REVIEW 3 major objections 4 minor 9 references
Electromagnetic Channel Statistics for Continuous-Aperture Array (CAPA) Systems
T0 review · 3 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash
Pith's one-line read The paper shows that a CAPA's fading SNR is a finite weighted sum of exponentials whose term count is the electromagnetic degrees of freedom $2L/\lambda$.
desk verdict The paper's central SNR derivation contains a load-bearing error, but the intended result is likely repairable. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the autocorrelation kernel $K(z,z') = \frac{1}{2\pi}\int_{-k_0}^{k_0} e^{j(z-z')\kappa_z}\,d\kappa_z$ acting on $L^2([-L/2,L/2])$; this is a sinc kernel, and its eigenvalues are the $\sigma_\ell$ scaled by the constant in (10). Landau's eigenvalue theorem for bandlimited kernels shows that, as $L\to\infty$, the fraction of eigenvalues exceeding any fixed positive threshold tends to the electromagnetic degrees of freedom $DOF = 2L/\lambda$, with a transition band whose width grows like $\log DOF$. That step spectrum is what truncates the infinite Karhunen--Loève expansion of the Gaussian field to a finite weighted sum of exponentials, and it is what makes the PDF and capacity integrals evaluable in closed form.
What would settle it
A direct check is to measure the SNR of a CAPA of length $L$ with a single-antenna receiver in an anechoic chamber built for isotropic Rayleigh fading, and compare the empirical PDF with Eq. (19). Equivalently, sample the spatial response $g(z)$ across the aperture, estimate its autocorrelation, and compute the eigenvalue spectrum: the prediction is that roughly $2L/\lambda$ eigenvalues sit near one and the rest near zero. A failure to see that step, or a systematic deviation of the SNR distribution from (19) beyond Monte Carlo error, would invalidate the central approximation. A cheaper boundary test is to set $L=\lambda$: the step is then weak, so the truncation in (18) should be visibly inaccurate, delimiting how large the aperture must be.
Extended reading notes
Core claim
Under the isotropic Gaussian scattering model of Eqs. (4)--(5), the normalized EM spatial response $g(z)$ is a zero-mean stationary complex Gaussian process with autocorrelation $R_g(z,z') = \frac{1}{2k_0}\int_{-k_0}^{k_0} e^{j(z-z')\kappa_z}\,d\kappa_z$. Its Karhunen--Loève expansion gives $\gamma \stackrel{d}{=} \bar{\gamma}\sum_{\ell=1}^{\infty}\sigma_\ell|\Phi_\ell|^2$, where the $\Phi_\ell$ are independent $\mathcal{CN}(0,1)$ variables and the $\sigma_\ell$ are the eigenvalues of that sinc kernel. By Landau's eigenvalue theorem these eigenvalues polarize: for $L\gg\lambda$, the leading $DOF = 2L/\lambda$ eigenvalues are near one and the rest are near zero, with a transition band of width proportional to $\log DOF$. The paper therefore approximates the infinite sum by the first $DOF$ terms and, using the series representation of a sum of independent exponentials, obtains the SNR PDF (19) and the average capacity (22), including the high-SNR asymptotics (23) with unit slope and the explicit power offset (24).
Load-bearing premise
The derivation requires the scattering to be isotropic and the angular response $W(k,\kappa)$ in Eq. (5) to be a zero-mean unit-variance complex Gaussian random field; without that, $g(z)$ is not a stationary Gaussian process with a sinc autocorrelation, and the Landau step spectrum with $2L/\lambda$ significant eigenvalues does not apply.
Editorial extensions
If this is right
- The high-SNR capacity slope of a single-user CAPA link is exactly $1$ bit/s/Hz per 3 dB regardless of aperture size; a larger aperture only improves the power offset $\mathcal{L}$ in (24).
- For design purposes, the average capacity of a CAPA under isotropic fading can be computed in closed form from $L$, $\lambda$, and transmit power, replacing Monte Carlo simulations.
- The same eigenvalue expansion gives other performance metrics such as outage probability directly, since the SNR PDF (19) contains all the statistical information.
- A CAPA of physical length $L$ under isotropic scattering has $2L/\lambda$ effective fading branches, so increasing the aperture beyond that does not add independent signal components; capacity gains come from better weighting of the existing branches.
Reading between the lines
- For non-isotropic but spatially stationary scattering, the kernel would be a filtered version of the sinc kernel, and Landau-type estimates would still predict a step spectrum; the effective number of branches would equal the measure of the angular support times the aperture length, so the paper's framework extends in spirit to directive scattering once the kernel is known.
- The truncation at $DOF$ is a convenience, not a necessity: Eq. (19) is the exact distribution of any finite weighted sum of exponentials, and the infinite sum is the limit as $DOF\to\infty$, so one could retain small tail eigenvalues if a particular aperture length requires it.
