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

Abstract

Uniform Manifold Approximation and Projection (UMAP) is applied to reduce the multipath dataset into 2-dimensions (2D) for visualization and clustering. Density-based spatial clustering of applications with noise (DBSCAN) is used as the clustering approach and the performance of different search radius epsilon ?. The proposed approach was used to cluster semi-urban scenarios of the COST2100 channel model (C2CM), which has many multipath components (MPCs). The approach is validated by comparing the clustering results to the ground truth and computing the Adjusted Rand Index (ARI) and the cluster-wise Jaccard index . The results suggest that lowering the search radius up to 0.3 achieved a median below 0.6 in the multiple-links scenarios due to the overlapping nature of clusters. Nevertheless, the median values above 0.7 and 0.8 for the ARI and Jaccard index , respectively for the single-link scenarios indicate the robsutness of the approach. ABSTRAK: Anggaran Manifold Seragam dan Unjuran (UMAP) 2-dimensi (2D) digunakan sebagai penggambaran dan pengelasan bagi mengurangkan set data pelbagai laluan. Aplikasi pengelasan ruangan bersama bunyi berdasarkan ketumpatan (DBSCAN) ini mengguna pakai pendekatan pengelasan dan prestasi pelbagai radius carian epsilon ?. Pendekatan yang dicadangkan ini digunakan bagi pengelasan senario separa-bandar model saluran COST2100 (C2CM), di mana komponen ini mempunyai banyak laluan (MPCs). Pendekatan ini disahkan dengan membandingkan dapatan pengelasan kepada kesahihan lapangan, pengiraan Indeks Rawak Terlaras (ARI) dan indeks Jaccard pengelasan ?. Dapatan menunjukkan pengurangan radius carian sehingga 0.3 dicapai pada median di bawah 0.6 dalam senario pelbagai pautan disebabkan oleh sifat pertindihan pengelasan. Walau bagaimanapun, nilai median di atas 0.7 dan 0.8 untuk ARI dan indeks Jaccard ?, masing-masing menunjukkan kaedah ini berkesan bagi senario pautan-tunggal.