Pith. sign in

REVIEW 4 major objections 4 minor 33 references

Bertrand's Representation of the Optimal Detector

T0 review · 4 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A new derivation expresses the optimal detector for Gaussian analytic signals as a correlation of Bertrand-class time-frequency distributions, a form well-suited to power-law chirp detection.

desk verdict The detector derivation is solid, but the cross-Bertrand identity that carries the main result is unproved and as printed inconsistent; the paper needs major repair before the representation claim can be trusted. read the letter →

arxiv 2509.10198 v1 pith:QRENKIJ4 submitted 2025-09-12 physics.data-an astro-ph.IM

Vladimir Lenok This is my paper · ORCID
classification physics.data-anastro-ph.IM MSC 94A1262M15
keywords optimaldetectionBertrand'sclasstime-frequencydistributionspower-lawchirpanalyticsignalslikelihoodratioGaussianphase-insensitive
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that the optimal detector—the likelihood-ratio test—for Gaussian analytic signals can be rewritten exactly as a correlation of Bertrand-class time-frequency distributions. This matters because Bertrand's class is built to localize power-law chirps, the type of dispersed signals seen in pulsar and fast-radio-burst observations, whereas the usual Wigner-Ville-based (Cohen-class) representation is tied to linear chirps. The author derives the optimal detector for analytic signals from first principles, then uses a cross-Bertrand Moyal-like identity to convert each pair of inner products into an integral over the time-frequency plane. If the identity holds, the detector becomes a sum of weighted time-frequency correlations, and simulations suggest it outperforms both the Wigner-Ville representation and spectrogram correlation when the signal phase is unknown.

What carries the argument

The load-bearing tool is the cross-Bertrand class (eq. 42), which generalizes Bertrand's time-frequency distribution to two different signals X and Y. The decisive step is a Moyal-like identity (eq. 43): the integral of B^L_{X1X2} (conjugated B^A_{X3X4}) times f^{2q} over the half-plane equals the product of two inner products, with the signal pairings permuted. By picking r=1/2, q=0 and applying Parseval's theorem, this identity turns the pair of inner products in the optimal detector into a single time-frequency correlation. The combination of the 'localized' weighting µ_L (which gives ideal power-law chirp localization) with the 'auxiliary' weighting µ_A (which supplies unitarity) is what

What would settle it

Evaluate numerically both sides of equation (43) for several simple analytic signals (e.g., two overlapping chirps with different delays) at r=1/2, q=0. If the integral of B^L_{X1X2} B^{A*}_{X3X4} dtdf is not equal to (∫ X1 X3* df)(∫ X2 X4* df)^*, then the representation (46)-(47) fails. A second check: compare the pairing structure of (45) with (43) — the printed version pairs x1x3 with x2x4 in one place and x1x2 with x3x4 in the other, so a direct derivation from the definitions of B^L and B^A is needed to identify the correct identity.

Watch

Extended reading notes

Core claim

The central result is equations (46) and (47). The random component of the log-likelihood ratio becomes a sum over the Karhunen-Loève modes: each term is a weight η_i^2/(2N0+η_i^2) times the time-frequency integral of B^L_{rr} with the conjugate of B^A_{φ_iφ_i}; the deterministic component is the analogous correlation between B^L_{rm} and B^A_{φ_iφ_i}. The derivation starts from the optimal detector for analytic signals under generalized complex Gaussian noise, expands inverse covariances via Karhunen-Loève, and then uses a cross-Bertrand Moyal-like identity to convert pairs of inner products into integrals over the time-frequency plane. For a fully known signal or a Rayleigh-fading signal,

Load-bearing premise

The entire representation depends on an unproved identity, introduced as a straightforward computation, that connects the integral of a product of cross-Bertrand distributions to a product of inner products; if that identity is wrong, the detector representation collapses.

