Research On V2V Channel Estimation Based On Deep Learning And Channel Super-Resolution Reconstruction

Vehicle-to-vehicle（V2V）communication scenarios have the characteristics of low vehicle antenna height and wide distribution of road scatterers,making the channel propagation environment complex and changeable.Meanwhile,there are also serious Doppler shifts,multipath fading,and path loss phenomena.However,current channel estimation methods cannot completely eliminate the influence of complex statistical characteristics in V2V communication scenarios,and it is difficult to accurately track the channel.To this end,this paper introduces the idea of image and video super-resolution reconstruction into channel estimation,combines deep learning to build channel super-resolution networks to estimate V2V channel response.Simulation verification was carried out under IEEE 802.11p.The main work is as follows:First,the pilot is extended to channel data with time-frequency correlation,and a low-resolution network input construction method based on averaging decision feedback and time-domain truncation is proposed.The interpolation strategy of averaging time-domain truncation is used for the data pilot,which is expanded into the channel dimension,and the channel parameters are updated by comparing the correlation between related channels.The simulation results show that the proposed low-resolution network input construction method can expand the pilot to channel data with time-frequency information.Secondly,on the basis of constructing a low-resolution network input,the channel response is represented as a two-dimensional matrix in the complex domain,and the channel response is reconstructed by combining deep learning and image super-resolution algorithms.Using the residual structure and convolutional neural network,a residual channel network is proposed to restore the high-resolution channel response.The simulation results show that the algorithm can reduce the bit error rate of channel estimation to 5×10^-3 when the signal-to-noise ratio is 30d B in the V2V channel scenario.Finally,on the basis of constructing low-resolution network input,in order to utilize the time-frequency correlation of the channel,the channel is divided into blocks and processed into video frames.A video super-resolution algorithm based on deep learning is proposed to reconstruct the channel response.Using the correlation between adjacent symbols,the channel is divided into multiple blocks,and the time-frequency characteristics of the channel are extracted using a network with convolutional long short-term memory as the core.The proposed temporal spectral channel network realizes V2V channel estimation.The simulation results show that the proposed algorithm can reduce the bit error rate of channel estimation to 4×10^-4 when the signal-to-noise ratio is 30d B in the V2V channel scenario.