Demo: A Flexible, Declarative Framework for Deploying and Reproducing 6G Experiments

Abstract

5G and emerging 6G systems are increasingly modular and programmable, combining heterogeneous RAN, core, and AI components through standardized interfaces [1]. While this flexibility enables new research opportunities, it also increases the complexity of configuring and deploying experimental infrastructures. In practice, experiments remain coupled to specific implementations, hardware, and testbeds (e.g., PAWR, SLICES), making deployment complex and limiting reproducibility [2]. We propose a flexible framework that simplifies the deployment and sharing of heterogeneous 6G infrastructures. At its core is a declarative meta-model, designed from 3GPP and O-RAN specifications and existing implementations, that provides a common abstraction for network components and their configuration. Implemented using Pydantic, it supports multiple representations, automated validation, and transparent reconciliation of high-level descriptions into implementation-specific parameters. A complementary deployment model captures runtime state. Standardised management interfaces and idempotent actions allow these models to be composed into arbitrary workflows, from infrastructure provisioning and 5G deployment to AI-RAN experiments. This enables heterogeneous components and implementations to be seamlessly combined while making experimental infrastructures easier to deploy, reproduce, and share. The framework provides a Python library of management units for provisioning resources, deploying core and RAN implementations, configuring radio units, and running experiments across heterogeneous environments. It deliberately prioritizes flexibility over orchestration, leaving workflow scheduling, global validation, and recovery to users or higher-level tools. Its simple interfaces support the development of APIs, CLIs, and LLM-based assistants that automate experiment-specific workflows. We demonstrate this approach with a CLI for interactively deploying tailored 5G environments on the SophiaNode testbed and with a LLM Chatbot to ease experiment deployments in SLICES-RI [3]. The management units can also be containerized for CI/CD and GitOps integration.