Technical Summary: Computational and Effective Degrees of Freedom for Spatially Stationary HMIMO Channel Modeling

Problem Statement

The rapid evolution of holographic MIMO (HMIMO) systems necessitates a physically compliant and mathematically tractable framework for modeling continuous aperture channels. While prior work established the spectral-order Nyström method with Gauss-Legendre quadrature (NGLQ) as a superior discretization technique over traditional Fourier series expansions, two fundamental theoretical gaps remained:

1. Lack of Rigorous Convergence Analysis: Previous selections of quadrature nodes relied on heuristic refinements of physical degrees of freedom (DoF) without a theoretical basis for the onset of numerical convergence.
2. Ill-Conditioning in Eigenvalue Decomposition: The Nyström discretization of smooth spatial correlation kernels leads to rapidly decaying eigenvalues. Truncating the eigenvalue decomposition (EVD) without a precise theoretical threshold results in numerical instability (ill-conditioning) when dividing by near-zero eigenvalues, or unnecessary computational complexity if too many modes are retained.

There is a critical need to rigorously define the Computational DoF (cDoF) required to initiate super-exponential convergence and the Effective DoF (eDoF) that characterizes the numerically resolvable rank of the channel under finite machine precision. Furthermore, extending these analyses from 1D linear arrays to 2D rectangular apertures with piecewise smooth boundaries presents significant mathematical challenges, particularly regarding the asymptotic distribution of eigenvalues for discontinuous test functions.

Methodology

The paper establishes a comprehensive theoretical framework integrating operator theory, asymptotic analysis, and numerical linear algebra:

1. NGLQ Discretization: The continuous Karhunen-Loève Decomposition (KLD) is discretized using the Nyström method with Gauss-Legendre quadrature. This preserves the continuous spatial correlation properties of the channel.
2. Operator-Theoretic Convergence Analysis:
   • For 1D arrays, the authors map the physical domain to a standard interval and analyze the interpolation remainder. They prove that the quadrature error exhibits super-exponential decay once the number of nodes exceeds a specific threshold.
   • For 2D rectangular arrays, a tensor-product approach is employed. The error is decomposed into 1D components, and the multivariate Paley-Wiener theorem is extended to bound the partial derivatives of the integrand.
3. Asymptotic Eigenvalue Analysis: To address the ill-conditioning of the EVD, the paper invokes the multidimensional Szegő-Widom asymptotic expansion. This allows for the derivation of a semi-analytical expression for the eDoF, accounting for anisotropic boundary truncation effects in 2D rectangular apertures.
4. Numerical Evaluation:
   • Isotropic Scattering: Closed-form kernels (e.g., sinc kernel) are used to validate the theoretical cDoF and eDoF thresholds.
   • Non-Isotropic Scattering: To eliminate interpolation error floors inherent in traditional Inverse Discrete Fourier Transform (IDFT) plus interpolation methods, the authors deploy the exact Non-Uniform Discrete Fourier Transform (NUDFT). This is applied to the von Mises-Fisher (vMF) scattering model to evaluate spatial correlation functions with machine-precision accuracy.

Key Contributions

1. Definition and Proof of Computational DoF (cDoF)

The paper defines the cDoF as the theoretical threshold for the onset of super-exponential quadrature convergence.

• 1D Result: The cDoF is derived as $M_c \approx \pi L / \lambda$. This reveals a fundamental $\pi/2$ oversampling penalty over the physical DoF ($2L/\lambda$).
• 2D Result: For separable 2D grids, this penalty compounds, resulting in a computational redundancy of approximately 68% compared to the physical DoF ($\pi A / \lambda^2$). This redundancy arises because separable tensor-product grids must resolve a rectangular bounding box in the wavenumber domain, whereas physical propagating waves are confined to a circular disk.
• Tightness: The threshold is proven tight under worst-case end-fire conditions, where the spatial frequency is maximized.

2. Derivation of Effective DoF (eDoF)

To resolve the ill-conditioning of the Nyström discretization, the paper derives a physically grounded, semi-analytical expression for the eDoF:

• 1D: The eDoF is approximated by the classical length law plus a logarithmic edge-correction term: $N_{eDoF} \approx \lfloor \frac{2L}{\lambda} + \frac{1}{\pi^2} \ln(\frac{1-\epsilon}{\epsilon}) \ln(\frac{\pi L}{\lambda}) \rfloor$.
• 2D: A new asymptotic expansion is proposed for 2D rectangular apertures, decoupling the area-law term from a cross-coupled logarithmic boundary term. This expression captures anisotropic boundary truncation effects and provides a theoretical prior for partial EVD solvers (e.g., Lanczos algorithm), significantly reducing computational complexity for large apertures.

3. Environment-Aware NGLQ Scheme

The framework is shown to adapt to specific channel conditions. By restricting the wavenumber domain support based on the maximum elevation ($\theta_{max}$) and azimuth ($\phi_{max}$) angles, the required number of GLQ nodes can be substantially reduced, offering a practical method to lower cDoF in sectorized deployments.

4. Elimination of Interpolation Error Floors

By utilizing the exact NUDFT for non-isotropic scattering models, the paper demonstrates that interpolation error floors can be entirely eliminated. This enables spectral convergence down to machine precision ($10^{-14}$ to $10^{-16}$), confirming the validity of the proposed framework even in complex, non-isotropic environments.

Results

• Convergence Verification: Numerical evaluations confirm that the quadrature error exhibits super-exponential decay immediately after the derived cDoF threshold ($M \approx \pi L / \lambda$).
• eDoF Accuracy: The proposed semi-analytical eDoF expressions serve as accurate asymptotic approximations. For large normalized apertures, the relative gap between the theoretical eDoF and the numerically observed rank drops significantly (e.g., from 37.2% at small scales to 5.5% at large scales).
• Stability: The proposed "numerically stable regime," which truncates eigenmodes based on the eDoF and machine precision, successfully prevents the amplification of numerical noise while maintaining high reconstruction accuracy.
• Non-Isotropic Performance: Simulations using the vMF model and NUDFT show that the Mean Squared Reconstruction Error (MSRE) reaches machine precision levels, validating the framework's effectiveness beyond isotropic assumptions.

Significance and Claims

The paper claims to provide the first comprehensive theoretical framework that unifies the concepts of cDoF and eDoF for continuous-to-discrete HMIMO channel modeling. Its significance lies in:

1. Rigorous Theoretical Foundation: It moves beyond heuristic node selection to provide mathematically proven convergence thresholds and eigenvalue distributions for 2D rectangular apertures with piecewise smooth boundaries.
2. Computational Efficiency: By identifying the eDoF, the paper offers a method to guide partial EVD solvers, reducing the complexity of channel reconstruction from $O(M^3)$ to approximately $O(M^2k)$, where $k$ is the effective rank.
3. Numerical Stability: It resolves the inherent ill-conditioning of Nyström discretization by defining a safe truncation regime, ensuring that discrete channel models are both accurate and numerically stable.
4. Practical Adaptability: The framework accommodates environment-aware constraints (sectorization) and non-isotropic scattering, making it applicable to realistic wireless network deployments.

The authors conclude that while the current work addresses spatially stationary (far-field) scenarios, the underlying physical constraints (band-limited wavenumber domain) suggest that the spectral convergence properties may extend to near-field scenarios, provided evanescent waves are neglected.