Deep Learning-based Channel State Estimation for V2V OFDM Communication: A Comparative Study of LSTM, BiLSTM, and GRU Networks

CSE is crucial for OFDM systems to handle multipath fading in wireless channels. While CS techniques like SOMP are computationally efficient, their performance is limited by basis mismatch and noise sensitivity. This paper presents a comprehensive comparison between SOMP and DL approaches using LSTM, BiLSTM, and GRU networks for CSE in V2V communication. The performance of the proposed DL models is rigorously evaluated in a realistic V2V communication scenario utilizing the 3GPP standard vehicular channel model within an OFDM system, with estimation accuracy assessed based on MSE. Experimental results demonstrate that the DL architectures significantly outperform SOMP, achieving a reduction in MSE by up to 15 dB and a reduction in BER by up to three orders of magnitude at high SNRs while maintaining robust performance in high-mobility highway environments. The study establishes DL, particularly the efficient GRU model, as a superior paradigm for accurate and adaptive channel estimation in modern wireless communication systems, thereby contributing to safer and more reliable V2V communication essential for next-generation intelligent transportation systems. The proposed models are trained to accurately estimate the CSI, which is subsequently utilized for the final detection of the transmitted data.