REVIEW 2 major objections 3 minor 7 references
Slepian-Bangs formula and Cramer Rao bound for circular and non-circular complex elliptical symmetric distributions
T0 review · 2 major / 3 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read The paper extends the Slepian-Bangs formula to non-circular complex elliptical symmetric distributions, giving a closed-form Fisher information matrix and showing that Gaussian noise is not always the worst case.
desk verdict This supplement genuinely extends the Slepian-Bangs formula to non-circular CES and corrects the Gaussian-worst-case belief, but the stochastic representation proof skips a Takagi step and has typos that need fixing. 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 stochastic representation $z = \mu + R A[\Delta_1 u + \Delta_2 u^*]$, where $u$ is uniformly distributed on the complex unit sphere, $R$ is a nonnegative radial variable, and $\Delta_1, \Delta_2$ are real diagonal matrices satisfying $\Delta_1^2+\Delta_2^2=I$ and $2\Delta_1\Delta_2=\Delta_\kappa$. This converts expectations with respect to the full density into expectations over the scalar $Q$, whose distribution is carried by $\xi_1 = E[Q\varphi(Q)^2]/M$ and $\xi_2 = E[Q^2\varphi(Q)^2]/[M(M+1)]$. Lemma 1, which computes $E[(\tilde{y}^H\tilde{A}\tilde{y})(\tilde{y}^H\tilde{B}\tilde{y})]$ for an augmented complex Gaussian vector, supplies the combinatorial coefficients in the Fisher information matrix. Together these pieces turn the Slepian-Bangs formula into a pair of moment computations.
What would settle it
Construct a non-circular CES density whose scatter matrix $\Sigma$ and pseudo-scatter matrix $\Omega$ are not simultaneously diagonalizable in the required real-diagonal sense, then compare its Fisher information with the formula of Result 2; if the formula fails, the representation theorem is the point of collapse. Alternatively, simulate a heavy-tailed non-circular sample with $\xi_2<1$ and check the prediction that its stochastic CRB exceeds the Gaussian SCRB.
Extended reading notes
Core claim
The central discovery is Result 2: for NC-CES observations with augmented covariance matrix $\tilde{\Gamma}$, the Fisher information matrix is given in closed form as a combination of terms involving $\tilde{\Gamma}^{-1}\tilde{\Gamma}_k$, $\tilde{\Gamma}^{-1}\tilde{\Gamma}_l$, and the mean derivatives, with coefficients fixed by the two density-generator moments $\xi_1$ and $\xi_2$. This generalizes the classical Slepian-Bangs formula, previously known only for Gaussian and circular CES cases. Result 4 then shows that the difference between the NC-CES and non-circular Gaussian information along covariance parameters is proportional to $\xi_2-1$ times a positive-definite quadratic form, so whenever $\xi_2<1$ the non-Gaussian information is strictly smaller and the corresponding stochastic CRB is larger. The proofs in Sections III through VI establish these statements for a general parameterization and then specialize them to the noisy mixture, yielding explicit SCRB expressions for the parameters of interest.
Load-bearing premise
The stochastic representation theorem assumes every NC-CES vector can be written with real diagonal matrices $\Delta_1$ and $\Delta_2$, and the proof asserts this diagonal form can be chosen without showing that it holds for arbitrary scatter and pseudo-scatter matrices.
Editorial extensions
If this is right
- For any NC-CES distribution, the stochastic CRB can be computed from the augmented covariance matrix and the two generator moments $\xi_1$ and $\xi_2$, without specifying the full density.
- The non-circular complex Gaussian and the circular CES bounds are both special cases of the new formula, so the result unifies previously separate derivations.
- When $\xi_2<1$, the NC-CES Fisher information is strictly smaller than the Gaussian information along covariance-parameter directions, meaning the Gaussian distribution does not always give the largest SCRB.
- For the noisy mixture model, the supplement derives closed-form SCRBs for the parameters of interest, covering direction-of-arrival and source parameter estimation for both circular and non-circular observations.
- The formulas hold for arbitrary parameterizations of the spatial signature matrix $A_\theta$, including near-field and far-field DOA models with scalar or vector sensors.
Reading between the lines
- Because the FIM depends on the density only through $\xi_1$ and $\xi_2$, the same formula could support a plug-in CRB: estimate those two moments from data and compute a bound without assuming a full parametric generator.
- The proof strategy for the noisy mixture may transfer to other array-processing models beyond the near-field and far-field DOA examples named in the supplement, such as models with structured or partially correlated sources.
