Pith. sign in

REVIEW 3 major objections 5 minor 17 references

Local Ambiguity Shaping for Doppler-Resilient Sequences Under Spectral and PAPR Constraints

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper proposes two optimization algorithms that shape a sequence's local ambiguity-function sidelobes under spectral and peak-power caps, beating the chirp baseline by more than 10 dB in weighted peak sidelobe level.

desk verdict Good problem and a plausible AM algorithm; but ALaMM's convergence proof has a real gap in the eigenvalue bound. read the letter →

arxiv 2506.01637 v1 pith:PBU6NEGR submitted 2025-06-02 eess.SP

Shi He , Lingsheng Meng , Yao Ge , Yong Liang Guan , David González G. , Zilong Liu This is my paper
classification eess.SP MSC 94A1290C26
keywords Doppler-resilientsequencedesignlocalambiguityfunctionshapingweightedpeaksidelobelevelspectralconstraintsPAPRintegratedsensingandcommunicationmajorization-minimizationalternatingminimization
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

This paper tries to establish a practical recipe for designing Doppler-resilient sequences — the waveforms a radar or integrated sensing and communication system transmits — whose delay-Doppler ambiguity sidelobes are low precisely in the local region of interest, while the sequence still obeys a spectrum guard (low energy in a specified stopband) and a peak-to-average power ratio cap. The authors propose two algorithms: an alternating-minimization scheme built on semidefinite relaxation that minimizes the weighted peak sidelobe level directly, and a much cheaper augmented-Lagrangian majorization-minimization scheme (ALaMM) that minimizes an $\ell_p$ approximation of the same metric. Their experiments with 128-point sequences show both schemes producing more than 10 dB lower weighted peak sidelobe level than the widely used chirp sequence, with stopband attenuation as good as the conventional filtered-polyphase approach. If the claims hold, waveform designers in 6G and integrated sensing and communication get a tunable PAPR-versus-sidelobe knob for robust sensing under spectral coexistence without sacrificing Doppler resilience.

What carries the argument

The load-bearing objects are the local ambiguity metric (1) built from the matrices $U_{k,l}=J_k\,\mathrm{Diag}(p(f_l))$ (delay shift matrix times Doppler phase vector), the spectral cost matrices $F_s=f_s f_s^\dagger$ formed from discrete-time Fourier transform vectors, and — for ALaMM — the quadratic majorizer $x^\dagger\Phi(x^{(t)})x$, whose construction needs closed-form maximum eigenvalues $\lambda_{\max}(\Lambda_l)=\max_k\{w_k a_{k,l}(N-|k|)\}$ and $\lambda_{\max}(L)=N^2$, where $L=\sum_{s\in S}\mathrm{vec}(F_s)\mathrm{vec}(F_s)^\dagger$. The identity $\lambda_{\max}(L)=N^2$ rests on assuming the stopband DTFT vectors are orthogonal, so that $L\,\mathrm{vec}(F_s)=N^2\,\mathrm{vec}(F_s)$. For the AM branch, the working machinery is the biconvex split of the epigraph problem into two coupled semidefinite subproblems solved alternately with the penalty $N^2-\mathrm{Tr}(X_1X_2)$ and converted back to a sequence by a rank-one check on the singular values. The projection $P(v)$ defined in (13) is what enforces the constant-modulus or PAPR constraint inside both algorithms.

What would settle it

Compute the largest eigenvalue of $L=\sum_{s\in S}\mathrm{vec}(F_s)\mathrm{vec}(F_s)^\dagger$ for the paper's settings ($N=128$, 50 evenly spaced stopband frequencies in $[0.1,0.2]$ normalized frequency): if it exceeds $N^2=16384$, then the quadratic surrogate $x^\dagger\Phi(x^{(t)})x$ in problem (12) does not majorize the augmented Lagrangian, and monitoring ALaMM's objective across iterations would show whether the reported convergence actually occurs.

