On channel estimation and memory inference in MIMO systems

Abstract

The design of robust channel estimation algorithms is essential to meet the increasing demands of next-generation wireless systems. Massive multiple-input multiple-output (MIMO) and ultra-dense antenna deployments present new opportunities for spatial multiplexing and spectral efficiency, but they also introduce challenges that are often overlooked in conventional estimation frameworks. One such challenge is the presence of mutual coupling (MC) among closely spaced antenna elements, which degrades channel state information (CSI) accuracy if not modeled properly. We address this issue by proposing a physically consistent Kalman filtering framework that accounts for MC effects in compact MIMO arrays through impedance-aware modeling. By embedding coupling transformations into the channel observation model, our method enables recursive tracking of the wireless channel while accurately reflecting the mutual coupling behavior of the antenna array. Simulation results demonstrate significant performance gains over conventional least squares and LMMSE estimators, particularly under strong spatial correlation and compact array configurations. Building on the spatial-domain improvements in the first part of the thesis, we next address the temporal behavior of the channel. Even with MC-aware estimation, pilot-based updates can be costly in fast-changing or resource-limited settings. Instead of estimating the full CSI at every step, we investigate whether the channel’s memory and its stability over time can be inferred directly from received signal sequences to guide smarter update decisions. While most estimation algorithms assume quasi-static or memoryless fading, realistic wireless environments exhibit time correlation that can be exploited for prediction and adaptation. We leverage Dynamic Mode Decomposition (DMD) to analyze the evolution of the CSI over time. By processing sequences of received signals, DMD identifies dominant temporal modes whose spectral properties reveal the channel's memory structure. We also compare DMD with principal component analysis (PCA) and simulation results demonstrate that, unlike PCA, the spectrum of the DMD operator effectively captures the channel memory. Together, these two contributions form a comprehensive estimation framework that integrates physical modeling with data-driven inference. The results highlight the importance of accounting for both spatial and temporal effects in wireless channel estimation and point toward more resilient and adaptive communication strategies for 6G and beyond.