Channel estimation for OFDM Systems based on RMSGAN

In this paper, a residual multi-scale generative adversarial network (RMSGAN) is introduced for orthogonal frequency division multiplexing (OFDM) systems, which conducts feature extraction via multi-scale feature fusion and local residual connections, rendering it suitable for OFDM channel estimation. In addition, the proposed channel estimation method achieves generalization for different pilot insertion methods through custom deconvolution layers. Simulation experiments reveal that in multipath fading channels, the RMSGAN-based channel estimation algorithm outperforms several existing image super-resolution-based channel estimation algorithms with regard to the mean square error (MSE) evaluation metric. Furthermore, it achieves comparable performance to the linear minimum mean square error (LMMSE) algorithm.