AI is a key component of the Open Radio Access Network (O-RAN) and is used for various optimization tasks. The network architecture provides data and control interfaces between the RAN Intelligent Controller (RIC) and the Radio Access Network (RAN), enabling AI agents to optimize the RAN across different time scales.

A typical application of AI in O-RAN is the xApp, which operates in the Near-Real-Time RIC to perform learning-based optimization at the NET, MAC, and PHY layers. We have developed a 5G O-RAN system comprising three RUs and 15 cellphones from diverse vendors to evaluate different online learning algorithms. These algorithms run within the xApp to optimize spectrum and energy efficiency in near-real-time, with a time scale ranging from 10 ms to 1 s.

The key features of our design include:

• Near-real-time online optimization for resource slicing, inter-cell interference management, and energy efficiency
• Enhanced optimization for actor and critic agents
• Diffusion policy for reinforcement learning
• Integrated hardware and software design

While the literature contains a substantial amount of work on WiFi CSI-based human recognition, existing approaches face two key limitations: i) they require at least two devices (one transmitter and one receiver); and ii) the phase information over time is not coherent. In this project, we developed a new hardware component for WiFi devices that enables a single WiFi device to detect human body boundaries. The accompanying video demonstrates how a single WiFi device can identify human skeletons in a dynamic scene. The key features of our design include:

• Human skeleton esetimation using a single WiFi device
• No channel state information (CSI) required
• Radar-like sensing (phase-coherent sensing)
• Integrated hardware and software design

Ubiquitous sensing technologies enable smart home environments to monitor individuals' health status in their daily lives. In this project, we developed a 6 GHz FMCW radar system capable of measuring heartbeat and breathing rate in a contactless manner. Our system leverages the fact that radio signals are influenced by inhaling and exhaling motions, as well as the skin vibrations caused by heartbeats. The accompanying video demonstrates our system's ability to measure heartbeats and breathing rates in real time. The key features of our design include:

• Real-time measurement of heartbeat and breathing rate
• Integrated hardware and software design
• Dedicated signal processing pipeline

Reliable 5G mmWave communications are critical for many applications. However, maintaining the reliability of mmWave communications is challenging in scenarios where networks are affected by radio interference or jamming attacks. In this project, we present a joint analog and digital beamforming scheme for a 5G mmWave receiver designed to decode data packets in the presence of jamming signals. The beamforming scheme incorporates two key techniques: i) an online-learning Bayesian optimization framework for analog beamforming, and ii) a Modified Minimum Mean Square Error (M-MMSE) detector for digital beamforming. The combination of analog and digital beamformings enables the receiver to suppress jamming signals in both the analog and digital domains, allowing it to successfully decode data packets despite interference. We have built a prototype of the proposed receiver on a 28 GHz testbed and extensively evaluated its performance in different scenarios. Key features of this system include:

• Joint analog and digital beamforming
• Online learning for beam selection
• Real-time jamming mitigation
• Realistic OTA evaluation

This real-time mmWave (28 GHz) communication system demonstrates that horizontal and vertical polarizations provide two spatial degrees of freedom (DoF) for transmitting two independent data streams. The components of this system include: i) USRP X310 devices for baseband signal processing, ii) Sivers EVK02004 for 28 GHz up/down conversion, and iii) computers for digital signal processing. Most of the code was written in C++. This mmWave communication system has the following features:

• Real-time video streaming
• Real-time beam steering at Tx and Rx
• Support 2x2 MIMO transmission
• Support Tx and Rx synchronization

In this project, we designed a practical underlay spectrum sharing scheme for cognitive radio networks (CRNs) in which primary users are unaware of the presence of secondary users. The key components of our scheme include two MIMO-based interference cancellation (IC) techniques to manage cross-network interference on the secondary network side. We implemented the proposed underlay spectrum sharing scheme on a GNURadio-USRP2 wireless testbed. Experimental results demonstrate that secondary users can achieve an average spectrum efficiency of 1 bps/Hz without degrading the performance of primary users in a real-world office environment. Key features of this system include:

• No across-system coordination is needed for spectrum sharing
• No inter-system synchronization is needed for spectrum sharing
• No central controller is needed for spectrum sharing

In WLANs, network-wide time and frequency synchronization among user devices is often considered essential for uplink MU-MIMO. Achieving such synchronization, however, incurs significant airtime overhead, compromising the throughput gains of MU-MIMO. In this demo, we demonstrate that time and frequency synchronization among user devices is not necessary for uplink MU-MIMO in WLANs. We developed a practical uplink MU-MIMO scheme that operates without requiring timing and clock frequency alignment among user devices. The key component of our solution is a new PHY design for the access point’s receiver, enabling it to decode concurrent signals from multiple asynchronous user devices. Experimental results indicate that uplink MU-MIMO is feasible in asynchronous wireless networks. Key features of this project include:

• No across-system coordination is needed for spectrum sharing
• No inter-system synchronization is needed for spectrum sharing
• No central controller is needed for spectrum sharing