Watch

Extended reading notes

Core claim

The central claim is that problem (2) — minimizing the weighted peak sidelobe level of the local discrete ambiguity function, $\mathrm{WPSL}=\max_{(k,f_l)\in\Gamma} w_k|x^\dagger U_{k,l}x|$, subject to the normalization $x^\dagger x=N$, per-sample modulus caps $|x_n|\le\sqrt{\gamma}$, and spectral caps $x^\dagger F_s x\le U_{\max}$ on $N_f$ stopband bins — can be solved to good engineering accuracy by two algorithms. The AM algorithm recasts the problem in epigraph form as a rank-one-constrained semidefinite program, splits it into two convex subproblems in the autocorrelation matrices $X_1$ and $X_2$, alternates between them with a rank-one penalty $N^2-\mathrm{Tr}(X_1X_2)$, and extracts the final sequence by singular value decomposition; the paper reports superior WPSL from AM at complexity $\mathcal{O}(2(2rL+N_f)N^6)$. The ALaMM algorithm replaces the max by its $\ell_p$-norm approximation ($p=22$ in the experiments), forms an augmented Lagrangian for the spectral constraints, majorizes the objective by a quadratic surrogate $x^\dagger\Phi(x^{(t)})x$ whose largest-eigenvalue factors are taken in closed form, and updates via an alternating projection that enforces the PAPR constraint, accelerated by SQUAREM, at complexity $\mathcal{O}((2rL+N_f)N^2)$. The numerical section claims the optimized unimodular sequences achieve more than 10 dB lower WPSL than the chirp sequence while matching the stopband attenuation of the filtered random polyphase sequence, and that relaxing the PAPR cap from 1 to 3 suppresses WPSL further.

Load-bearing premise

A load-bearing premise set down in Section III-B is that the 50 stopband sample frequencies act as mutually orthogonal probes, which would make one matrix in the ALaMM derivation have largest eigenvalue exactly $N^2$; in the paper's own experiments those probes are not orthogonal, so that exact value does not hold.

Editorial extensions

If this is right

  • A unimodular sequence can carry a deep stopband notch and a shaped local ambiguity function at the same time, something the chirp and filtered-polyphase baselines each achieved separately.
  • The PAPR cap is a tunable design knob: relaxing it from 1 to 3 lowers WPSL further, so a system designer can trade transmitter efficiency against sidelobe suppression on the same algorithm.
  • ALaMM lowers per-iteration complexity from $\mathcal{O}(2(2rL+N_f)N^6)$ to $\mathcal{O}((2rL+N_f)N^2)$, the difference between precomputing waveform tables and redesigning sequences on the fly as the spectral environment changes.
  • Because both methods shape only the local ambiguity zone $\Gamma$ rather than the whole delay-Doppler plane, they specialize naturally to the small delay-Doppler window that matters in high-mobility links.

Reading between the lines

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

  • Editorial extension: replacing the closed-form $\lambda_{\max}(L)=N^2$ with the numerically computed largest eigenvalue of $L$ would keep ALaMM's majorization guarantee on arbitrary, even non-uniform, stopband grids; the paper's experiments do not use such a replacement.
  • Editorial extension: because the AM branch never uses the stopband-orthogonality shortcut, a hybrid that runs ALaMM's cheap iterations from a few AM-certified starting points could deliver both accuracy and speed.
  • Editorial extension: the same WPSL-under-spectrum machinery transfers to MIMO radar and multi-user ISAC, where the objective would be the joint peak of cross-ambiguity sidelobes, with $U_{k,l}$ replaced by the appropriate cross-shift matrices.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper addresses the design of Doppler-resilient sequences with low local ambiguity function sidelobes, measured by a weighted peak sidelobe level (WPSL), under frequency-stopband and peak-to-average-power-ratio (PAPR) constraints. Two algorithms are proposed: an alternating-minimization (AM) method based on semidefinite relaxation and a low-complexity augmented-Lagrangian-assisted majorization-minimization (ALaMM) method. Numerical experiments for N=128, a delay-Doppler zone of k in [-5,5] and normalized Doppler in [-2,2], and a stopband of [0.1,0.2] with 50 bins indicate that both algorithms achieve lower WPSL than a chirp sequence while maintaining stopband attenuation comparable to a filtered random polyphase sequence.

