Map-assisted TDOA Localization Enhancement Based On CNN

Yiwen Chen,Tianqi Xiang,Xi Chen,Xin Zhang

2024-01-31

Abstract:For signal processing related to localization technologies, non line of sight (NLOS) multipaths have a significant impact on the localization error level. This study proposes a localization correction method based on convolution neural network (CNN), which extracts obstacle features from maps to predict the localization errors caused by NLOS effects. A novel compensation scheme is developed and structured around the localization error in terms of distance and azimuth angle predicted by the CNN. Four prediction tasks are executed over different building distributions within the maps for typical urban scenario, resulting in CNN models with high prediction accuracy. Finally, a thorough comparison of the accuracy performance between the time difference of arrival (TDOA) localization algorithm and the results after the error compensation reveals that, generally, the CNN prediction approach demonstrates great localization error correction performance, improving TDOA accuracy by 75%. It can be observed that the powerful feature extraction capability of CNN can be exploited by processing surrounding maps to predict the localization error distribution, showing great potential for further enhancement of TDOA performance under challenging scenarios with rich multipath propagation.

Signal Processing

What problem does this paper attempt to address?

The paper mainly addresses the performance degradation issue of the Time Difference of Arrival (TDOA) method in wireless positioning technology under Non-Line-of-Sight (NLOS) conditions. Specifically, when wireless signals are affected by obstacles causing reflection, diffraction, or scattering, traditional TDOA positioning algorithms exhibit significant errors. To solve this problem, the researchers propose a positioning error compensation method based on Convolutional Neural Network (CNN).

The core idea of this method is to use CNN to extract environmental features from the map and predict how these features affect the TDOA positioning results, thereby predicting the distribution of positioning errors. In this way, a model can be pre-trained to predict positioning errors in different scenarios, and this predicted error can be used to correct the TDOA positioning results during actual positioning, thereby improving positioning accuracy.

The specific steps include:
1. **Dataset Construction**: Construct a map dataset containing different building distributions and the corresponding TDOA positioning error labels.
2. **Offline Model Training**: Use the constructed dataset to train the CNN model so that it can predict positioning errors based on the input map.
3. **Online Error Compensation**: In practical applications, first obtain the map information of the surrounding environment, then input it into the trained CNN model to get the predicted positioning error, and finally use these errors to correct the TDOA positioning results.

Experimental verification shows that this method can significantly improve the accuracy of TDOA positioning, reducing positioning errors by about 75% in some cases. This indicates that by directly extracting environmental information from maps using deep learning technology, the performance of TDOA positioning technology under NLOS conditions can be effectively enhanced.