Line-of-Sight Detection for 5G Wireless Channels
Jayawardana, Palihawadana Arachchige Dinu Nirmal (2023)
Jayawardana, Palihawadana Arachchige Dinu Nirmal
2023
Master's Programme in Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2023-04-26
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202304143718
https://urn.fi/URN:NBN:fi:tuni-202304143718
Tiivistelmä
With the rapid deployment of 5G wireless networks across the globe, precise positioning has become essential for many vertical industries reliant on 5G. The predominantly non-line-of-sight (NLOS) propagation instigated by the obstacles in the surrounding environment, especially in metro city areas, has made it particularly difficult to achieve high estimation accuracy for positioning algorithms that necessitate direct line-of-sight (LOS) transmission. In this scenario, correctly identifying the line-of-sight condition has become extremely crucial in precise positioning algorithms based on 5G. Even though numerous scientific studies have been conducted on LOS identification in the existing literature, most of these research works are based on either ultra-wideband or Wi-Fi networks. Therefore, this thesis focuses on this hitherto less investigated area of line-of-sight detection for 5G wireless channels.
This thesis examines the feasibility of LOS detection using three widely used channel models, the Tapped Delay Line (TDL), the Clustered Delay Line (CDL), and the Winner II channel models. The 5G-based simulation environment was constructed with standard parameters based on 3GPP specifications using MATLAB computational platform for the research. LOS and NLOS channels were defined to transmit random signal samples for each channel model where the received signal was subjected to Additive White Gaussian Noise (AWGN), imitating the authentic propagation environment. Variable channel conditions were simulated by randomly alternating the signal-to-noise ratio (SNR) of the received signal.
The research mainly focuses on machine learning (ML) based LOS classification. Additionally, the threshold-based hypothesis was also deployed for the same scenarios as a benchmark. The main objectives of the thesis were to find the statistical features or the combination of statistical features of the channel impulse response (CIR) of the received signal, which provide the best results and to identify the most effective machine learning method for LOS/NLOS classification. Furthermore, the results were verified through actual measurement samples obtained during the NewSense project.
The results indicate that the time-correlation feature of the channel impulse response used in isolation would be effective in LOS identification for 5G wireless channels. Additional derived features of the CIR do not significantly increase the classification accuracy. Positioning Reference Signals (PRS) were found to be more appropriate than Sounding Reference Signals (SRS) for LOS/NLOS classification. The study reinforced the significance of selecting the most suitable machine learning algorithm and kernel function as relevant for the task of obtaining the best results. The medium Gaussian support vector machines ML algorithm provided the overall highest precision in LOS classification for simulated data with up to 98% accuracy for the Winner II channel model with PRS. The machine learning algorithms proved to be considerably more effective than conventional threshold-based detection for both simulated and real measurement data. Additionally, the Winner II model with the richest features presented the best results compared with CDL and TDL channel models.
This thesis examines the feasibility of LOS detection using three widely used channel models, the Tapped Delay Line (TDL), the Clustered Delay Line (CDL), and the Winner II channel models. The 5G-based simulation environment was constructed with standard parameters based on 3GPP specifications using MATLAB computational platform for the research. LOS and NLOS channels were defined to transmit random signal samples for each channel model where the received signal was subjected to Additive White Gaussian Noise (AWGN), imitating the authentic propagation environment. Variable channel conditions were simulated by randomly alternating the signal-to-noise ratio (SNR) of the received signal.
The research mainly focuses on machine learning (ML) based LOS classification. Additionally, the threshold-based hypothesis was also deployed for the same scenarios as a benchmark. The main objectives of the thesis were to find the statistical features or the combination of statistical features of the channel impulse response (CIR) of the received signal, which provide the best results and to identify the most effective machine learning method for LOS/NLOS classification. Furthermore, the results were verified through actual measurement samples obtained during the NewSense project.
The results indicate that the time-correlation feature of the channel impulse response used in isolation would be effective in LOS identification for 5G wireless channels. Additional derived features of the CIR do not significantly increase the classification accuracy. Positioning Reference Signals (PRS) were found to be more appropriate than Sounding Reference Signals (SRS) for LOS/NLOS classification. The study reinforced the significance of selecting the most suitable machine learning algorithm and kernel function as relevant for the task of obtaining the best results. The medium Gaussian support vector machines ML algorithm provided the overall highest precision in LOS classification for simulated data with up to 98% accuracy for the Winner II channel model with PRS. The machine learning algorithms proved to be considerably more effective than conventional threshold-based detection for both simulated and real measurement data. Additionally, the Winner II model with the richest features presented the best results compared with CDL and TDL channel models.