What does this research mean for the field?

Training classification models on the PFS-REM dataset using five key features reduces execution time by over 50% while maintaining 99.7% accuracy in distinguishing between line-of-sight, non-line-of-sight, and multipath UWB signal conditions. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The research aims to develop a dataset that effectively classifies signal conditions to enhance indoor positioning accuracy using UWB signals.

April 5, 2026International journal of electrical and computer engineering systems0 citationsOpen Access

PFS-Rem

GHGita Indah HapsarTelkom University RMRendy MunadiTelkom University BEBayu ErfiantoTelkom University

Key Points

The research aims to develop a dataset that effectively classifies signal conditions to enhance indoor positioning accuracy using UWB signals.
Developed the PFS-REM dataset to classify LOS, NLOS, and multipath conditions.
Applied PFS to select key features: RXPower, FPPower, CIR, SNR, and Ranging.
Utilized REM to achieve a 93.3% reduction in ranging errors.
Conducted measurements in controlled environments with various antenna orientations.
Achieved 99.7% accuracy in classification models, notably using Random Forest.
Demonstrated over 50% reduction in execution time compared to models using 12 features.
Highlighted distinct signal characteristics for LOS, NLOS, and multipath scenarios.

Abstract

Ultra-wideband (UWB) signals face challenges owing to complex obstacles in non-line-of-sight (NLOS) and multipath propagation. Distinguishing between NLOS and multipath signal propagations is crucial for differentiating the positioning error corrections. The application of edge and real-time computing presents a challenge in producing a model that is both simple and highly accurate. This study presents PFS-REM, a compact multi-class dataset designed to classify LOS, NLOS, and multipath conditions while reducing model complexity. The dataset was developed using PFS to select five key features: RXPower, FPPower, CIR, SNR, and Ranging from a secondary dataset, followed by REM to minimize ranging errors by 93.3% using an ESP32 UWB Pro module with a DW1000 chip. Measurements were conducted in controlled environments with varied antenna orientations (direct, perpendicular, and opposite) to simulate real-world conditions. The experimental results show that classification models, notably RF (with 99.7% accuracy), trained on PFS-REM achieve an over 50% reduction in execution time compared to models using 12 features, while maintaining high performance. Feature analysis revealed distinct signal characteristics across line-of-sight (LOS), non-line-of-sight (NLOS), and multipath scenarios, enhancing the dataset reliability. This approach supports the development of lightweight models suitable for edge computing and real- time indoor positioning applications, addressing the trade-off between accuracy and computational efficiency in complex environments.

AI에게 질문

Bookmark

View Full Paper