Editorial extensions

If this is right

  • If equation (43) is correct, the optimal detector for Gaussian analytic signals can be computed as a weighted sum of time-frequency correlations, extending the known Cohen-class time-frequency formulation to power-law chirp signals.
  • For a fully known signal, the detector reduces to a correlation between B^L_{rg} and B^{A*}_{gg}, a generalization of earlier path-integration chirp detectors to arbitrary power-law chirps.
  • For a Rayleigh-fading signal, the detector is a single correlation between B^L_{rr} and B^{A*}_{gg}, which is naturally phase-insensitive.
  • The representation is not unique: µ_L and µ_A can be swapped, or replaced by the unitary µ_U, yielding equivalent detector forms.
  • Numerical simulations with a dispersed chirp signal indicate that the Bertrand representation matches the matched filter in phase-sensitive detection and outperforms both Wigner-Ville and spectrogram correlation when the phase is unknown.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the identity (43) is stated without proof and its printed specialization appears to contain a weight mismatch (f^2 vs plain df) and a signal-pairing mismatch, the entire representation hinges on a missing derivation; a direct numerical check of (43) on simple analytic signals would settle it.
  • If the identity is repaired to hold for general k, the same construction would yield optimal detectors for hyperbolic and other chirp models beyond the Unterberger case studied in the simulation.
  • The phase-agnostic improvement seen in simulation comes from replacing Re with |·|; that substitution is heuristic, so the reported advantage may depend on the specific signal, noise level, and threshold choice.
  • The single simulated chirp is perfectly localized by the Unterberger distribution; testing on non-localized, noisy, or multi-component chirps would clarify how general the performance gain is.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper derives an optimal detector for analytic Gaussian signals using a generalized multivariate complex normal distribution and a likelihood-ratio argument. It then claims to represent this detector in the time-frequency domain as correlations of Bertrand-class distributions, specifically equations (46)-(47). The derivation relies on a new cross-Bertrand identity (43) which is said to follow by 'a straightforward computation.' The paper also gives two analytic examples (fully known signal and Rayleigh fading signal) and a numerical simulation for a power-law chirp with a phase-agnostic detector.

Significance. If the main representation were correct, it would extend Flandrin's Cohen-class formulation of optimal detection to Bertrand's class, which is better adapted to power-law chirps and hence relevant to pulsar and fast radio burst detection. The derivation of an optimal detector for analytic signals in Sections III-V is a useful contribution in itself. However, the central bridge between the detector and Bertrand's class, equations (43)-(45), is unproved and, as printed, inconsistent. The claimed representation (46)-(47) is therefore not established, and the numerical conclusions rest on an unsupported heuristic.

major comments (4)
  1. [Section VI, Eqs. (43)-(45)] The cross-Bertrand identity (43) is asserted without proof and is internally inconsistent as printed. With r=1/2 and q=0, the factor f^{2r+1}=f^2 on the right-hand side of (43) is silently dropped in (44). Moreover, the time-domain pairing in (45) is (x1,x2),(x3,x4), whereas the frequency-domain right-hand side of (43) pairs (X1,X3),(X2,X4). These two pairings cannot both follow from the same identity. Since (46)-(47) are based on (44)-(45), this is a load-bearing error. A correct derivation of a cross-Bertrand Moyal-type formula, with explicit hypotheses, is required.
  2. [Section VI, Eq. (46)] Even if a corrected version of (43) were supplied, the identity yields frequency integrals weighted by f^{2r+1}, not the unweighted inner products appearing in (38)-(39). For r=1/2 the weight is f^2, and no parameter choice in the printed formulas removes it. Thus the equality between (46) and (38) is not established. The detector would need to be reformulated with weighted inner products, or a different Bertrand-class parameterization would be needed.
  3. [Section VII-A, Eq. (55)] For a fully known signal (K=0), the Karhunen-Loève eigenvalues η_i^2 are zero, not E_g. The coefficient in (55) should therefore be 1/(2N0), not 1/(2N0+η_g^2). The statement that η_g^2 equals the energy of the expected signal confuses an eigenvalue of the covariance matrix with a signal energy, making the example incorrect.
  4. [Section VIII] The phase-agnostic detectors are obtained by replacing Re{·} with |·| in (66)-(68). This heuristic is not derived from the likelihood ratio and is not claimed to be optimal. The numerical comparison uses a single signal (64) with one value of b and one noise model; the conclusion that Bertrand's representation is 'superior' for phase-agnostic detection is therefore not supported by the simulations as presented.
minor comments (4)
  1. [Abstract] Typo: 'more effective then' should be 'more effective than'.
  2. [Section II] Typo: 'complex conjuration' should be 'complex conjugation'; 'power-low chirps' should be 'power-law chirps' (two occurrences).
  3. [Section VIII] In the enumeration of equations, '(66), (66)' should be '(66), (67), (68)'.
  4. [Section VII-A] Typo: 'cross-Bernard's class' should be 'cross-Bertrand's class'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the unproved cross-Bertrand identity is a verification gap, not a circular reduction.

