer, introducing the principle of t-DMO
[5] Native AI toolkit that defines the needed functions for Intent Management:
• Intent life cycle handling: LLM, Reasoning function
• Closed Control Loop (CCL): Reinforcement Learning with Human Feedback (RLHF)
[6] Novel and programmable overlay networking technologies for Compute Inter-Connection (CIC) using SD-WAN.