MT-GCNN: Multi-Task Learning with Gated Convolution for Multiple Transmitters Localization in Urban Scenarios.

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Accurate localization is crucial for Internet of Things (IoT) services.
• Urban environments present challenges for multiple transmitter localization, including non-line-of-sight (NLOS) propagation and sparse sensor deployment.

Purpose of the Study:

• To propose a novel deep multi-task learning scheme, the Multi-Task Gated Convolutional Neural Network (MT-GCNN), for multiple transmitters localization.
• To effectively learn NLOS propagation features and improve localization accuracy in challenging urban scenarios.

Main Methods:

• The MT-GCNN framework decomposes the localization problem into coarse and fine correction tasks using deep multi-task learning.
• An improved gated convolution module is employed to extract features from sparse sensing data.
• A joint loss function is utilized for optimizing the multi-task network during training.

Main Results:

• The MT-GCNN model jointly predicts classified grids and biases for enhanced localization performance.
• Numerical simulations demonstrate superior accuracy and robustness compared to existing algorithms in urban NLOS conditions.

Conclusions:

• The proposed MT-GCNN framework offers a significant advancement in multiple transmitter localization, particularly in complex urban environments.
• Deep multi-task learning effectively addresses NLOS propagation and sensor sparsity challenges, leading to more reliable localization solutions.