- If a real heavy-tailed non-circular generator has $\xi_2<1$, then designs based on the Gaussian CRB will systematically underestimate achievable variance, suggesting a practical diagnostic: estimate $\xi_2$ before trusting Gaussian-based bounds.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This manuscript is a detailed-proof supplement to the authors' accepted letter [1]. It gives proofs of six results: a stochastic representation theorem for non-circular complex elliptically symmetric (NC-CES) random vectors (Result 1); the Slepian-Bangs formula, i.e., the Fisher information matrix (FIM), for NC-CES distributions (Result 2); the moment identity E[Q phi(Q)] = -M and the inequality M^2 <= M xi_1 (Eq. (9) of [1]); the claim that the Gaussian distribution is not always the least favorable for the stochastic CRB (Result 4); and closed-form stochastic CRBs for noisy mixture models in circular and non-circular cases (Results 5 and 6). The central derivation in Section III expresses the FIM in terms of the augmented covariance matrix and the density-generator moments xi_1 and xi_2, following the template of Besson and Abramovich for circular CES. The paper is self-contained modulo the statements of results in [1], and the algebraic derivations are presented in considerable detail.
Significance. If the representation theorem is properly justified, the main formula is a genuine and useful unification: it contains the non-circular Gaussian and the circular CES cases as special cases, and it is stated in a form directly usable for CRB computations. The proof of Result 4 is also valuable because it gives a concrete parameter regime (xi_2 < 1) in which the common 'Gaussian is least favorable' intuition fails. The supplement is careful in its FIM derivation: the regularity condition (19), the moment identities (22)-(28), and the use of Lemma 1 for the fourth-order moments are internally consistent and appear correct. The main weakness is not the FIM computation itself but the foundational stochastic representation theorem in Section II, whose proof is incomplete as written.
major comments (2)
- [Section II, Eq. (15)] The proof of Result 1 does not justify the reduction to real diagonal matrices (Delta_1, Delta_2). The text states that solutions of (15) 'can be looked for' in real-valued diagonal form and then proceeds to (16), but no argument is given that this restriction is without loss of generality for arbitrary scatter matrix A A^H and pseudo-scatter matrix A Delta_kappa A^T. This is load-bearing because the FIM derivation in Section III uses the canonical augmented representation ~z = ~mu + R ~Gamma^{1/2} ~u, which is obtained from the diagonal choice. A rigorous proof should show either that any solution of (14) yields the same augmented covariance matrix ~Gamma (so the distribution depends only on ~Gamma) or provide a direct simultaneous congruence argument. It should also show that the scalar equations (16) are solvable, i.e., that |kappa_i| <= 1, which follows from the positive semidefiniteness of [[I, Delta_kappa], [Delta_kappa, I]]. As written, the representation theorem is not established for all NC-CES distributions.
- [Section II, Eqs. (13)-(18)] The algebra in the reduction contains several errors that obscure the step. Eq. (13) should read Psi Phi^T + Phi Psi^T, not Psi Phi^T + Psi Phi^T. The change of variables should be Psi' = A^{-1} Psi and Phi' = A^{-1} Phi, not Psi' = A Psi. Eq. (15) is missing primes and should be I = Psi' Psi'^H + Phi' Phi'^H and Delta_kappa = Psi' Phi'^T + Phi' Psi'^T. Finally, the eigenvalue decomposition before Eq. (18) should have blocks I + Delta_kappa and I - Delta_kappa, not two copies of I + Delta_kappa. These are presentation issues in a load-bearing part of the proof, and they should be corrected together with the missing justification in the previous comment.
minor comments (3)
- [Section VI, Eq. (52)] In the final trace formula, the matrix H is used without being defined. From the preceding line, H should be R_s A_theta^H Sigma^{-1} A_theta R_s (or an equivalent simplification); please define it explicitly.
- [Section VII] The proof of Result 6 is mostly an appeal to [7, th. 1] with substitutions, and the 'key form expression' for Pi_V^perp is stated without derivation. For a supplement labelled 'detailed proofs', a fuller derivation of the non-circular noisy-mixture CRB would be helpful, even if it follows the template of Result 5.
- [Global] There are several typographical slips: 'cheek' should be 'check' (Section III), 'manducation' should be 'manipulation' (Section VI), and 'foll ows' should be 'follows'. These do not affect the mathematics but should be cleaned up.
