AI-AIDED JOINT VISIBILITY REGION DETECTION AND CHANNEL ESTIMATION FOR EXTREMELY LARGE APERTURE ARRAYS

Author

Muhammad Zia Hameed, Southern Illinois University CarbondaleFollow

Date of Award

8-1-2024

Degree Name

Master of Science

Department

Electrical and Computer Engineering

First Advisor

Baduge, Gayan Aruma

Abstract

Massive multiple input multiple output(MIMO) technology, now commercially deployed in 5G wireless systems, enhances spectral and energy efficiency. While compact antenna arrays assume wide-sense stationary (WSS) channels, recent studies reveal that extremely large aperture arrays (ELAAs) exhibit non-WSS characteristics, leading to variability in received signal properties such as average power. This study begins by examining the unique propagation characteristics of ELAAs and the importance of visibility regions (VRs) in channel estimation and precoder designs. A comprehensive case study and analytical analysis demonstrate the effectiveness of VR-aware system design approaches, leading to significant improvements in ELAAs system performance, including minimized inter-user interference and enhanced achievable sum rates.Building on these insights, this research introduces a hybrid deep learning architecture combining one-dimensional convolutional neural networks (1D-CNN) and long-short term memory (LSTM) networks for precise VR classification. This hybrid model, using a diverse dataset generated through Monte-Carlo simulations to capture various VR combinations, shows higher accuracy in classifying VRs, which is crucial for improving channel estimation and designing VR-aware precoders. The model's efficacy is validated through rigorous numerical results, highlighting its robustness and efficiency. Subsequently, a hybrid deep learning architecture, 1D CNN-LSTM, is presented to estimate the channel for each user at the ELAA through a data-driven deep learning technique. A diverse dataset, generated through Monte-Carlo simulations of the system and channel model, captures various VR combinations to improve channel estimation.Furthermore, this study develops a multi-task learning (MTL) 1D CNN-LSTM framework to simultaneously address VR classification and channel estimation in ELAA systems. This MTL model leverages uplink pilot signals as inputs, enhancing both tasks' accuracy and performance. The proposed approach demonstrates superior performance in terms of lower mean squared error (MSE) and higher achievable rates compared to conventional methods.The results confirm the potential of deep learning models to significantly improve ELAA-based communication systems' performance, providing a robust foundation for future research. Future work should focus on developing adaptive learning techniques, integrating ELAA systems with emerging technologies, enhancing dataset generation for better model robustness, and validating these frameworks in real-world environments.