- Because the SNR is a sum of $DOF$ independent exponentials with generally distinct scales, the outage CDF near zero behaves like $x^{DOF}/(DOF!\prod \sigma_\ell)$, implying a diversity order of $DOF$ for fixed-rate transmission, even though the high-SNR capacity slope is capped at 1.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript analyzes the statistics of the received SNR of a continuous-aperture array (CAPA) under isotropic scattering. Using a planar-wave EM channel model from the holographic-MIMO literature, it derives the spatial autocorrelation of the channel response as a sinc kernel, invokes Landau's eigenvalue theorem to show that the kernel's eigenvalues have a step-like behavior with DOF = 2L/λ, and then truncates the Karhunen–Loève expansion at DOF terms. On this basis it claims that the matched-filter SNR is statistically equivalent to a finite weighted sum of independent exponential random variables, and it gives closed-form expressions for the SNR PDF and for the average capacity, together with a high-SNR expansion. Numerical results plot the eigenvalues and compare the analytical capacity with simulations and with conventional MIMO.
Significance. The topic is timely: CAPA and holographic MIMO channel statistics are of active interest, and a simple closed-form description of fading statistics for electrically large continuous apertures would be valuable. The paper has notable strengths: it builds on an established electromagnetic channel model, uses no fitted parameters, and its numerical eigenvalue plots support the step-like spectrum predicted by Landau's theorem. However, the central matched-filter SNR derivation contains a mathematical error that, as submitted, invalidates the claimed PDF and capacity formulas. The error appears to be readily repairable, but the manuscript as it stands does not prove its main result.
major comments (3)
- [§2.1, Eq. (3)] The matched-filter SNR calculation is wrong. With j(t) = h*(r,t)/√(∫_A |h(r,t)|² dt), the matched-filter output satisfies |∫_A h(r,t) j(t) dt|² = ∫_A |h(r,t)|² dt, so the SNR is (P/σ²)∫_A |h(r,t)|² dt, not (P/σ²)|∫_A h(r,t) dt|² as written immediately after Eq. (3). This is not a minor notational slip: it changes the SNR by the ratio |∫ h|²/∫|h|², which is not equal to unity for the random channel considered here.
- [§3.3, Eq. (17)] The asserted equivalence |∫_A h(r,t) dt|² d= ∫ |g(z)|² dz in Eq. (17) is false for the sinc-correlated Gaussian process derived in §3.2. For L ≫ λ, E|∫g|² = ∫∫ sinc(k₀(z−z′)) dz dz′ ≈ λ/2, whereas E∫|g|² = L; the former saturates while the latter grows linearly in L. The Karhunen–Loève expansion in Eq. (16) applies to ∫|g|², not to |∫g|², so the truncation at DOF terms in Eq. (18) is not established for the printed SNR formula. With the correct matched-filter SNR, the result can be repaired as γ d= (P/σ²)∑_{ℓ=1}^{DOF} σ_ℓ|Φ_ℓ|², but Eq. (18) as written omits the P/σ² factor and Eq. (19) is the PDF of the unscaled sum, not of the SNR in Eq. (3). Consequently Eqs. (19)–(24) do not follow from the stated system model without additional scaling and re-derivation.
- [§4, Figure 3] The numerical validation should be clarified. If the 'Simulation' curves are generated from the same Karhunen–Loève truncation used to derive the analytical formulas, then Figure 3 validates the closed-form evaluation of the truncated sum but does not independently validate the truncation of the physical channel. After correcting Eq. (17), the simulations should be rerun against the physical channel model and the accuracy of the DOF truncation should be explicitly reported.
minor comments (4)
- [§3.3, Eqs. (19)–(24)] The symbol γ is used both for the random SNR and for the fixed transmit SNR P/σ²; introduce a distinct symbol such as η or γ̄ for the transmit SNR throughout the PDF and capacity expressions.
- [Figure 2 caption] The caption contains a duplicated phrase: 'exhibit a step-like behavior: exhibit a step-like behavior:'.
- [§3.2, Eq. (15)] The application of Landau's theorem to the finite-interval sinc kernel would benefit from a sentence explaining why the theorem's hypotheses are satisfied for the kernel K(z,z′) defined on [−L/2,L/2]×[−L/2,L/2].
- [§4, parameter setup] The paper should state explicitly how the MIMO comparison is made in terms of total aperture length and power normalization, since the claimed capacity gain over conventional MIMO depends on this normalization.
Circularity Check
No significant circularity: the SNR distribution and capacity derivations rely on external mathematical results and an external channel model, with no fitted parameters or self-citation load-bearing steps.
full rationale
The derivation chain is self-contained: the channel model in Eqs. (4)-(5) is taken from Pizzo et al. (external), the autocorrelation in Eq. (10) is derived directly from that model, Landau's theorem (external, Ref. [6]) provides the eigenvalue step behavior and DOF, and the PDF in Eq. (19) invokes Moschopoulos (external, Ref. [8]). No constants are fitted to the capacity or SNR data; the eigenvalues are computed from the assumed kernel, and the numerical results are validated against independent simulations. The paper's self-citations ([1]-[3]) appear only as background. A possible algebraic issue in Eq. (17) relating |∫g|^2 and ∫|g|^2 is a correctness concern, not a circular one, because it does not define a quantity in terms of the target result or fit a parameter to it. Therefore, no circular step is present.
