Grouping of COST 2100 Indoor Multipaths Using Simultaneous Clustering and Model Selection

Abstract

Simultaneous Clustering and Model Selection (SCAMS) is introduced to cluster multipaths from COST 2100 channel model (C2CM). SCAMS determines not only the number of clusters but also the membership of the clusters. The study is the first to report clustering of multipaths that consider simultaneously the number of clusters and the membership of the clusters. Cluster identification and cardinality classification are dependent on the values of and , the parameters that weigh the penalty terms to avoid the trivial solution (all 1 matrix) of the affinity matrix. The clustered multipaths are compared with the reference multipaths that can be found in IEEE DataPort. The accuracy of the clustering approach is examined using the Jaccard index ( ). The proposed clustering approach can achieve higher accuracy compared to popular multipath clustering approaches.

Joint Identification of the Clustering and Cardinality of Wireless Propagation Multipaths

“…Table 1 shows the Jaccard indices of the eight-channel scenarios. The results are the completion of the previous works [14], [15]. Indoor channel scenarios have better accuracy both for the number of clusters and membership of clusters due to lesser number of multipaths and multipath clusters generated by the enclosed space where reflections of signals are limited.…”

Wireless Propagation Multipath Clustering: On Simultaneously Solving the Membership and the Number of Clusters

“…Semi-urban B2 has twenty indices (in cyan) out of thirty above the mean for the number of clusters while seventeen indices (in cyan) out of thirty above the mean for the membership of the clusters. SCAMS can be used as an alternative to cluster multipaths in semi-urban scenario but it gives lower accuracy compared to the results obtained in [14] for indoor scenario. The relative frequency of the Jaccard indices for the number of clusters for both semi-urban B1 and B2 LOS single link is shown in Table 1 while that of the membership of the clusters for also both semi-urban B1 and B2 LOS single link is presented in Table 2.…”

Optimal Clustering of Wireless Multipaths by Uniform Manifold Approximation and Projection-Assisted Dbscan

“…Additionally, a comparative study of different algorithms and their performance is presented in [9]. The use of Simultaneous Clustering and Model Selection (SCAMS) is proposed in [10,11] to cluster the C2CM generated datasets. Finally, a visualization tool using 3D point cloud data to locate small interacting objects in a microcell was proposed to include the visualization process in the wireless channel characterization [12].…”