Boosting indoor white space identification by exploiting the substructure of unexplored data

In 2008, the Federal Communications Commission (FCC) issued a historic ruling to allow unlicensed devices to operate in the locally vacant TV channels (also called TV white spaces or simply white spaces). The FCC requires that unlicensed users must not interfere with the incumbents, thus we need to identify the white spaces. While most prior works have focused on exploring the outdoor white spaces, the indoor story is largely open for investigation. WISER [1] is the first and best known solution to identify the indoor white spaces by leveraging the power of clustering algorithms. However, WISER uses a simple agglomerative clustering algorithm, which assumes the measurement data is precise (i.e., no noise in the data) and cannot exploit the substructure of the data. To overcome these drawbacks, in this paper, we modify two powerful clustering algorithms (Fuzzy C-Means and Affinity Propagation) to boost the performance of indoor white space identification. Our experimental results based on real-world traces show that: (1) Fuzzy C-Means can reduce the false alarm rate of WISER by 50%. (2) Affinity Propagation can obtain 5% more white spaces than WISER and simplify the system design without manually setting the number of clusters.