Objectives

The solution will integrate the following elements: i) Data Lake will support collecting data from multiple sources, technologies and network layers using heterogeneous interfaces to take advantage of correlations in data; ii) Sandbox domains will be defined for model training and hosting simulators; iii) Intents will enable control loops and network automation. Special attention will be paid to the cooperation among intents defined for services and infrastructure; iv) AI pipelines create data analytics functions. An AIFO will monitor and manage AI pipelines; v) Explainable AI models will allow human operators to understand the decisions made by the AI algorithms; and vi) AI security based on encryption and token enforcement will provide data and AI models and inference integrity, confidentiality and right of use to distributed AI nodes in the AI pipelines.

At the orchestration/control layer, the goal is to design and develop an enhanced architecture that spans across various technology and administrative domains and is characterized by 3 layers of abstraction. A specific non-RT orchestrator will be in charge of each segment of the network (RAN, transport, and edge and cloud computing) that are divided into areas. Each area with a near-RT controller strategically placed close the data plane to speed up the local control loop and guarantee the service level. On top of the hierarchy, an e2e service orchestrator will coordinate the domain orchestrators and provide an e2e vision, resulting in a reduction of CAPEX from resource optimization. The entire architecture will support IBN, enabling service lifecycle management to both customer and infrastructure intents.

The e2e service orchestrator will be a central part of the ubiquitous intelligent environment characterizing the whole IBON framework; novel modules for AIOps and AIFO will be developed and integrated in the service orchestrator to: i) deploy proper AI pipelines with logical nodes in different domains and locations of the network to implement automation based on the current network status and service requirements (intents); ii) apply AI-based service (re)configuration for infrastructure and customer service assurance; and iii) apply complex e2e AI-based network (re)configuration operations to guarantee optimal resource allocation and ensure resource availability, without compromising service requirements. 

At the data plane, the goal is to increase data plane measurability and predictability by designing and implementing massive monitoring/telemetry of key performance parameters of optical transmission, service, and state of network elements. The platform will be able to adapt the data acquisition rate to specific monitoring objectives.

The goal is to enable distributed and communication and energy efficient AI model training and inference algorithms at the network edge, while considering application level requirements, communication, and compute resource constraints. To support that, IBON will design and implement a dynamic platform offering computation from edge to cloud and connectivity that rapidly adapts to changing application needs, mobility conditions of terminals and to the overall changing network status. IBON will provide a set of primitives and tools, so applications are able to use them to measure the performance e2e and automatically and dynamically scale and place computation functions in the closest edge node and make use of geo-replication functionalities to create truly distributed resilient systems with perceived zero latency; the required changes will be performed real-time.

The project will integrate the innovations targeted in the previous objectives and build a PoC designed to show the IBON ubiquitous AI from terminals to the core. In addition, IBON will disclose the approach, concepts, design methodologies, architectures, and performance results to a wide range of relevant stakeholders by means of communication, dissemination, standardization, and exploitation activities, while emphasizing openness and transparency.