Deep Learning-Based Improved Cascaded Channel Estimation and Signal Detection for Reconfigurable Intelligent Surfaces-Assisted MU-MISO Systems

Abstract

Reconfigurable intelligent surface (RIS) consists of cost-effective passive elements which can be utilized in different scenarios in next-generation wireless communication. Deep learning (DL) algorithm plays a vital role in channel estimation (CE) due to the learning capability of DL tools to tackle the CE challenge. Bi-directional long-short term memory (Bi-LSTM) model collects data from both past (backward) and future (forward) simultaneously to improve prediction accuracy and provide an additional feature extraction capability. To take advantage of the Bi-LSTM model, in this paper, we proposed a Bi-LSTM model based CE and signal detection for RIS empowered multi-user multiple input single output downlink orthogonal frequency division multiplexing systems. To measure the performance of the proposed model, it is trained by two different deep neural network (DNN) optimization algorithms. Additionally, the proposed model is compared with four existing DNN models. The least square and minimum mean square error estimators are used to investigate CE and signal detection for performance comparison. The proposed Bi-LSTM model based RIS CE is capable of learning and generalizing rapidly and outperforms the comparable estimators when different pilots and cyclic prefix values are available. Simulation results confirm the effectiveness of the proposed model for CE and signal detection.