Toward Experimentation-as-a-Service in 5G/6G: The Plaza6G Prototype for AI-Assisted Trials

This paper introduces Plaza6G, the first operational Experiment-as-a-Service platform developed at CTTC that unifies cloud resources with 5G/6G infrastructure and leverages an LLM-based assistant to enable natural language-driven, automated, and reproducible wireless experimentation.

The Gist

Imagine you want to test a new app, like a video game or a ride-sharing service, but you need to see how it performs in different "worlds." Maybe you want to test it in a crowded subway (bad signal), a high-speed train (fast but shaky connection), or a remote desert (very slow connection).

In the past, doing this was like trying to build a miniature city in your living room just to test a toy car. You needed expensive equipment, a team of engineers to set it up, and days of manual work before you could even start driving.

Plaza6G is a new platform created by researchers in Barcelona that changes this completely. Think of it as "Netflix for Wireless Network Testing."

Here is how it works, broken down into simple concepts:

1. The "Magic Restaurant" (Experiment-as-a-Service)

Imagine a restaurant where you don't need to be a chef to order a complex meal. You just walk up to the counter (or use an app) and say, "I want a spicy burger with extra cheese, cooked for a group of 50 people."

In the old days, you'd have to buy the meat, the spices, the grill, and hire a chef to do it. With Plaza6G, you just type your request into a website. The system instantly "cooks" up a complete 5G network for you. It's called Experiment-as-a-Service (ExaS) because you aren't buying the hardware; you are renting the experience of testing your app on a network, exactly when you need it.

2. The "AI Sous-Chef" (The LLM Assistant)

The coolest part is how you talk to the system. You don't need to know complex computer code (like Python or C++). You can just talk to it in plain English.

• You say: "Set up three different 5G networks and check if my video app works fast enough on all of them."
• The AI (The "Plaza6G Brain"): Understands exactly what you mean. It acts like a super-smart sous-chef who knows the entire kitchen. It automatically grabs the right servers, sets up the virtual radio towers, and connects your phone to the network.

It uses a special type of AI (called an LLM) that has been trained on technical manuals and past experiments. It's like having a librarian who knows every book in the library and can instantly find the right recipe for your specific test.

3. The "Time-Traveling Simulator" (The Four Modes)

Plaza6G offers four different ways to test, like different levels of a video game:

• Simulation (The Movie): You test your app in a purely digital world. It's fast and cheap, like watching a movie of your app running.
• Emulation (The Video Game): You run your app on a computer that pretends to be a real phone network. It's very close to reality but still virtual.
• In-Lab (The Soundproof Room): This is where it gets real. They have a special room with no echoes (an anechoic chamber) where they put real phones and real radio towers. They can control the "weather" inside the room—making the signal weak or strong at will—to see how your app handles bad connections.
• Outdoors (The Real World): They have a real 5G network set up on rooftops outside their building. You can test your app while walking around the campus, dealing with real trees, buildings, and interference.

4. The "Instant Setup" (Why it matters)

Before Plaza6G, setting up a test like this could take days of manual work by experts.
With Plaza6G, it takes less than 10 minutes.

The paper shows two examples:

1. The Automated Runner: A software developer writes code for a new app. Every time they save their code, Plaza6G automatically tests it against three different 5G networks to make sure it doesn't crash. If it passes, the app gets updated. If it fails, the developer gets a report. It's like a self-driving car that tests itself before you buy it.
2. The Human Explorer: A researcher wants to see how a video stream behaves when the signal gets worse and worse. They use the web portal to set up a real phone and a real tower in the soundproof room. Then, they manually turn a dial to make the signal "worse" and watch how the video quality drops, all while chatting with the AI assistant to get data.

The Big Picture

Plaza6G is a bridge between the Cloud (where we store data) and the Air (where wireless signals travel).

• For Developers: It means they can test their apps on 5G/6G networks without needing a PhD in engineering or a million dollars in equipment.
• For Researchers: It means experiments can be repeated exactly the same way by anyone, anywhere, making science more reliable.
• For the Future: It paves the way for 6G, where networks will be so complex that only an AI-assisted, automated system can manage the testing effectively.

In short, Plaza6G turns the complex, messy world of wireless testing into something as easy as ordering a pizza: You ask, it delivers, and you get to enjoy the result.

Technical Summary

1. Problem Statement

The increasing complexity of next-generation (xG) wireless networks (5G/6G) has outpaced the capabilities of traditional experimentation environments. Current challenges include:

• Fragmentation: Existing testbeds often separate cloud computing resources from realistic radio infrastructure, making end-to-end validation difficult.
• High Barrier to Entry: Most platforms require specialized domain expertise, manual scripting, and complex configuration, limiting accessibility for developers and researchers.
• Lack of Reproducibility: Manual interventions and fragmented interfaces hinder the reproducibility of experiments and the integration of network testing into modern software development lifecycles (CI/CD).
• Scalability Issues: Scaling experiments across simulation, emulation, and physical over-the-air (OTA) testing within a single unified framework remains a significant challenge.

