Experimental Validation of Clustering-based Multipath Component Classification with 28 GHz Indoor Measurements

Abstract

In high-frequency wireless systems, reliable classification of multipath components (MPCs) remains challenging due to environment-dependent propagation. This paper introduces a real-time two-stage clustering method that couples a constraint-based preprocessing step with probabilistic clustering on a reduced feature set. The first stage suppresses noise and prunes features derived from delay, angle, and power-domain observations, yielding compact representations. The second stage conducts lightweight probabilistic assignment that accommodates residual uncertainty without strong prior assumptions. Experimental validation on measurement data indicates that the method further improves stability and suggests that real-time MPC classification can be achieved without sacrificing fidelity.