eSNR-Adjusted Channel Decorrelation Preprocessing for AMP Data Detection in Highly Correlated THz MIMO Systems

The approximate message passing (AMP)-based data detection is a highly effective solution for terahertz (THz) multiple-input multiple-output (MIMO) communications, enabling reliable data detection at ultra-high data rates. However, in the uplink of THz MIMO systems, high channel correlation leads to performance degradation and computational inefficiencies. To address these challenges, we develop correlated probability estimation (CPE) for the standard AMP iterative data detection algorithm (AMP-IDA), achieving Bayesian-optimal (BO) bit error rate (BER) performance in highly correlated THz channels. To mitigate the significant computational complexity of CPE, we propose an effective signal-to-noise ratio (eSNR)-adjusted channel decorrelation preprocessing (ACDP) method, which leverages whitening transformation and convex optimization, mitigating the impact of row correlation without prior knowledge of correlation indices. By integrating eSNR-ACDP with the low-complexity standard AMP-IDA, we design the ACDP-AMP-IDA, which attains BER close to the BO benchmark with significantly reduced complexity. Compared to orthogonal AMP (OAMP) algorithms, ACDP-AMP-IDA outperforms standard OAMP by up to 8 dB and achieves performance comparable to OAMP with linear minimum mean square error (MMSE) while incurring only 3%–6% of its runtime. Additionally, it surpasses existing AMP-IDA-based and MMSE detectors by over 10 dB and guarantees robust convergence across various transmitter-receiver distances in uplink THz MIMO systems.