Channel Estimation for AFDM System Based on Deep Residual Network

Abstract

Affine Frequency Division Multiplexing (AFDM) demonstrates significant advantages in high-mobility doubly dispersive channels. Conventional pilot-assisted channel estimation schemes necessitate high pilot power, resulting in a high pilotto-data power ratio (PDR). Reducing the pilot power makes weak multipath components difficult to be detected due to noise and data interference, thereby degrading channel estimation accuracy. To address this issue, this paper proposes an iterative interference cancellation-based AFDM channel estimation scheme employing a deep residual network (ResNet). The proposed scheme utilizes the ResNet to detect multipath components under severe interference and iteratively cancels residual interference to enhance channel estimation accuracy. Simulation results demonstrate that the proposed scheme improves channel estimation accuracy under a lower PDR than existing schemes and achieves a tradeoff between system power overhead and reliability.