Channel estimation in IRS-assisted OTFS communication via residual attention network

Abstract

For intelligent reflecting surface (IRS) based communication, channel estimation methods have predominantly focused on low-mobility and static scenarios. However, in dynamic scenarios where mobility and channel variations take place, accurate channel estimation becomes a challenging task. To address this limitation, this paper proposes a novel approach for channel estimation in dynamic IRS-aided communication scenarios by leveraging the advantages of orthogonal time-frequency space (OTFS) modulation. The proposed approach converts the time-frequency domain channel representation into the delay-Doppler (DD) domain using OTFS modulation. By doing so, the channel estimation problem is transformed into estimating the DD channel, which is more suitable for dynamic scenarios. To estimate the DD channel, a residual attention-based channel estimation (RACE) model is proposed. The RACE model outperforms existing deep learning methods and conventional approaches. It achieves a lower normalized mean square error compared to other methods.