2. Methodology and Architecture

The authors present Plaza6G, the first operational Experiment-as-a-Service (ExaS) platform developed at CTTC (Barcelona). It unifies cloud-scale computing with real 5G/6G infrastructure under an AI-assisted orchestration model.

Core Architecture

The platform adopts a three-layer architecture:

1. User Layer: Provides access via a Web Portal and REST API. Crucially, it integrates a Large Language Model (LLM) assistant that interprets natural language requests.
   • LLM Backend: Deployed locally at CTTC, utilizing Retrieval-Augmented Generation (RAG) for up-to-date technical knowledge and Low-Rank Adaptation (LoRA) for continuous domain fine-tuning. This allows users to describe experiments in natural language (e.g., "Deploy my app across three 5G cores...") rather than writing complex scripts.
2. Orchestration Layer: A policy-driven scheduler that translates user intents into executable experiment graphs, managing resource allocation and lifecycle (provisioning, execution, teardown).
3. Infrastructure Layer: A heterogeneous pool of resources including:
   • Compute: >3,000 CPU cores, 30 vGPUs (NVIDIA L40S), 500 TB storage, and Kubernetes support.
   • Network: Multiple 5G core implementations (Free5GC, Open5GS, Cumucore, OAI-CN) supporting network slicing.
   • Radio: Programmable RANs (Amarisoft Callbox, O-RAN, srsRAN) and both emulated (UERANSIM) and physical UEs (Android smartphones).

Experiment Modalities

Plaza6G supports four distinct modalities under unified orchestration, allowing progressive validation:

1. Simulation: Discrete-event simulators (ns-3, Komondor).
2. Emulation: Virtualized protocol stacks (srsRAN, UERANSIM).
3. In-lab OTA: Physical equipment in a four-chamber anechoic facility for repeatable, controlled radio propagation conditions.
4. Outdoors: A dual-site private 5G network operating in sub-6 GHz and mmWave bands for realistic field trials.

3. Key Contributions

• First Operational ExaS Platform: Plaza6G is the first system to merge cloud-scale resources with real wireless infrastructure under an AI-assisted model, bridging the gap between cloud elasticity and realistic wireless testing.
• AI-Driven Zero-Touch Orchestration: By leveraging LLMs with RAG and LoRA, the platform allows users with minimal domain expertise to deploy complex, multi-domain experiments via natural language or API, eliminating manual scripting barriers.
• Unified Multi-Modal Framework: It uniquely integrates simulation, emulation, controlled lab testing, and outdoor field trials into a single platform, enabling seamless transition from theoretical models to field deployment.
• Experiment-as-Code (ExaC) & Reproducibility: All actions are logged as machine-readable descriptors, ensuring full traceability, version control, and reproducibility. The platform supports automated CI/CD integration.

4. Results and Demonstrations

The paper validates Plaza6G through two representative use cases:

Use Case A: Automated CI/CD Network Acceptance (Emulation)

• Scenario: A developer uses the REST API to validate an application (iperf3) across three different 5G core implementations (Open5GS, Free5GC, OAI-CN) simultaneously.
• Process: The LLM interprets the request, generates an experiment plan, and the scheduler deploys three isolated emulated networks.
• Outcome:
  • Performance: All cores achieved mean throughput significantly exceeding the 50 Mbps acceptance threshold (reaching up to ~350 Mbps).
  • Efficiency: Setup time was reduced to under 10 minutes per experiment, representing an order-of-magnitude improvement over manual procedures.
  • Integration: Demonstrated successful "network acceptance testing" where CI/CD pipelines can automatically proceed to staging only after KPI thresholds are met.

Use Case B: Controlled Over-the-Air (OTA) Experiment (In-lab)

• Scenario: A researcher uses the Web Portal to deploy a physical 5G network (Free5GC + Amarisoft Callbox + Android UE) inside an anechoic chamber.
• Process: The system automatically provisions resources and establishes connectivity. The user then interacts manually (via Vysor) to test application behavior under programmable path loss and attenuation.
• Outcome:
  • Enabled interactive, exploratory studies on how channel degradation affects throughput, latency, and Quality of Experience (QoE).
  • The LLM assistant provided real-time contextual data (signal conditions, stats) to aid the user's interpretation without restricting the experiment to predefined workflows.

5. Significance and Future Outlook

• Democratization of xG Research: By lowering the expertise barrier through AI assistance, Plaza6G makes advanced 5G/6G experimentation accessible to a broader range of developers and researchers.
• Reproducibility Standard: The machine-readable descriptors and unified logging establish a new standard for reproducible wireless research, addressing a critical gap in the field.
• Future Directions:
  • Policy-Aware Orchestration: Optimizing for cost, energy, and radio resource usage.
  • Safety Validation: Verifying LLM-generated actions to ensure correctness.
  • Federated Testbeds: Integrating with external testbeds (e.g., via ETSI OpenSlice) for geographically distributed, multi-domain experiments.

In conclusion, Plaza6G represents a paradigm shift from manual, fragmented testbeds to an automated, AI-driven, and unified service model, accelerating the validation and deployment of next-generation wireless technologies.