full rationale

The central claim is a mathematical representation claim, not an empirical prediction with fitted inputs. The optimal detector (37)-(40) is derived from a generalized complex normal likelihood ratio and a Karhunen-Loeve expansion of the covariance matrices; no parameter is fit to the output of the detector and no benchmark result is used to define the detector. The Bertrand representation (46)-(47) does depend entirely on the cross-Bertrand identity (43)-(45), which is introduced as 'a straightforward computation' with no derivation. That is a serious omitted proof and the printed identity appears internally inconsistent (the r=1/2, q=0 specialization changes f^{2r+1}=f^2 to plain df, and the Parseval statement pairs x1x2 with x3x4 while (43) pairs X1X3 with X2X4). However, an unproved or even false supporting identity is a correctness risk, not circularity: the identity is not a fitted parameter, not a self-citation, and not a definition of the quantities being derived. References [11] and [19] are prior work by other authors, not self-citations, and they are used as context rather than as the load-bearing derivation. The simulation's choice of a signal perfectly localized by the Unterberger distribution is a selection bias for the phase-agnostic comparison, but the detector is not fit to the simulation outcome and the phase-sensitive representations are checked against the matched filter, so this is not a circular fit either. No step reduces by construction to its own input.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim depends on the new cross-Bertrand identity (43), which is asserted rather than derived and appears mis-specialized in (44) and (45). The detector derivation itself uses standard mathematics: complex normal distribution, Karhunen-Loeve expansion, Parseval. The simulation introduces hand-chosen parameters such as b, the window shape, and N0. No new physical entities are postulated.

free parameters (3)
  • Bertrand class parameters r and q = r=1/2, q=0 in Section VI
    Free class parameters chosen to make the TF-plane weight f^{2q}=1. But with r=1/2 the spectral weight f^{2r+1}=f^2 in (43), so the claimed simplification to plain inner products is not achieved as printed.
  • Bertrand class parameter k = k=-1 for the Unterberger distribution in the simulation
    Design choice that selects which power-law chirp family is perfectly localized. k=-1 matches the pulsar dispersion model t(f) ~ 1/f^2.
  • Signal curvature parameter b in the simulation = 4*10^5*(2*pi)^2
    Hand-chosen value in Section VIII for the cos(sqrt(b alpha)) phase. It is example-specific and not fitted to the detector, but it fixes the simulation scenario.
assumptions (5)
  • standard math The multivariate complex normal distribution (19) with arbitrary covariance V describes the analytic signals.
    Used in Section IV as the probability model for noise and signal; a generalization from [25].
  • domain assumption Analytic signals have the block covariance structure (89) to (99): the imaginary part is the Hilbert transform of the real part, and E[z_alpha z_beta]=0.
    Derived in Appendices A and B for stationary signals; this structure enables the simplification from (23) to (26).
  • ad hoc to paper The cross-Bertrand unitary relation (43) holds with the stated parameters; the text says 'a straightforward computation' but no proof is given.
    This is the load-bearing new identity behind the main representation (46) and (47). It is unproved as printed.
  • ad hoc to paper The Parseval reduction (45) correctly maps (43) to time-domain inner products with the pairing x1x2 and x3x4.
    This appears to conflict with (43), which pairs x1x3 and x2x4. If (45) is false, the main representation does not follow.
  • ad hoc to paper Phase-agnostic detection can be obtained by replacing Re with |.| in the detector formulas.
    Section VIII substitutes the absolute value operator without a derivation that this is the optimal unknown-phase test; it is a heuristic modification.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bertrand's Representation of the Optimal Detector." pith.science (2026). https://pith.science/paper/QRENKIJ4

@misc{pith2026250910198,
  author       = {Pith},
  title        = {Pith review of: Bertrand's Representation of the Optimal Detector},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QRENKIJ4}},
  note         = {Machine review of arXiv:2509.10198}
}
read the original abstract

It is shown how the optimal detector of Gaussian signals can be represented in terms of Bertrand's class of time-frequency distributions. In this representation, the detector is a correlation between the corresponding time-frequency distributions. Since Bertrand's class is related to the power-law chirp signals, the new representation can be useful for their detection. The new approach is shown to be more effective then other time-frequency methods for the case of phase-insensitive detection. The finding provides a complementary representation to Cohen's class representation in the time-frequency domain already known in the literature.

Figures

Figures reproduced from arXiv: 2509.10198 by the authors.

