Colosseum: Large-Scale Wireless Experimentation Through Hardware-in-the-Loop Network Emulation

Francesco Restuccia, Leonardo Bonati, Pedram Johari, Michele Polese, Salvatore D'Oro, Subhramoy Mohanti, Miead Tehrani Moayyed, Davide Villa, Shweta Shrivastava, Chinenye Tassie, Kurt Yoder, Ajeet Bagga, Paresh Patel, Ventz Petkov, Michael Seltser, Francesco Restuccia 0001, Abhimanyu Gosain, Kaushik R. Chowdhury, Stefano Basagni, Tommaso MelodiaDyspan 2021 · DOI

TL;DR

Colosseum is a groundbreaking, publicly accessible wireless testbed that bridges the gap between simulation and real-world experimentation. It allows researchers to emulate complex, large-scale wireless scenarios with hardware-in-the-loop fidelity, enabling the development and validation of next-generation wireless technologies.

What Problem Does This Paper Address?

Developing and testing new wireless technologies, especially for 5G and beyond, is increasingly challenging. Traditional network simulations often lack the fidelity to accurately capture real-world complexities like hardware imperfections, interference, and diverse channel conditions. Conversely, building and accessing physical testbeds at scale is prohibitively expensive and logistically difficult. This creates a significant gap for researchers trying to validate innovative solutions in realistic, large-scale environments. Colosseum aims to fill this gap by providing a programmable, hardware-in-the-loop emulation platform that is accessible to the research community.

Methods

Colosseum's architecture is built around 128 Standard Radio Nodes (SRNs), each equipped with powerful CPUs, GPUs, and USRP X310 Software-Defined Radios (SDRs). These SRNs are connected to a Massive Channel Emulator (MCHEM), which uses FPGA-based Finite Impulse Response (FIR) filters to accurately model complex wireless channel conditions, including path loss, fading, and multipath effects. The system also includes a Traffic Generator (TGEN) for emulating IP traffic flows and a management infrastructure for resource allocation and experiment control.

Key methodological aspects include:

  • Hardware-in-the-Loop Emulation: By integrating real SDRs with a sophisticated channel emulator, Colosseum provides a more realistic environment than pure software simulations.
  • Programmable RF Scenarios: The MCHEM can reproduce a vast array of channel conditions, allowing researchers to model diverse environments from urban alleys to open terrains.
  • Softwarized Network Stacks: Users can deploy custom, containerized network protocol stacks (e.g., cellular, Wi-Fi) on the SRNs, enabling experimentation with novel software-defined networking approaches.
  • Large-Scale Emulation: With 128 SRNs and the ability to emulate up to 65,535 wireless channels over areas up to 1 km², Colosseum supports experiments at an unprecedented scale.
  • Interactive and Batch Modes: The platform offers both interactive modes for rapid prototyping and debugging, and batch modes for automated, large-scale data collection and benchmarking.

Datasets

Not discussed in the paper. The paper focuses on the platform's capabilities for creating and utilizing experimental data, rather than pre-existing datasets.

Key Insights and Ideas

  • Bridging the Simulation-Reality Gap: Colosseum's hardware-in-the-loop approach is crucial for validating wireless solutions that are sensitive to physical layer characteristics and hardware imperfections, which simulations often miss.
  • Enabling Large-Scale, Repeatable Experiments: The platform's scale and precise emulation capabilities allow for reproducible experiments that were previously impossible outside of expensive, dedicated facilities.
  • Democratizing Advanced Wireless Research: By making such a powerful testbed publicly available, Colosseum significantly lowers the barrier to entry for researchers working on cutting-edge wireless technologies.
  • Facilitating AI/ML in Wireless: The ability to generate large, diverse, and realistic datasets makes Colosseum an ideal "wireless data factory" for training and evaluating AI/ML algorithms for network optimization and control.
  • Supporting Emerging Technologies: The platform is designed to support a wide range of use cases, from cellular and Wi-Fi to spectrum sharing and UAV communications, making it a versatile tool for future research.

Key Results

The paper demonstrates Colosseum's effectiveness through several exemplary use cases:

  • Cellular Networking: Throughput measurements show how Colosseum can accurately model downlink throughput under varying traffic conditions (e.g., 1 Mbps, 2 Mbps, 4 Mbps per user). It also demonstrates the impact of user mobility and distance from base stations on spectral efficiency, showing tight confidence intervals that confirm repeatability.
  • O-RAN: The platform was used to instantiate an O-RAN-managed cellular infrastructure, showcasing how AI-driven DRL agents can dynamically optimize scheduling policies for slices like URLLC, achieving superior performance compared to static policies.
  • Wi-Fi: Experiments measuring Signal-to-Interference-plus-Noise Ratio (SINR) for Wi-Fi nodes showed a degradation of 3.18 dB on average due to interference from cellular users, highlighting the platform's ability to model spectrum sharing scenarios.
  • Spectrum Sharing: The paper illustrates how Colosseum can emulate heterogeneous networks with cellular and Wi-Fi nodes sharing the same spectrum, demonstrating the interference caused by Wi-Fi transmissions on cellular users' Channel Quality Information (CQI) and uplink error rates.
  • Unmanned Aerial Vehicles (UAVs): The platform was used to emulate UAV networks, showing that using a maximum weighted matching algorithm for relay node selection resulted in higher SINR for squads compared to random allocation.

Limitations and Future Work

Limitations:

  • Emulation Area: Currently, Colosseum is designed to emulate terrains up to 1x1 km. This is a consequence of the maximum propagation delay that can be emulated by the MCHEM's FPGAs.
  • Directional Transmissions and Higher Frequencies: Colosseum was initially designed for sub-6 GHz omnidirectional transmissions and cannot currently emulate directional transmissions or higher frequency bands.

Future Work:

  • Expanding Emulation Area: Plans are underway to extend the emulation capabilities to cover larger terrains, up to 100x100 km, to better represent NRDZ scenarios.
  • Supporting mmWave and Directional Transmissions: The platform will be enhanced to emulate beamforming-based, directional communications for mmWave networks.
  • Increased Bandwidth: A new Colosseum quadrant will be developed to emulate very large baseband bandwidths, up to 2 GHz, to support 5G and beyond requirements.
  • AI Jumpstart Integration: Integration of advanced AI hardware (NVIDIA DGX A100 nodes) will enable faster training of large wireless datasets, real-time AI-driven 5G signal processing, and model-free adaptation of wireless networks.
  • NRDZ Support: Colosseum will be expanded to support new spectrum bands, scenarios, and power levels for National Radio Dynamic Zones (NRDZs), acting as a management and optimization system for spectrum sharing.

Summary updated

BibTeX
@inproceedings{restuccia2021colosseum,
  author = {Francesco Restuccia and Leonardo Bonati and Pedram Johari and Michele Polese and Salvatore D'Oro and Subhramoy Mohanti and Miead Tehrani Moayyed and Davide Villa and Shweta Shrivastava and Chinenye Tassie and Kurt Yoder and Ajeet Bagga and Paresh Patel and Ventz Petkov and Michael Seltser and Francesco Restuccia and Abhimanyu Gosain and Kaushik R. Chowdhury and Stefano Basagni and Tommaso Melodia},
  title = {{Colosseum: Large-Scale Wireless Experimentation Through Hardware-in-the-Loop Network Emulation}},
  booktitle = {Dyspan},
  year = {2021},
  doi = {10.1109/dyspan53946.2021.9677430},
}
Figure