Data Clustering Using Matrix Factorization Techniques for Wireless Propagation Map Reconstruction

This paper develops an efficient data clustering technique by transforming and compressing the measurement data to a low-dimensional feature matrix, based on which, matrix factorization techniques can be applied to extract the key parameters for data clustering. For the application of wireless propagation map reconstruction, a theoretical result is developed to justify that the feature matrix is a composite of several unimodal matrices, each containing key parameters for an individual propagation region. As a result, instead of iterating with $N$ data points at each step, the proposed scheme provides a low complexity online solution for data clustering based on the feature matrix with dimension much smaller than $N.$