Clustering Analysis for Internet of Spectrum Devices: Real-World Data Analytics and Applications

Internet of Spectrum Devices (IoSD) has been proposed as a bridging network among various spectrum-monitoring devices and massive spectrum-utilizing devices to enable a highly efficient spectrum sharing and management paradigm for future wireless networks. Spectrum data analytics is one of the key enabling techniques in IoSD. Correlations between spectrum state evolutions of different frequency points measured by an IoSD have been exploited to realize joint time–frequency spectrum prediction for improving the prediction accuracy. However, this kind of interrelationship has not been utilized efficiently to enhance the positive influences or avoid the negative influences when inferring the spectrum state. To fill the above gap, characteristics of spectrum state evolutions in the frequency domain are first modeled as multidimensional feature vectors in this article. Then, extensive clustering analyzes based on bisecting the $K$ -means clustering and the agglomerative hierarchical clustering are conducted on spectrum state evolutions with multidimensional feature vectors. Real-world experiments demonstrate that the proposed multidimensional features can represent the characteristics of spectrum state evolutions in a more comprehensive way. Furthermore, clustering with the proposed vectors is integrated to the joint time–frequency spectrum inference problem to form the clustering-based joint spectral–temporal-spectrum-prediction (C-JSTSP) scheme. Experiments verify that the proposed C-JSTSP scheme can improve the inference performance on both the inference accuracy and the runtime overhead.