Subarray based Wideband Beamforming and Variational Sparse CSI Estimation for Low-Resolution MU THz MIMO Systems

Technical Summary: Subarray-based Wideband Beamforming and Variational Sparse CSI Estimation for Low-Resolution MU THz MIMO Systems

Problem Statement
The paper addresses the challenges inherent in Terahertz (THz) multiple-input multiple-output (MIMO) systems operating in the 0.3–10 THz band. These systems face a "dual-wideband" problem characterized by both frequency-dependent propagation effects (such as molecular absorption and frequency-selective path loss) and spatial-wideband effects (differential delays across large antenna apertures). These factors lead to the "beam-squint" effect, where the effective angle of arrival/departure varies across subcarriers, causing beam misalignment in conventional phase-shifter-based beamformers. Furthermore, practical THz systems are constrained by hardware limitations, specifically the use of low-resolution Analog-to-Digital Converters (ADCs) to manage power consumption, which introduces nonlinear quantization distortions. Existing sparse recovery techniques often fail to jointly address the dual-wideband nature of the channel, the off-grid angular domain, and the nonlinearities introduced by low-resolution quantization.

Methodology
The authors propose a unified framework integrating channel estimation and beamforming design within a Variational Bayesian Inference (VBI) paradigm.

1. System and Channel Modeling:

   • A dual-wideband channel model is formulated for a partially connected subarray architecture, incorporating Root Raised Cosine (RRC) pulse shaping to account for band-limited transmission.
   • The model explicitly includes frequency-dependent array steering vectors to capture the beam-squint effect.
   • To handle low-resolution ADCs, the nonlinear quantization process is linearized using Bussgang decomposition, assuming the input to the quantizer follows a Gaussian distribution (justified by the Central Limit Theorem due to the superposition of multiple user signals and noise).

2. Sparse Channel Estimation:

   • The channel estimation problem is cast as a compressive sensing task exploiting angular sparsity.
   • To overcome the "basis mismatch" caused by off-grid angles, the authors employ a Taylor-series-based super-resolution dictionary (TBsD). This approximates the continuous steering vector using a first-order Taylor expansion around grid points, allowing for higher resolution without the computational cost of an excessively fine grid.
   • A Variational Bayesian Inference algorithm is developed to estimate the sparse channel coefficients. This approach uses a hierarchical Gaussian-Gamma prior to promote sparsity and approximates the intractable posterior distribution with a tractable variational distribution. The algorithm iteratively updates the mean and covariance of the channel estimates and the hyperparameters of the prior.

3. Beamforming and Beam-Squint Mitigation:

   • The estimated beamspace channel is directly utilized to extract dominant Angle of Arrival (AoA) and Angle of Departure (AoD) pairs.
   • These angles configure a True Time Delay (TTD)-based hybrid transceiver. Unlike conventional phase shifters, TTD elements introduce frequency-dependent delays that inherently compensate for the beam-squint effect, ensuring consistent beam alignment across all subcarriers.
   • The design accounts for practical hardware impairments, including insertion loss in TTD elements and the quantization of delay values.

Key Contributions

• Unified Framework: The work presents a joint channel estimation and beamforming framework that simultaneously addresses dual-wideband effects, low-resolution ADC quantization, and off-grid angular sparsity.
• Variational Bayesian Estimation: Unlike greedy algorithms (e.g., OMP) or reweighting methods (e.g., MFOCUSS), the proposed VBI approach provides a probabilistic framework that quantifies uncertainty and avoids manual tuning of regularization parameters. It integrates likelihood information to refine the posterior distribution.
• Taylor-Series Dictionary: The introduction of the TBsD formulation enables accurate sparse channel estimation for off-grid angles, improving resolution and robustness compared to conventional on-grid dictionaries.
• TTD-Enabled Beamforming: The paper proposes a low-latency, pilot-efficient method to configure TTD elements directly from the estimated beamspace channel, eliminating the need for separate angle estimation modules.
• Theoretical Bounds: The Multi-User Bayesian Cramér-Rao Lower Bound (BCRLB) is derived to benchmark the performance of the proposed estimator.

Results
Extensive simulations validate the proposed framework:

• Estimation Accuracy: The proposed VBI-based estimator achieves Normalized Mean Squared Error (NMSE) performance that closely approaches the derived BCRLB, significantly outperforming Least Squares (LS), Orthogonal Matching Pursuit (OMP), and MFOCUSS, particularly in low SNR regimes and with limited pilot overhead.
• Robustness to Quantization: The system demonstrates resilience to low-resolution ADCs (1-bit to 4-bit). While 1-bit ADCs show error flooring at high SNR, 2-bit and higher resolutions approach the performance of infinite-resolution ADCs, validating the effectiveness of the Bussgang-based linearization.
• Beam-Squint Mitigation: The TTD-based hybrid transceiver design effectively aligns beams across subcarriers, restoring array gain that is otherwise lost in conventional phase-shifter designs. This results in substantial improvements in Spectral Efficiency (SE) compared to state-of-the-art schemes like Delay-Phase Precoding (DPP) and SOMP, which do not fully account for low-resolution constraints or joint estimation.
• Convergence: The variational inference algorithm exhibits monotonic convergence, typically reaching stable solutions within 5–8 iterations.
• Trade-offs: While the proposed Bayesian method incurs higher computational complexity and runtime compared to greedy algorithms, the paper notes this is a trade-off for superior estimation accuracy and robustness.

Significance
The paper claims that its primary significance lies in bridging the gap between theoretical wideband channel models and practical hardware constraints in THz communications. By integrating variational Bayesian learning with TTD-based beamforming, the work offers a robust solution for Multi-User (MU) THz MIMO systems that are limited by power-constrained low-resolution ADCs. The framework eliminates the need for separate, complex angle estimation stages, providing a streamlined, end-to-end design that mitigates the critical beam-squint effect while maintaining high spectral efficiency. The results suggest that this approach is well-suited for next-generation wireless systems where accurate channel state information and precise beam alignment are critical for achieving ultra-high data rates.