An Efficient Convolutional Denoising Autoencoder-Based BDS NLOS Detection Method in Urban Forest Environments.

Area of Science:

• Satellite Navigation Systems
• Deep Learning Applications
• Signal Processing

Background:

• BeiDou Navigation Satellite System (BDS) is crucial for autonomous driving, but urban environments cause positioning errors due to non-line-of-sight (NLOS) signals.
• Existing deep learning (DL) methods for NLOS detection face challenges with large labeled datasets and noise sensitivity, limiting accuracy and generalization.
• The need for robust and accurate NLOS detection is critical for reliable autonomous vehicle navigation in complex settings.

Purpose of the Study:

• To propose a novel deep neural architecture, CDAENet, for effective NLOS signal detection in urban forest environments.
• To address the limitations of supervised learning and improve the robustness of DL models against noisy data.
• To enhance the positioning accuracy of the BeiDou Navigation Satellite System (BDS) in challenging urban settings.

Main Methods:

• Developed a convolutional denoising autoencoder network (CDAENet) utilizing unsupervised deep learning for signal denoising and feature extraction.
• Implemented a denoising autoencoder to reduce time-series signal dimensionality and enhance robustness against input noise.
• Integrated a Multi-Layer Perceptron (MLP) algorithm for identifying non-linear characteristics within BDS signals.

Main Results:

• The CDAENet model achieved over 95% satellite detection accuracy on a real urban forest dataset.
• Demonstrated an approximate 8% improvement in accuracy compared to existing machine learning methods.
• Showcased approximately a 3% accuracy enhancement over current deep learning-based approaches for NLOS detection.

Conclusions:

• The proposed CDAENet model effectively detects NLOS signals in complex urban environments, significantly improving BDS positioning accuracy.
• Unsupervised learning and denoising autoencoders enhance model robustness and reduce reliance on large labeled datasets.
• CDAENet offers a promising solution for reliable autonomous driving by mitigating positioning errors caused by NLOS signals.