VOLUME 72 : 2024

Channel modeling can be used to evaluate the performance of wireless communications systems. The
European Cooperation in Science and Technology (COST) 2100 channel model (C2CM) can reproduce the
multiple-input multiple-output (MIMO) channels’ stochastic properties over time, frequency, and space.
Multipath components with similar properties in delay and angles form multipath clusters. Multipaths
have been clustered by shallow (non-deep learning) approaches over the years. The rise of deep learning
approaches makes them good candidates in multipath clustering, but studies in this area remain rare.
Thus, this study investigates the performance of Deep Divergence-Based Clustering (DDC) in grouping
the multipaths from the COST 2100 dataset and measuring the performance with fourteen well-known
shallow approaches. Ten different validation metrics evaluate the clustering results. DDC has the highest
scores in ACC (0.3935), AMI (0.5346), and FMI (0.3102) in the semi-urban scenarios. Results indicate
that the performance of DDC is close to the shallow clustering approaches. Thus, DDC can be used in
clustering multipaths.