Testbeds & Datasets

Experimental wireless platforms and public datasets supporting reproducible systems research.

Experimental wireless platforms and public datasets for reproducible research across Wi-Fi and mmWave MIMO systems.

Experimental Testbeds

Sub-6 GHz Wi-Fi ISAC testbed with distributed CSI and BFI capture stations

Sub-6 GHz Wi-Fi ISAC Testbed

IEEE 802.11ac/ax · CSI & BFI · SU/MU-MIMO · Commercial Wi-Fi

A configurable multi-node Wi-Fi ISAC platform for comparative collection of uncompressed Channel State Information (CSI) and standards-compliant compressed Beamforming Feedback Information (BFI). Distributed capture stations support controlled sensing experiments across devices, subjects, locations, orientations, and LoS/NLoS conditions.

Measurement Stack

  • CSI: Powered by Nexmon CSI for per-frame channel measurements from supported Broadcom Wi-Fi chipsets.
  • BFI: Powered by Wi-BFI for extracting beamforming feedback angles and reconstructing feedback matrices from IEEE 802.11ac/ax frames.

Research Enabled

  • Single- and multi-subject Wi-Fi sensing
  • Cross-environment domain adaptation
  • CSI-versus-BFI sensing comparisons
  • Beamforming-feedback radio fingerprinting
  • BeamSense, Si-FI, and CSI-BFI-HAR

WiSEC: Multiband Wi-Fi Sensing and Communication Testbed

18 IEEE 802.11ax NICs · 36 antennas · 2.4/5/6 GHz · Up to 160 MHz

A modular COTS Wi-Fi platform for large-scale multi-antenna and multiband experimentation. WiSEC integrates 18 independently configurable Intel NICs with two antennas each; the NICs can operate as separate devices or be combined into larger MIMO configurations, including a reconfigurable 6 × 6 antenna array.

System Capabilities

  • Independent operation across the 2.4, 5, and 6 GHz Wi-Fi bands
  • Reconfigurable antenna geometries and distributed MU-MIMO deployments
  • CSI and BFI acquisition from ongoing or triggered Wi-Fi transmissions
  • Simultaneous real-time collection and processing with onboard compute

Research Enabled

  • BeamID and DeepCSIv2 radio fingerprinting
  • Large-scale beamforming and MU-MIMO evaluation
  • Multiband MIMO sensing and communication
  • AI-driven channel analysis and domain adaptation

m3MIMO: Fully Digital mmWave MU-MIMO Testbed

57–64 GHz · Up to 1 GHz bandwidth · 8 × 8 MIMO · RFSoC SDRs

A fully digital mmWave platform built from three custom Zynq UltraScale+ RFSoC-based software-defined radios with Pi-Radio transceivers. Two radios provide eight transmit and receive streams each, while the third supports four channels, enabling flexible communication and sensing experiments.

System Capabilities

  • Point-to-point, SU-MIMO, and two-user MU-MIMO operation
  • Fully digital access to high-resolution mmWave channel measurements
  • OFDM-based frequency-domain multiplexing across up to 1 GHz bandwidth
  • Open-source control and experimentation software

Research Enabled

  • Communication-native mmWave sensing
  • MAGIC micro-gesture recognition
  • M3-CFR dataset collection and domain adaptation
  • Tracking-based beamforming and MIMO channel estimation

Sensing-Assisted Edge Computing Testbed

Wi-Fi 6 sensing · 160 MHz · 10K 360° video · Edge AI

A joint wireless-sensing and visual edge-computing platform that uses Wi-Fi measurements to localize and track environmental changes. The sensed locations are mapped to regions of interest in ultra-high-resolution video, allowing the system to offload and process only the visually relevant parts of each frame.

System Capabilities

  • 160 MHz Wi-Fi 6 sensing for object localization and tracking
  • Synchronized capture with a 10K 360° camera
  • Wireless-to-visual mapping for sensing-assisted region-of-interest selection
  • Controlled anechoic-chamber and real-world entrance-hall deployments

Research Enabled

SOAR outdoor UAV semantic offloading testbed with distributed wireless nodes and edge-computing devices

SOAR: Semantic-Aware UAV Edge Offloading Testbed

UAV experimentation · Multi-user MIMO · Semantic offloading · Edge AI

An outdoor aerial edge-computing platform for evaluating task-oriented wireless offloading under changing propagation conditions. SOAR couples application context with multi-user MIMO control so that communication resources can be adapted to the reliability needs of vision tasks rather than fixed network-level targets.

System Capabilities

  • Outdoor UAV operation with distributed wireless and edge-computing nodes
  • Line-of-sight and non-line-of-sight propagation experiments
  • Adaptive antenna, spatial-stream, and bandwidth configuration
  • Context-aware control using distributional deep reinforcement learning

Research Enabled

Distributed USRP software-defined radios deployed for an indoor over-the-air experiment

USRP-Based Over-the-Air Testbed

USRP SDRs · Indoor OTA experiments · Real wireless channels

A configurable software-defined radio platform for prototyping and evaluating wireless systems through indoor over-the-air experiments. Distributed USRPs support measurements across different transmitter and receiver locations.

System Capabilities

  • Indoor over-the-air transmission and reception
  • Configurable SDR-based wireless experiments
  • Measurements across different transmitter and receiver locations

Research Enabled

  • Evaluation of learning-based wireless systems
  • Physical-layer prototyping over real channels
  • OTA experiments supporting the PhyDNNs project

Research Datasets

M3-CFR

58 GHz mmWave · 8 × 8 MIMO CFR · Micro-gestures

A bi-static mmWave MIMO CFR dataset collected with the MAGIC platform for fine-grained recognition under domain shifts. It contains 10 micro-gesture classes, 3 subjects, 3 indoor environments, 3,000 labeled gesture instances, and approximately 1.5 million CFR frames.

CSI-BFI-HAR

Wi-Fi · Paired CSI and BFI · Human activity recognition

Paired channel state information and beamforming feedback traces for 20 activities. The collection supports single- and simultaneous multi-subject recognition across multiple indoor environments, device placements, orientations, and LoS/NLoS conditions.

DeepCSIv2

Wi-Fi · Compressed MIMO feedback · Radio fingerprinting

A 60.9 GB collection of over-the-air Wi-Fi traces and reconstructed beamforming feedback matrices for physical-layer identification of client devices. The release includes raw captures, processed V matrices, extraction tools, and the learning pipeline.

BeamID

Wi-Fi · MIMO beamforming feedback · Domain adaptation

Complex-valued beamforming feedback measurements for domain-adaptive radio fingerprinting. Organized by client radio and collection location, the dataset supports source-domain training and few-shot adaptation to changing deployment environments.