Joint phase shift and beamforming channel estimation for reconfigurable intelligent surface multiple-input multiple-output systems

Channel estimation plays a crucial role in optimizing signal quality within wireless communication systems. Reconfigurable Intelligent Surfaces improve signal propagation through passive elements due to their low cost and high energy efficiency. With the increasing number of mobile users, 5G faces challenges, making accurate CSI acquisition in RIS-MIMO systems essential yet challenging for three reasons. This thesis aims to enhance channel estimation accuracy and capacity across various simulation scenarios. Channel estimation in RIS-MIMO systems faces three main challenges. First, multipath propagation causes interference due to reflections, diffractions, and scattering. Second, the least square estimation at pilot positions produces low-resolution, noisy images, reducing system accuracy. Finally, signals reflecting off the RIS undergo phase shifts and amplitude changes, requiring more pilot signals and passive elements, leading to high training overhead. These challenges highlight the complexity of RIS-MIMO and the need for advanced solutions to maximize its potential. Using different symbol mapping techniques, this thesis first introduces the Least Square (LS) estimator for various multipath fading channels. Performance was evaluated based on Bit Error Rate (BER), Throughput, and Mean Square Error (MSE). Results indicate that the proposed LS method outperforms the traditional LS in Rayleigh, Rician, and AWGN channels, especially for 64-QAM in downlink scenarios. Using a diamond pilot pattern, the proposed LS reduces BER at 20 dB by approximately 50% for AWGN, 40% for Rayleigh, and 33.33% for Rician compared to the traditional LS with a block pilot pattern. Second, we introduced the Super Resolution Image Restoration Channel Network (SRIR-ChNet) algorithm, which addresses the cascaded downlink RIS-MIMO channel estimation by framing it as an image super-resolution task to improve the accuracy of low-resolution estimates and noisy channel representations. The results indicate that SRIR-ChNet achieves MSE values between 10−4 and 10−3 , outperforming Generative Adversarial Networks for convolutional blind denoising networks(GAN-CBD) and CAE-ChannelNet. Additionally, SRIR-ChNet has a total time complexity of 0.86×10−2 s, which is lower than CAE-ChannelNet and GAN-CBD. Finally, we proposed the PSBA-H model for cascaded and separated channels and developed a system to manage the challenges associated with active and passive elements in the RIS. We analyze two scenarios: the estimation of the downlink cascaded MIMO channel from BS to UE via the RIS and the separate estimation of the BS-RIS and RIS-UE channels. The results show that the PSBA-HT,K,R method for the downlink cascaded channel achieves superior capacity performance, reaching 14.3 bits/Hz at 30 dB SNR with K=32 elements, outperforming DDL (11.2 bits/Hz) and DQN (11 bits/Hz). In separate downlink channels, PSBA- HT,K reaches approximately 19 bits/Hz with K=64 elements and 13 bits/Hz with K=32 at 30 dB SNR. Similarly, for separate uplink channels, the PSBA-HK,R method achieves nearly 18.5 bits/Hz at 30 dB SNR with K=64 elements and 12.3 bits/Hz with K=32. The results indicate that the proposed PSBA-H methods for the cascaded channel take about 0.0075 seconds, while the separate channels range from 0.0092 to 0.0095 seconds. The PSBA-H model outperforms DDL and DQN in capacity and computational efficiency due to joint optimization of the phase shift and beamforming matrix toward accurate channel estimates in RIS-MIMO systems.

Text FK 2025 4 - Full Text.pdf Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (7MB) Text FK 2025 4.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (8MB)

Actions (login required)

View Item