Accurate channel estimation remains a fundamental challenge for next-generation orthogonal frequency-division multiplexing (OFDM) systems, especially in environments with high mobility, sparse pilot allocation, and time-varying multipath fading. Traditional pilot-based methods, such as Least Squares (LS) and Minimum Mean Square Error (MMSE), are easy to implement but suffer from noise sensitivity, high computational costs, and a reliance on prior channel statistics. Recently, deep learning techniques have shown promising results; yet many do not fully capture the joint spatial and temporal characteristics of wireless channels or overlook realistic pilot-grid structures. This study introduces a pilot-grid-aware hybrid deep learning model that integrates Convolutional Neural Networks (CNNs) for spatial feature extraction with Long Short-Term Memory (LSTM) networks for temporal sequence modeling. The model is trained using the Root Mean Square Propagation (RMSprop) optimizer, selected for its robustness in adapting learning rates and effectively capturing dependencies across both spatial and temporal dimensions. Benchmark evaluations against conventional estimators and theoretical limits under different modulation formats demonstrate that the proposed model consistently achieves a lower estimation error and bit error rate across a wide range of signal-to-noise ratios. The results confirm that the hybrid architecture provides a scalable and reliable solution for future wireless systems including sixth-generation (6G) networks, vehicular communications, and satellite applications where both adaptability and robustness are essential.
Massalay et al. (Mon,) studied this question.