Practical Waveform Generation and Labeling for Radio Frequency Applications

When developing and testing radio frequency (RF) signal detection applications, having high quality ground truth labels for the time and frequency coordinates and metadata of any transmission are crucial for quantifying the system’s performance. In addition to testing systems, this ability is necessary for high-quality datasets to be generated for Machine Learning (ML) applications. This demonstration showcases an over-the-air, stand-in waveform generation toolset used for testing a spectrum observer being developed for the IARPA SCISRS program. This tool is not only able to create and transmit data with software-defined radios, but also keep track of the precise time and intended frequency of transmission to the fidelity of the hardware available, for validating reporting of the spectrum observer. There are two goals of this demonstration: first, to share with the community a tool for generating high quality data for both testing and training complex RF systems; second, to solicit feedback of how this tool can further be developed to better support the Radio Frequency Machine Learning (RFML) area of study.