Spectrum Sharing That Only Works in Theory

Radio spectrum — the electromagnetic frequencies used for wireless communication — is an exhaustible shared resource allocated through static licensing that leaves much spectrum underutilized while licensed bands are congested. Dynamic spectrum sharing (DSS) promises to solve this by allowing multiple users to access the same frequencies in real time based on actual usage rather than rigid licenses. However, deployed DSS systems (CBRS/3.5 GHz in the US, TV white spaces) only work under centralized coordination with cooperative users who voluntarily share information about their transmissions. When users are non-cooperative (competitive carriers, military/commercial coexistence, international border zones), or when the number of autonomous devices exceeds what centralized databases can manage (billions of IoT devices), existing DSS approaches fail — causing harmful interference, unfair access, or extreme latency in spectrum allocation decisions.

The 2023 National Spectrum Strategy identifies dynamic sharing as essential for meeting growing wireless demand without displacing critical federal (military, weather satellite, radar) spectrum users. The estimated economic value of spectrum access exceeds $500 billion annually in the US. The explosion of IoT devices (projected 30+ billion by 2030), autonomous vehicles requiring ultra-reliable low-latency communication, and 6G's use of higher frequencies all demand spectrum access models that scale beyond current centralized approaches. Without scalable DSS, the US faces a spectrum crisis where critical services compete destructively for limited frequencies.

The CBRS (Citizens Broadband Radio Service) three-tier sharing system uses a centralized Spectrum Access System (SAS) and Environmental Sensing Capability (ESC) to manage access around incumbent Navy radar. This works for the limited CBRS deployment but requires each user to register, report, and comply with centralized decisions — a model that breaks down with millions of autonomous, low-cost IoT devices that lack the intelligence or connectivity for real-time coordination. Cognitive radio approaches using spectrum sensing allow devices to detect and avoid occupied channels, but sensing is unreliable (hidden terminal problem) and slow relative to the millisecond dynamics of modern wireless systems. Game-theoretic models assume rational, self-interested players with knowledge of other players' strategies — assumptions that fail in heterogeneous networks where devices range from simple sensors to sophisticated base stations. Reinforcement learning approaches to spectrum access show promise in simulation but suffer from exploration-exploitation tradeoffs in live spectrum: exploration (trying different channels) causes real interference to other users.

Distributed spectrum access protocols that guarantee coexistence properties (bounded interference, minimum throughput) without requiring inter-user communication or centralized coordination — analogous to how CSMA/CD enables Ethernet without central scheduling. Physical-layer techniques that make transmissions inherently interference-tolerant (ultra-wideband, spread spectrum with modern coding) could reduce the cost of imperfect sharing decisions. Federated learning approaches where devices learn spectrum models locally and share only model updates (not raw sensing data) could enable coordination without privacy or bandwidth costs. Formal verification of spectrum sharing protocols could provide guaranteed performance bounds that regulators need to approve non-cooperative sharing.

A student team could implement and benchmark three distributed spectrum access algorithms (game-theoretic, multi-armed bandit, deep RL) on a software-defined radio testbed (USRP) with 4-8 non-cooperative nodes, measuring throughput, fairness, and convergence time under realistic interference conditions. Alternatively, a team could develop a simulation framework modeling the transition from centralized CBRS-style sharing to fully distributed access as user density increases, identifying the breaking points of centralized approaches. Relevant disciplines include electrical engineering, computer science, game theory, and machine learning.