Data-driven baseband processing and sensing for smart & efficient virtualized RAN in 5G and beyond
Data-Driven algorithms are transforming the wireless baseband. Using data-driven algorithm design, we can replace numerous algorithms in the 4G and 5G baseband with more efficient and accurate learned alternative which save cost, increase performance, make better use of power, heat, cooling, and antenna apertures, and ultimately provide a better experience for users and a better value proposition for operators. Beyond using data-driven solutions within current standards, AI/ML and data-driven design have begun to influence how future standards are envisioned in both 5G Advanced as well as within 6G candidate waveforms and technologies. In this latter case (and outside of standards settings), we are beginning to see AI-Native solutions, or waveform components which are designed and learned for and by AI/ML optimization from the ground up. This approach offers to provide significant performance benefits in many applications and will be an exciting area of fastmoving applied research over the coming years as 6G continues to take form.
In this talk, we will cover key data-driven AI/ML software solutions which will make 4G/5G vRAN infrastructure more performant as well as highlighting how powerful AI/ML based spectrum sensing and broad sensing-making capabilities are becoming as an enabler for spectral efficiency, interference-free operations, and security within both public and private RAN deployments and other applications. We will highlight some of our recent work in these areas, initiatives in industry and standards bodies looking at including these technologies and discuss our vision for how these technologies will continue to evolve and shape future wireless systems. Finally, we would like to share our data-driven wideband spectrum activity recognition challenge open-dataset competition. This competition helps to provide a new broad definition and scoring metric for spectrum activity sensing making in diverse environments as both a challenge and a benchmark task to researchers in the area to help evolve the dialogue and focus beyond more simplistic tasks such as modulation classification.