Generative Adversarial Network-Based Joint Mapping and Localization for Millimeter-Wave Communication Systems.

Area of Science:

• Wireless Communication
• Artificial Intelligence
• Robotics and Navigation

Background:

• Millimeter-wave (mmWave) systems offer high bandwidth but require precise localization and mapping.
• Existing joint localization and mapping (JLAM) methods face challenges in accuracy and efficiency.
• Angle difference of arrival (ADOA) measurements provide rich spatial information for localization.

Purpose of the Study:

• To propose a novel generative adversarial network (GAN)-based JLAM method utilizing ADOA measurements for mmWave systems.
• To enhance the accuracy of mobile terminal (MT) localization and indoor map estimation.
• To improve upon existing JLAM algorithms in terms of localization error and room boundary estimation.

Main Methods:

• Developed a GAN-based JLAM method with a deep auto-encoder discriminator.
• Modeled the generator as an explicit geometric ADOA function, not a black-box neural network.
• Exploited 2D distribution of high-dimensional ADOA vectors to learn data distribution and AP topology.

Main Results:

• Achieved an average localization error of approximately 0.25 m under representative simulation conditions.
• Demonstrated a significant error reduction of about 58% compared to the JADE algorithm (0.60 m error).
• Showcased more accurate room boundary estimation compared to the JADE algorithm.

Conclusions:

• The proposed GAN-based JLAM method is effective for mmWave communication systems.
• The method accurately estimates MT positions and indoor maps by recovering AP topology.
• This novel approach offers superior performance in localization accuracy and map detail over existing methods.