Enhancing Channel Estimation in Terrestrial Broadcast Communications Using Machine Learning

Abstract

Artificial Intelligence (AI) and Machine Learning (ML) approaches have emerged as viable alternatives to conventional Physical Layer (PHY) signal processing methods. Specifically, in any wireless point-to-multipoint communication, accurate channel estimation plays a pivotal role in exploiting spectrum efficiency with functionalities such as higher-order modulation or full-duplex communication. This research paper proposes leveraging ML solutions, including Convolutional Neural Networks (CNNs) and Multilayer Perceptrons (MLPs), to enhance channel estimation within broadcast environments. Each architecture is instantiated using distinct procedures, focusing on two fundamental approaches: channel estimation denoising and ML-assisted pilot interpolation. Rigorous evaluations are conducted across diverse configurations and conditions, spanning rural areas and co-channel interference scenarios. The results demonstrate that MLP and CNN architectures consistently outperform classical methods, yielding 10 and 20 dB performance improvements, respectively. These results underscore the efficacy of ML-driven approaches in advancing channel estimation capabilities for broadcast communication systems.