Lightweight and High-Performance Vehicular Channel Estimation With Liquid Neural Networks

Vehicular communication demands precise and efficient channel estimation for robust connectivity. While deep learning techniques, particularly long short-term memory (LSTM) networks, have recently improved channel estimation results by effectively capturing temporal dependencies, their high computational demands pose significant challenges, especially in resource-constrained environments such as Internet of Things devices. This paper introduces a novel channel estimation approach using liquid neural networks to address these limitations. It specifically focuses on closed-form continuous-depth (CfC) networks optimized through a neural architecture search process to balance performance and complexity. The proposed CfCbased estimator enhances channel tracking accuracy and reduces computational overhead compared to traditional LSTM-based methods.