Abstract—The automotive industry has developed Dedicated Short-Range Communication (DSRC) technology for specific Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) communication applications. However, the effectiveness of DSRC communication is highly dependent on other integrated standards for interoperability, which is still room for research, especially under high mobility. Deep learning has recently played a crucial role in boosting the system performance in 5G-and-beyond networks. This paper utilizes deep learning to improve the channel estimation quality of DSRC systems under time-variant and frequency-selective channels and applying a post filtering process to enhance the quality of reconstructed images. We consider an OFDM-based system where the propagation channels are roughly estimated at the receiver by a low-cost least squares method. Then, the channel estimation quality is enhanced by a data-driven approach exploiting supervised learning. Numerical results manifest the added benefits of deep learning for improving the channel estimation quality and boosting the Bit Error Ratio (BER) compared to the traditional estimation methods. Besides, a post-filter is necessary to remove artifacts and residual errors in recovered image data. Quantitatively, the support of deep learning improves the channel estimation quality by about 30%. At the same time, the post-filtering process enhances the reconstruction quality in terms of the Peak-Signal-to-Noise Ratio (PSNR) up to 4dB.
 

Index Terms—DSRC technology, OFDM, channel estimation, deep learning, image data