Figure 1
Figure 1. A spectrogram of a bright pulse of the Crab Pulsar detected by [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Simulated detection efficiencies are shown for three different optimal [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references

  1. [1]

    D. R. Lorimer and M. Kramer,Handbook of Pulsar Astronomy. Cam- bridge University Press, 2005

    2005

  2. [2]

    J. J. Condon and S. M. Ransom,Essential Radio Astronomy. Princeton University Press, 2016

    2016

  3. [3]

    Discoveries and timing of pulsars in NGC 6440,

    L. Vleeschower, B. W. Stappers, M. Bailes, E. D. Barr, M. Kramer, Ransom, and et al., “Discoveries and timing of pulsars in NGC 6440,” Monthly Notices of the Royal Astronomical Society, vol. 513, pp. 1386– 1399, June 2022

    2022

  4. [4]

    The second data release from the European Pulsar Timing Array. I. The dataset and timing analysis,

    EPTA Collaboration, “The second data release from the European Pulsar Timing Array. I. The dataset and timing analysis,”Astronomy & Astrophysics, vol. 678, p. A48, Oct. 2023

    2023

  5. [5]

    The NANOGrav 15 yr data set: observations and timing of 68 millisecond pulsars,

    Nanograv Collaboration, “The NANOGrav 15 yr data set: observations and timing of 68 millisecond pulsars,”The Astrophysical Journal Letters, vol. 951, p. L9, July 2023

    2023

  6. [6]

    Single-pulse studies of three millisecond pulsars,

    N. T. Palliyaguru, B. B. P. Perera, M. A. McLaughlin, S. Osłowski, and G. L. Siebert, “Single-pulse studies of three millisecond pulsars,” Monthly Notices of the Royal Astronomical Society, vol. 520, pp. 2747– 2756, Apr. 2023

    2023

  7. [7]

    Boashash,Time-Frequency Signal Analysis and Processing

    B. Boashash,Time-Frequency Signal Analysis and Processing. Oxford: Academic Press, second ed., 2016

    2016

  8. [8]

    MeerKAT correlator-beamformer: a real-time processing back-end for astronomical observations,

    A. van der Byl, J. Smith, A. Martens, J. Manley, T. van Balla, A. Rust, and et al., “MeerKAT correlator-beamformer: a real-time processing back-end for astronomical observations,”Journal of Astronomical Tele- scopes, Instruments, and Systems, vol. 8, no. 1, p. 011006, 2021

    2021

Show all 33 references
  1. [9]

    The LOFAR correlator: implementation and performance analysis,

    J. W. Romein, P. C. Broekema, J. D. Mol, and R. V . van Nieuwpoort, “The LOFAR correlator: implementation and performance analysis,” in Proceedings of the 15th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, PPoPP ’10, (New York, NY , USA), p. 169–178...

    2010

  2. [10]

    polyphase filter bank generator

    R. van Nieuwpoort, “polyphase filter bank generator.” https://github.com/ NLeSC/polyphase-filter-bank-generator

  3. [11]

    A time-frequency formulation of optimum detection,

    P. Flandrin, “A time-frequency formulation of optimum detection,”IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 36, pp. 1377–1384, Sept. 1988. Conference Name: IEEE Transactions on Acoustics, Speech, and Signal Processing

    1988

  4. [12]

    On the optimality of the Wigner dis- tribution for detection,

    S. Kay and G. Boudreaux-Bartels, “On the optimality of the Wigner dis- tribution for detection,” inICASSP ’85. IEEE International Conference on Acoustics, Speech, and Signal Processing, vol. 10, pp. 1017–1020, Apr. 1985

    1985

  5. [13]

    On detection-estimation procedures in the time-frequency plane,

    P. Flandrin, “On detection-estimation procedures in the time-frequency plane,” inICASSP ’86. IEEE International Conference on Acoustics, Speech, and Signal Processing, vol. 11, pp. 2331–2334, Apr. 1986

    1986

  6. [14]

    Time-frequency receivers for locally optimum detection,

    P. Flandrin, “Time-frequency receivers for locally optimum detection,” in ICASSP-88., International Conference on Acoustics, Speech, and Signal Processing, pp. 2725–2728 vol.5, Apr. 1988. ISSN: 1520-6149

    1988

  7. [15]

    The use of hy- perbolic time-frequency representations for optimum detection and pa- rameter estimation of hyperbolic chirps,

    A. Papandreou, S. Kay, and G. Boudreaux-Bartels, “The use of hy- perbolic time-frequency representations for optimum detection and pa- rameter estimation of hyperbolic chirps,” inProceedings of IEEE-SP International Symposium on Time- Frequency and Time-Scale Analysis, pp. 369...

