Distributed Consensus for Coherent Spectrum Sensing

In this Phase I effort, Tiami, LLC, aims to develop and demonstrate a hardware proof of concept for a completely distributed spectrum sensing scheme that leverages consensus learning amongst radio frequency (RF) sensors. The algorithm is based on low-bandwidth message exchange between one-hop neighbors, spans multiple RF bands, is agnostic to the sensing modality, and is resilient to link disruptions. Wireless multi-hop networks are widely used by the Navy for tactical data links and exchange of track measurements or engage-on-remote data. Spectrum domain awareness is crucial for link reliability and anti-jamming measures, however, the distributed nature of multi-hop networks and the need for robustness rules out a centralized spectrum sensing approach. Distributed spectrum sensing is a natural solution, but maintaining coherency across non-colocated nodes with intermittent links is a major technical challenge. Our proposed multi-agent consensus learning scheme for distributed spectrum sensing works as follows. M sensors or agents attempt to collaboratively learn the spectrum occupancy state of N frequency bands. Each sensor exchanges its local sensing likelihood per band with its one-hop neighbors in an iterative process, and updates its local statistic based on a weighted combination of its neighbor observations. Weights are adjusted based on network topology and outlier detection. Convergence is achieved when connected nodes arrive at a consensus.