Deep residual networks for wireless transmitter localization using sparse boundary measurements

Abstract

Transmitter localization is a critical challenge in wireless communication, impacting network optimization, security, and spectrum management. Traditional localization techniques, such as triangulation and fingerprinting, require extensive data collection and computational resources, making them impractical for large-scale deployments. This thesis explores a deep learning-based localization method that leverages a Residual Neural Network (ResNet) to predict the position of a wireless transmitter using sparse boundary signal strength measurements. Unlike conventional methods that rely on full-area data collection, the proposed approach significantly reduces measurement complexity while maintaining high localization accuracy. The model is trained on a dataset of simulated radio propagation maps generated using the Dominant Path Model (DPM) and evaluated on both synthetic and real-world measurement scenarios. Experimental results demonstrate an average localization error of 7.23 meters, with a standard deviation of 3.32 meters, making it competitive with state-of-the-art deep learning localization techniques. Additionally, a confidence metric is introduced to assess prediction reliability in non-line-of-sight conditions. The findings highlight the potential of deep residual learning for cost-effective transmitter localization in complex environments with limited access. This research contributes to the development of deep learning solutions for wireless signal analysis, paving the way for enhanced network management and interference mitigation strategies.

Date	5 Dec 2025
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Richard Al Hadi (Supervisor)