Mismatch Analysis and Cooperative Calibration of Array Beam Patterns for ISAC Systems

Technical Summary: Mismatch Analysis and Cooperative Calibration of Array Beam Patterns for ISAC Systems

Problem Statement
Integrated Sensing and Communication (ISAC) systems rely heavily on accurate angle estimation for tasks such as localization, orientation estimation, and simultaneous localization and mapping (SLAM). However, sensing performance is frequently degraded by model mismatches arising from geometric errors (e.g., position and orientation inaccuracies) and hardware impairments (e.g., mutual coupling and amplifier non-linearity). These mismatches lead to inaccurate beamforming, reduced array gain, and amplified angular errors that severely impact system performance, particularly in millimeter-wave (mmWave) scenarios. Existing calibration methods often focus on beam pattern reconstruction accuracy or assume ideal system models, failing to explicitly quantify the impact of beam mismatch on sensing performance or to address the scalability needs of distributed, hybrid beamforming architectures.

Methodology
The paper proposes a comprehensive framework for in situ beam pattern calibration from a sensing-oriented perspective, utilizing downlink pilot signals in a distributed ISAC scenario.

1. System and Beam Models: The authors consider a downlink OFDM system with a base station (BS) equipped with a uniform planar array and multiple user equipments (UEs). They define two beam representation models: an ideal steering-based codebook and a practical model that accounts for hardware limitations via a learned codebook ($W$) and a gain matrix ($\Gamma$). The calibration task is formulated as estimating $W$ and $\Gamma$ to minimize the discrepancy between the actual and mismatched models.
2. Sensing-Oriented Performance Metric: Unlike traditional metrics based on beam pattern similarity, the authors propose a novel metric based on angle estimation error. This is derived by calculating a "pseudo-true" parameter vector that minimizes the Kullback-Leibler (KL) divergence between the true model and the mismatched model. This metric directly reflects the degradation in sensing performance caused by calibration errors.
3. Loss Functions and Optimization:
   • Response Error Loss (REL): A standard loss function minimizing the difference between the reconstructed and reference beam patterns, used as a baseline.
   • Angle Error Loss (AEL): A differentiable loss function approximated from the sensing-based metric. It avoids the computational burden of bilevel optimization by matching the normalized projection of observations, enabling gradient-based optimization.
4. Cooperative Calibration Framework: To address scalability and communication overhead, a cooperative calibration strategy is introduced. Multiple UEs perform local calibration updates based on their received data and local loss functions. These local updates are then fused globally at the BS to refine the global calibration parameters. This approach reduces the need for centralized data collection and large matrix inversions.
5. Algorithms: The paper details optimization algorithms using Alternating Optimization (AO) and Gradient Descent (GD). The GD-based approach, particularly when combined with the AEL, is shown to be effective for refining calibration parameters.

Key Contributions

• Cooperative Calibration Problem Formulation: The paper formulates a distributed beam calibration problem where multiple UEs collaboratively optimize the BS antenna array's beam pattern using communication pilot signals.
• Novel Sensing-Oriented Metric: It introduces a performance metric based on angle estimation bias (derived via KL divergence minimization) rather than simple pattern similarity, providing a more relevant measure of calibration quality for ISAC systems.
• Differentiable Loss Function: A differentiable Angle Error Loss (AEL) is proposed to facilitate iterative optimization, allowing for the refinement of communication-based calibration to improve sensing performance.
• Experimental Validation: The proposed models and algorithms are validated using real-world beam pattern measurements collected in an anechoic chamber, covering both 2D and 3D calibration scenarios.

Results
Experimental results using measured data from an anechoic chamber demonstrate significant improvements in angle estimation accuracy:

• 2D Calibration: The angle estimation error was reduced from 1.01° (uncalibrated) to 0.11°.
• 3D Calibration: The angle estimation error was reduced from 5.19° (uncalibrated) to 0.86°.
• Performance Comparison: The proposed Model M4 (Joint Codebook and Gain Matrix Calibration) combined with the AEL-based loss function achieved the best performance in terms of beam similarity, angle error, and codebook gain loss compared to benchmark models (M1-M3) and other optimization methods (AO vs. GD).
• Cooperative Strategy: The cooperative calibration approach achieved performance comparable to global centralized calibration while significantly reducing communication overhead (scaling from $M \times G \times T$ to $M \times G \times N$ complex entries).

Significance and Claims
The paper claims to establish a solid foundation for robust and adaptive beam calibration in future ISAC-enabled 6G systems. By prioritizing angular sensing accuracy over traditional pattern similarity, the proposed framework addresses a critical gap in current calibration literature. The introduction of a differentiable, sensing-oriented loss function enables the use of scalable, gradient-based optimization algorithms suitable for distributed implementations under hybrid beamforming constraints. Furthermore, the cooperative calibration strategy demonstrates that distributed updates from multiple UEs can effectively calibrate large-scale arrays without the prohibitive communication overhead of centralized methods. The authors note that while the current work focuses on specific scenarios, future research should consider calibration algorithms that account for multipath propagation and frequency-dependent beam patterns in wideband systems.