This paper presents a comparative analysis of two Generalized Block Orthogonal Matching Pursuit (G-BOMP) techniques for recovering block sparse signals in noisy compressive sensing environments. While the conventional G-BOMP relies on maximum correlation for block selection, the proposed method integrates QR decomposition to enhance numerical stability and prevent redundant block selection. Through extensive MATLAB simulations, the improved G-BOMP with QR decomposition demonstrates a 31.31% higher noise reduction compared to the standard approach, achieving a recovery error reduction of 67.90% versus 36.59% under identical conditions. Complexity analysis reveals comparable computational costs, but the QR-based method exhibits superior robustness in ill-conditioned measurement matrices. The results underscore the efficacy of QR decomposition in mitigating noise propagation and improving reconstruction accuracy, particularly in high-noise scenarios.
This paper proposes a Federated Tomlinson-Harashima Precoding (F-THP) framework for multiuser MIMO systems under a Wi-Fi 6 architecture. Unlike traditional centralized THP, which requires full channel state information (CSI) from all access points (APs) and involves transmitting large matrices over the backhaul, this paper proposes a federated model that enables distributed channel knowledge processing at edge devices. Local channel updates are aligned through Gram-Schmidt transformation into a common basis before aggregation, reducing communication overhead by up to 25% compared to centralized THP. Simulation results demonstrate that F-THP achieves comparable bit error rate (BER) performance to centralized THP under Rayleigh fading while offering scalability and privacy advantages. Specifically, F-THP shares only transformed subspace components instead of raw channel matrices, ensuring that up to 25% of sensitive CSI data remains confidential. This makes F-THP particularly suitable for real-time Wi-Fi 6 operation in enterprise and IoT edge environments.