Demo: Intelligent Radar Detection in CBRS Band in the Colosseum Wireless Network Emulator

TL;DR

This paper demonstrates a framework using the Colosseum wireless network emulator to study spectrum sharing between radar systems and cellular networks in the CBRS band, employing machine learning for radar detection.

Contribution

It introduces a high-fidelity spectrum-sharing scenario with an ML-based radar detection agent deployed in a complex environment using the Colosseum emulator.

Findings

Achieved 88% radar detection accuracy

Detection time averaged 137 milliseconds

Validated spectrum sharing feasibility in CBRS band

Abstract

The ever-growing number of wireless communication devices and technologies demands spectrum-sharing techniques. Effective coexistence management is crucial to avoid harmful interference, especially with critical systems like nautical and aerial radars in which incumbent radios operate mission-critical communication links. In this demo, we showcase a framework that leverages Colosseum, the world's largest wireless network emulator with hardware-in-the-loop, as a playground to study commercial radar waveforms coexisting with a cellular network in CBRS band in complex environments. We create an ad-hoc high-fidelity spectrum-sharing scenario for this purpose. We deploy a cellular network to collect IQ samples with the aim of training an ML agent that runs at the base station. The agent has the goal of detecting incumbent radar transmissions and vacating the cellular bandwidth to avoid interfering with the radar operations. Our experiment results show an average detection accuracy of 88%, with an average detection time of 137 ms.