Channel Estimation for Wideband Multi-RIS-Assisted mmWave Massive-MIMO OFDM System With Beam Squint Effect

Reconfigurable intelligent surfaces (RISs) and massive multiple-input multiple-output (massive-MIMO) systems are promising technologies for improving the energy efficiency of millimeter-wave (mmWave) communication. Furthermore, in urban areas, where there is high obscurity, multiple RISs can be deployed to circumvent blockages between communicating nodes. However, deploying both multi-RIS and massive-MIMO systems significantly increases the dimensionality of a wireless communication channel and thus, accurate channel state information (CSI) acquisition by channel estimation (CE) becomes non-trivial mainly due to the passive nature of the RISs. Additionally, existing wideband RIS-assisted CE schemes ignore the beam squint effect despite its severe CE performance degradation. Therefore, in this paper, a beam squint aware machine learning (ML)-based uplink CE scheme for wideband multi-RIS-assisted mmWave massive-MIMO orthogonal frequency division multiplexing (OFDM) system is proposed. Specifically, to reduce the beam squint effect, the bandwidth of the system is divided into subbands, and thereafter, a denoising convolutional neural network bidirectional long-short term memory (DnCNN-Bi-LSTM) scheme is proposed for cascaded uplink CE. For certain parameter settings, the proposed beam squint aware DnCNN-Bi-LSTM CE scheme achieves better normalized minimum mean squared error (NMSE) performance than the state-of-the-art beam squint aware CE methods.