    1994

  8. [16]

    Detection and estimation of generalized chirps using time-frequency representations,

    A. Papandreou, G. Boudreaux-Bartels, and S. Kay, “Detection and estimation of generalized chirps using time-frequency representations,” inProceedings of 1994 28th Asilomar Conference on Signals, Systems and Computers, vol. 1, pp. 50–54 vol.1, Oct. 1994. ISSN: 1058-6393

    1994

  9. [17]

    Time-frequency distributions—a review,

    L. Cohen, “Time-frequency distributions—a review,”Proceedings of the IEEE, vol. 77, no. 7, pp. 941–981, 1989

    1989

  10. [18]

    A class of affine Wigner functions with extended covariance properties,

    J. Bertrand and P. Bertrand, “A class of affine Wigner functions with extended covariance properties,”Journal of Mathematical Physics, vol. 33, pp. 2515–2527, July 1992

    1992

  11. [19]

    On the time–frequency detection of chirps,

    E. Chassande-Mottin and P. Flandrin, “On the time–frequency detection of chirps,”Applied and Computational Harmonic Analysis, vol. 6, pp. 252–281, Mar. 1999

    1999

  12. [20]

    Geometry of affine time-frequency distributions,

    P. Flandrin and P. Goncalv `es, “Geometry of affine time-frequency distributions,”Applied and Computational Harmonic Analysis, vol. 3, pp. 10–39, Jan. 1996

    1996

  13. [21]

    F. W. King,Hilbert Transforms, vol. 1. Cambridge University Press, 2009

    2009

  14. [22]

    Oppenheim and R

    A. Oppenheim and R. Schafer,Discrete-Time Signal Processing. Pear- son, third ed., 2014

    2014

  15. [23]

    H. L. van Trees,Detection, Estimation, and Modulation Theory — Part III. John Wiley & Sons, Ltd, 2001

    2001

  16. [24]

    Y . D. Shirman and V . N. Manzhos,Theory and Techniques of Radar Information Processing against Interference Background. Radio i Svyaz’, 1981. IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. XXX, NO. XXX, AUGUST XXXX 12

    1981

  17. [25]

    The multivariate complex normal distribution—a generalization,

    A. van den Bos, “The multivariate complex normal distribution—a generalization,”IEEE Transactions on Information Theory, vol. 41, pp. 537–539, Mar. 1995. Conference Name: IEEE Transactions on Information Theory

    1995

  18. [26]

    Discrete time and frequency Wigner- Ville distribution: Moyal’s formula and aliasing,

    E. Chassande-Mottin and A. Pai, “Discrete time and frequency Wigner- Ville distribution: Moyal’s formula and aliasing,”IEEE Signal Process- ing Letters, vol. 12, pp. 508–511, July 2005. Conference Name: IEEE Signal Processing Letters

    2005

  19. [27]

    Interval Estimation for a Binomial Proportion,

    L. D. Brown, T. T. Cai, and A. DasGupta, “Interval Estimation for a Binomial Proportion,”Statistical Science, vol. 16, pp. 101–133, May

  20. [28]

    C. W. Helstrom,Statistical Theory of Signal Detection. Oxford: Pergamon Press, 2nd ed., 1968

    1968

  21. [29]

    L. A. Vainshtein and D. E. Vakman,Frequency Division in the Theory of Oscillations and Waves. Nauka, 1983

    1983

  22. [30]

    W. B. Davenport and W. L. Root,An Introduction to the Theory of Random Signals and Noise. McGraw-Hill, 1958

    1958

  23. [31]

    C. W. Therrien,Discrete Random Signals and Statistical Signal Pro- cessing. Prentice Hall, 1992

    1992

  24. [32]

    Further decomposition of the Karhunen-Lo `eve series representation of a stationary random process,

    W. Ray and R. Driver, “Further decomposition of the Karhunen-Lo `eve series representation of a stationary random process,”IEEE Transactions on Information Theory, vol. 16, pp. 663–668, Nov. 1970. Conference Name: IEEE Transactions on Information Theory. Vladimir Lenokwas born...

    1970

  25. [2001]

    Publisher: Institute of Mathematical Statistics

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.