Density Based User Clustering for Wireless Massive Connectivity Enabling Internet of Things

This paper considers the problem to handle the expected large number of users in future wireless mobile communication systems with massive multiple input multiple output. We focus on the recently proposed joint spatial division and multiplexing scheme introducing a clustering step performed on all users before the user-selection and precoding steps as in traditional systems. From literature it is known that the so far used k-means clustering has some drawbacks, e. g. the number of clusters has to be known a-priori to achieve good performances. We overcome this problem by using the density-based clustering of applications with noise (DBSCAN) approach and show how the input parameters for this algorithm can be derived. Performance results confirm that DBSCAN outperforms k-means clustering. A second conclusion is that clustering can be beneficial in terms of sum spectral efficiency for realistic user deployments reducing also complexity by enabling independent per-user group signal processing.