Assumptions & free parameters
assumptions (4)
- domain assumption Isotropic Gaussian scattering field: H(k,kappa) = As(k0)/sqrt(gamma(kx,kz)gamma(kappa_x,kappa_z)) W(k,kappa), with W a zero-mean unit-variance complex Gaussian random field.
- domain assumption Scatterers confined between CAPA and user; plane-wave angular response with no evanescent waves; wavenumber k0 = 2*pi/lambda.
- standard math Landau's eigenvalue theorem, including the asymptotic polarization count in Eq. (15).
- ad hoc to paper The infinite Karhunen-Loeve expansion can be truncated at DOF terms with negligible error (Eq. (18)).
Cite this review
Pith. "Pith review of Electromagnetic Channel Statistics for Continuous-Aperture Array (CAPA) Systems." pith.science (2026). https://pith.science/paper/L4MR7LGV
@misc{pith2026250206980,
author = {Pith},
title = {Pith review of: Electromagnetic Channel Statistics for Continuous-Aperture Array (CAPA) Systems},
year = {2026},
howpublished = {\url{https://pith.science/paper/L4MR7LGV}},
note = {Machine review of arXiv:2502.06980}
}
read the original abstract
The channel statistics of a continuous-aperture array (CAPA)-based channel are analyzed using its continuous electromagnetic (EM) properties. The received signal-to-noise ratio (SNR) is discussed under isotropic scattering conditions. Using Landau's theorem, the eigenvalues of the autocorrelation of the EM fading channel are shown to exhibit a step-like behavior. Building on this, closed-form expressions for the probability distribution of the SNR and the average channel capacity are derived. Numerical results are provided to validate the accuracy of the derivations.
Figures
Reference graph
Works this paper leans on
-
[1]
Near-field communications: A comprehensive survey,
Y . Liu, C. Ouyang, Z. Wang, J. Xu, X. Mu, and A. L. Swindlehurst, “Near-field communications: A comprehensive survey,” IEEE Commun. Surveys Tuts., Early Access, 2024
work page 2024
-
[2]
CAP A: Continuous-aperture arrays for revolutionizing 6G wireless communications,
Y . Liu, C. Ouyang, Z. Wang, J. Xu, X. Mu, and Z. Ding, “CAP A: Continuous-aperture arrays for revolutionizing 6G wireless communications,” arXiv preprint arXiv:2412.00894, 2024
arXiv 2024
-
[3]
A primer on near-field communications for next- generation multiple access,
C. Ouyang, Z. Wang, Y . Chen, X. Mu, and P . Zhu, “A primer on near-field communications for next- generation multiple access,” Proc. IEEE , vol. 112, no. 9, pp. 1527–1565, Sep. 2024
work page 2024
-
[4]
Spa- tial characterization of electromagnetic random chan- nels,
A. Pizzo, L. Sanguinetti, and T. L. Marzetta, “Spa- tial characterization of electromagnetic random chan- nels,” IEEE Open J. Commun. Soc. , vol. 3, pp. 847– 866, 2022
work page 2022
-
[5]
Spatially-stationary model for holographic MIMO small-scale fading,
A. Pizzo, T. L. Marzetta, and L. Sanguinetti, “Spatially-stationary model for holographic MIMO small-scale fading,” IEEE J. Sel. Areas Commun. , vol. 38, no. 9, pp. 1964–1979, Sep. 2020
work page 1964
-
[6]
On Szegö’s eingenvalue distribution theorem and non-Hermitian kernels,
H. J. Landau, “On Szegö’s eingenvalue distribution theorem and non-Hermitian kernels,” J. d’Analyse Mathématique, vol. 28, no. 1, pp. 335–357, 1975
work page 1975
-
[7]
On Landau’s eigenvalue theorem and information cut-sets,
M. Franceschetti, “On Landau’s eigenvalue theorem and information cut-sets,” IEEE Trans. Inf. Theory , vol. 61, no. 9, pp. 5042–5051, Sep. 2015
work page 2015
-
[8]
The distribution of the sum of independent gamma random variables,
P . G. Moschopoulos, “The distribution of the sum of independent gamma random variables,” Ann. Inst. Statist. Math. (Part A) , vol. 37, no. 1, pp. 541–544, 1985
work page 1985
Show all 9 references
-
[9]
Solving Fredholm integral equations of the second kind in MA TLAB,
K. Atkinson and L. Shampine, “Solving Fredholm integral equations of the second kind in MA TLAB,” ACM Trans. Math. Softw., vol. 34, no. 4, Jul. 2008
2008
Reviewed August 8, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.