A Hybrid CNN-BiLSTM Channel Estimation Framework for DVB-T2 Systems Under Time-Varying Fading Conditions

Abstract

Digital Video Broadcasting – Second Generation Terrestrial (DVB-T2) remains a cornerstone of terrestrial broadcast television, yet its performance degrades significantly in high-mobility scenarios due to rapid channel variations that violate the quasi-static assumption underlying conventional pilot-based estimation. This manuscript presents a novel hybrid deep learning framework combining convolutional neural networks (CNNs) and bidirectional long short-term memory (BiLSTM) networks for robust channel estimation in DVB-T2 systems operating under time-varying frequency-selective fading. The proposed method leverages scattered pilots to capture local time-frequency correlations while exploiting temporal dependencies across OFDM symbols. Simulation results demonstrate that the proposed approach achieves a 4.2 dB improvement in bit-error-rate performance at a signal-to-noise ratio of 20 dB compared to conventional least-squares and linear minimum mean-square error estimators under vehicular speeds of 120 km/h. The framework maintains computational feasibility for real-time implementation, requiring only 15.6 ms per frame on standard hardware. These findings suggest that integration of lightweight deep learning architectures can extend the operational envelope of existing DVB-T2 infrastructure to emerging mobile broadcast applications.