Indoor positioning with multi-domain CSI-based deep attention networks for MIMO wireless systems.

Area of Science:

• Wireless Communications and Signal Processing
• Artificial Intelligence in Engineering
• Robotics and Autonomous Systems

Background:

• Accurate indoor positioning is crucial for emerging technologies like augmented reality and autonomous robotics.
• Existing Channel State Information (CSI)-based methods show promise but require enhancement for complex indoor environments.
• Massive Multiple Input Multiple Output (mMIMO) systems offer potential for improved positioning accuracy and robustness.

Purpose of the Study:

• To develop and evaluate an AI-driven indoor positioning method utilizing CSI for mMIMO systems.
• To investigate the effectiveness of integrating uni-domain and multi-domain CSI features.
• To introduce a Deep Attention Network (DAN) for enhanced User Equipment (UE) positioning.

Main Methods:

• Extraction of channel impulse, channel frequency, and angular response domain features from CSI data.
• Development of a Deep Attention Network (DAN) to process and integrate multi-domain CSI features.
• Evaluation of DAN against baseline and multi-domain Convolutional Neural Network (CNN) models using a public mMIMO dataset.

Main Results:

• The multi-domain Deep Attention Network (DAN) demonstrated superior positioning accuracy compared to CNN approaches.
• DAN effectively leverages attention mechanisms for integrating diverse CSI feature domains.
• A trade-off exists between enhanced positioning performance and increased inference complexity with DAN.

Conclusions:

• Attention mechanisms and multi-domain CSI features hold significant potential for advancing indoor UE positioning systems.
• The proposed AI-driven DAN method offers a robust solution for accurate indoor localization in mMIMO environments.
• Further research may focus on optimizing DAN for reduced computational overhead while maintaining high accuracy.