Non-line-of-sight identification from WiFi CSI using particle swarm with multi-dimensional feature fusion

Identifying line-of-sight (LOS) and non-line-of-sight (NLOS) conditions can reduce multipath propagation's negative impacts, ease communication link burdens, and improve wireless application performance. This paper proposes a lightweight, reliable and efficient NLOS recognition scheme focusing on channel state information (CSI), named PNLOS-MFC (NLOS identification based on particle swarm optimization algorithm with multi-dimensional feature fusion). We explore various optimization algorithms (PSO, GSA, GWO, ABC) to optimize SVM parameters, ultimately selecting the PSO-SVM-RBF structure with the best NLOS identification performance, accelerating channel identification. Next, to enhance the recognition rate of NLOS signals, we select the best CSI feature clusters by comparing and analyzing the recognition rate of NLOS signals in nine sub- clusters. Finally, the proposed model is compared to five different machine learning classifiers and traditional NLOS recognition algorithms, demonstrating superior performance with detection rates of 98.75% and 96.87% for NLOS in office and corridor scenarios.