Optimal Crowd-Augmented Spectrum Mapping via an Iterative Bayesian Decision Framework

Recent spectrum measurements show that geo-location spectrum databases are inaccurate in Metropolitan areas by missing a large number of TV white spaces (TVWS). To counter this challenge, this paper introduces a crowd-augmented spectrum database, whose spatial resolution and accuracy are continuously and opportunistically augmented by the spectrum sensing data from the crowd of mobile users. To augment the accuracy of our database to a desired level, an iterative Bayesian decision framework is proposed, which coherently combines two major modules over multiple rounds, including (1) Bayesian spatial prediction for optimal prediction of the power spectrum density (PSD) values at different spatial locations under parameter uncertainty and (2) Bayesian experimental design for efficient selection of the locations with high utility for additional sampling. The proposed framework is implemented and verified within our university main campus. Abundant spectrum opportunities are discovered, compared with the traditional geo-location databases, e.g., Google spectrum database.