A Two-Stage CNN Based Channel Estimation for OFDM System
Annapurna Pradhan, Susmita Das, Deepak Dayalan
Abstract
The knowledge regarding the wireless channel parameters is essential for ensuring reliable end-to-end communication. Therefore, correct channel estimation becomes essential to recover the transmitted message efficiently. In this paper, we have proposed a Convolutional Neural Network (CNN) based architecture to estimate the channel using the knowledge regarding pilot locations in the time and frequency domain. We have considered the time-frequency response of the pilot positions as an image and used deep learning (DL) based image processing techniques to recover the complete information for correct channel estimation. The quality of the image is enhanced by a CNN based super-resolution network. Then the output of the first CNN is passed through an image restoration network to have high resolution image in a pipe-lined architecture. Moreover, the performance of the proposed DL based channel estimation has been evaluated in terms of mean square error. It can be observed from the simulation results that the proposed two-stage CNN based approach outperforms other baseline channel estimation methods.