Circularity Check
No circularity: the Slepian-Bangs formula is derived from the stochastic representation and standard Gaussian moment identities, not from the result being proved.
full rationale
The derivation is self-contained in the relevant sense. Result 2's FIM expression is obtained in Section III by differentiating the NC-CES density, substituting the extended stochastic representation ~z = ~μ + R~Γ^{1/2}~u, and evaluating expectations in Q and ~u using standard identities: E(Qφ(Q)) = -M, E(˜u˜u^H) = I/M, and the augmented Gaussian fourth-moment Lemma 1. The target FIM formula does not enter as an input; the quantities ξ1 and ξ2 are defined from the density generator, not fitted to data. Citations to prior work are used only as proof templates or auxiliary identities: [5] supplies the regularity-condition template, [2] supplies a standard complex-Gaussian expectation identity used in Lemma 1, and [7] is a template for the application in Section VII. None of these citations assumes the Slepian-Bangs formula being proved. The proof of Result 1 does assert without a complete existence argument that solutions to (15) can be sought in real diagonal form; this is a rigor gap in the canonical-form argument, not a circular reduction, since the diagonal solution is exhibited rather than derived from the conclusion. Result 4 follows algebraically from the derived FIM and the positive definiteness of the displayed matrix, not from a re-imported Gaussian comparison. Accordingly, no step reduces by construction to its own input, and the circularity score is 0.
Assumptions & free parameters
assumptions (3)
- domain assumption A NC-CES random vector z admits the stochastic representation z = µ + R[Ψu + Φu*] with u uniform on the complex unit sphere and R independent of u.
- domain assumption The density of a CES distribution depends on the observation only through the quadratic form Q = (1/2)(\tilde z - \tilde µ)^H \tilde Γ^{-1} (\tilde z - \tilde µ) and the density generator g.
- domain assumption The regularity conditions for the Fisher information matrix hold: the pdf is differentiable in the parameters and integration can be interchanged with differentiation.
Cite this review
Pith. "Pith review of Slepian-Bangs formula and Cramer Rao bound for circular and non-circular complex elliptical symmetric distributions." pith.science (2026). https://pith.science/paper/LLWXC64T
@misc{pith2026190803544,
author = {Pith},
title = {Pith review of: Slepian-Bangs formula and Cramer Rao bound for circular and non-circular complex elliptical symmetric distributions},
year = {2026},
howpublished = {\url{https://pith.science/paper/LLWXC64T}},
note = {Machine review of arXiv:1908.03544}
}
read the original abstract
This paper is mainly dedicated to an extension of the Slepian-Bangs formula to non-circular complex elliptical symmetric (NC-CES) distributions, which is derived from a new stochastic representation theorem. This formula includes the non-circular complex Gaussian and the circular CES (CCES) distributions. Some general relations between the Cramer Rao bound (CRB) under CES and Gaussian distributions are deduced. It is proved in particular that the Gaussian distribution does not always lead to the largest stochastic CRB (SCRB) as many authors tend to believe it. Finally a particular attention is paid to the noisy mixture where closedform expressions for the SCRBs of the parameters of interest are derived.
Reference graph
Works this paper leans on
-
[1]
H. Abeida and J.P . Delmas, ”Slepian-Bangs formula and Cr am´ er Rao bound for circular and non-circular complex ellip tical symmetric distributions,” accepted to IEEE Signal Process. Lett. , August 2019
work page 2019
-
[2]
H. Abeida and J.P . Delmas, ”MUSIC-like estimation of dir ection of arrival for noncircular sources,” IEEE Trans. Signal Process. , vol. 54, no. 7, pp. 2678-2690, 2006
work page 2006
-
[3]
E. Ollila and V . Koivunen, ”Generalized complex ellipti cal distributions,” in Proc. SAM W orkshop, pp. 460-464, July 2004
work page 2004
- [4]
-
[5]
O. Besson and Y . I. Abramovich, ”On the Fisher informatio n matrix for multivariate elliptically contoured distribu tions,” IEEE Signal Process. Lett. , vol. 20, no. 11, pp. 1130-1133, Nov. 2013
work page 2013
-
[6]
The stochas tic CRB for array processing: A textbook derivation,
P . Stoica, A. G. Larsson, and A. B. Gershman, “The stochas tic CRB for array processing: A textbook derivation,” IEEE Signal Process. Lett. , vol. 8, no. 5, pp. 148-150, May 2001
work page 2001
-
[7]
H. Abeida and J.P . Delmas, ”Direct derivation of the stoc hastic CRB of DOA estimation for rectilinear sources,” IEEE Signal Process. Lett. , vol. 24, no. 10, pp. 1522-1526, 2017
work page 2017
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.