2017–2021 · Co-author / RL & systems
CBRS & Dynamic Spectrum Sharing
Reinforcement-learning-aided spectrum sharing in the CBRS band. Use-or-share frameworks, channel selection for unlicensed LTE, and the measurement scaffolding behind them.
- CBRS
- Reinforcement Learning
- Unlicensed LTE
- Spectrum Access
The 3.5 GHz CBRS band introduced a three-tier spectrum-sharing regime (incumbents, PAL licensees, GAA users) and immediately raised practical questions: how do PAL holders and GAA users actually coexist without throwing away spectrum? How do unlicensed LTE eNBs pick channels well when they share the band with Wi-Fi and other LTE operators?
This line of work, done across my Samsung internship years and into early full-time, used reinforcement learning to attack both questions. The “use-or-share” policy work showed RL-aided listen-before-talk can increase spatial reuse at lower-tier nodes while keeping interference at higher tiers bounded.
Selected publications
- Enabling a “Use-or-Share” Framework for PAL–GAA Sharing in CBRS Networks via Reinforcement Learning — IEEE Transactions on Cognitive Communications and Networking (2019). Journal paper formalizing the RL-aided LBT policy and showing the resulting spatial-reuse / interference trade-off.
- Opportunistic Channel Access Using Reinforcement Learning in Tiered CBRS Networks — IEEE DySPAN 2018. The original RL-for-CBRS paper that the journal work extended.
- Machine Learning Enhanced Channel Selection for Unlicensed LTE — IEEE DySPAN 2019. Channel selection for unlicensed LTE eNBs that takes into account Wi-Fi APs and other LTE operators by sharing channel-utilization statistics.
- Method and Apparatus for Improving Coexistence Performance by Measurements in Wireless Communication Systems — U.S. Patent 11,032,717 (2021). Granted patent on the measurement scaffolding behind coexistence policies.