High-Frequency Skywave Source Geolocation Using Deep Learning-Based TDOA Estimation and Bias-Regularized Semidefinite...

Area of Science:

• Radio Science
• Signal Processing
• Machine Learning

Background:

• High-frequency (HF) skywave propagation enables beyond-line-of-sight communication via ionospheric reflection.
• Time-difference-of-arrival (TDOA) based geolocation is crucial for localizing non-cooperative HF emitters.
• Ionospheric variability, multipath dispersion, and low SNR challenge reliable TDOA estimation.

Purpose of the Study:

• To develop an integrated framework for robust TDOA estimation and geolocation of HF emitters.
• To improve the accuracy and reliability of HF emitter localization under challenging propagation conditions.

Main Methods:

• Constructed an improved wideband ionospheric channel model integrating the International Reference Ionosphere (IRI) with region-specific calibration and stochastic perturbations.
• Designed a convolutional neural network (CNN) for TDOA estimation, utilizing I/Q waveforms, GCC feature maps, and channel-state information (CSI).
• Formulated geolocation as a bias-regularized constrained least-squares problem, solved using semidefinite programming (SDP) relaxation.

Main Results:

• The proposed CNN-based TDOA estimator demonstrated robust performance under severe noise and multipath conditions.
• Simulations showed the estimator consistently outperformed competing algorithms and approached the Cramér-Rao lower bound (CRLB) at high SNR.
• Field-recorded data showed a 51% reduction in mean absolute TDOA deviation compared to GCC-PHAT.
• The end-to-end pipeline achieved a mean geolocation error of 19.67 km.

Conclusions:

• The integrated framework offers a significant advancement in HF emitter geolocation.
• The deep learning approach provides robust TDOA estimation in challenging ionospheric environments.
• The developed method enhances the accuracy and reliability of beyond-line-of-sight localization systems.