Low Complexity Deep Learning Aided Channel Estimation Architecture for Vehicular Networks

The channel estimation (CE) in ultra-reliable vehicular communication is challenging due to dynamically varying channel conditions. In this direction, various data pilot-aided (DPA) and deep neural network (DNN) approaches have been proposed to improve the CE obtained using the known preamble transmitted at the beginning of the data frame. In this work, we propose an efficient architecture for the time domain reliable test frequency domain interpolation (TRFI) CE scheme by exploiting the tri-diagonal matrix algorithm (TDMA) approach and demonstrate 99.4% reduction in latency on Zynq system on chip (ZSoC). Using TRFI-TDMA, we develop two efficient architectures of TRFI-DNN to achieve the desired trade-off between complexity and latency. We demonstrate the superiority of the proposed architectures in terms of bit error rate for different channels, signal-to-noise ratios, and word length.