A DNN-Based OFDM Channel Estimation Algorithm Without Training Overheads

In this paper, we propose a channel estimation algorithm for OFDM systems based on a deep neural network to reduce overheads in model training. In this method, the channel estimation problem is formulated as an image repair problem, where a channel matrix containing pilot values is regarded as an incomplete picture, and then a specially designed deep neural network based on the deep image prior (DIP) is exploited to reconstruct complete and noise-removed channel images from the incomplete picture. While reducing complexity and training overheads, the method also ensures estimation accuracy. Simulation results show the superior performance and effectiveness of the proposed channel estimation algorithm.