Significance. If the claims hold, the paper provides a useful engineering contribution: a flexible PAPR-controlled sequence design framework for integrated sensing and communication under spectral coexistence. The AM algorithm is a sensible adaptation of existing alternating-minimization ideas from autocorrelation PSL design to local AF WPSL design. The ALaMM algorithm, if its convergence were rigorously established, would offer a practically attractive complexity reduction from O(N^6) to O((2rL+Nf)N^2) per iteration. The numerical results, while limited, show meaningful WPSL suppression and stopband compliance, and the authors provide a link to the optimized sequences, which is a commendable step toward reproducibility.

major comments (3)
  1. [Section III-B, after Eq. (12)] The claim that the DTFT stopband vectors are orthogonal, so that lambda_max(L)=N^2 and L vec(F_s)=N^2 vec(F_s), is false for the reported experimental setup. For the matrix L=sum_s vec(F_s) vec(F_s)^dagger, the Rayleigh quotient with y=vec(F_s)/N gives y^dagger L y = (1/N^2) sum_i |f_i^dagger f_s|^4, which exceeds N^2 whenever any i != s has nonzero inner product. With N=128 and 50 frequency points in [0.1,0.2], these inner products are generally nonzero, so the true largest eigenvalue of L is larger than N^2. Because the majorization in problem (12) relies on lambda_max(L) as an upper bound for the curvature of the quadratic penalty term, using N^2 underestimates the spectral norm and the surrogate may fail to majorize. Consequently, the descent property of Algorithm 2 and the claimed O((2rL+Nf)N^2) complexity are not supported as written. A correct upper bound, or a numerical eigenvalue computation, is required; the latter would alter the complexity claim.
  2. [Section III-A, Algorithm 1] The AM algorithm's convergence to a solution of problem (2) is not established. The penalty reformulation with weight eta and the biconvex decomposition in problems (5)-(6) require a proof that the iterates converge to a feasible rank-one solution of problem (4) as eta and the stopping thresholds are adjusted. As written, the algorithm declares success only when the heuristic conditions in step 10 (sigma1/sigma0 <= epsilon_r) hold, but there is no argument that these conditions are eventually satisfied or that they yield a global or even local optimum of the original nonconvex problem. The central claim that AM solves the WPSL minimization problem therefore rests on an unproven convergence step.
  3. [Section IV, Figures 1-3] The numerical evaluation is limited to a single configuration and does not include error bars, multiple random initializations, or a comparison to the most relevant prior method [12], which addresses a closely related PSL-with-spectral-constraint problem. In particular, the claim that the proposed methods achieve 'significantly lower WPSL while maintaining comparable stopband attenuation' would be strengthened by including a state-of-the-art baseline and reporting the spread of results across random initializations. This comment does not question the positive empirical findings but asks for a more complete comparison to make the stated conclusions load-bearing.
minor comments (5)
  1. [Abstract and Section I] The phrase 'low local Ambiguity Function (AF) sidelobes' should be expanded to 'low weighted peak sidelobe level (WPSL) of the local AF' at first use for consistency with the problem statement.
  2. [Section III-A, Eq. (3)] There is a typo: 'N-dimentional' should be 'N-dimensional'.
  3. [Section III-B, Eq. (12)] The definition of L as sum_s vec(F_s) vec(F_s)^dagger is not explicitly labeled; please add an equation number and a clear statement that lambda_max(L) is being substituted in the definition of P(x^(t)).
  4. [Section IV and Reference [13]] Reference [13] is a FIR filter implementation paper, not a canonical reference for stopband filtering of polyphase sequences; please cite a more specific source or describe the filtering procedure in the text.
  5. [Algorithm 2 and Eq. (14)] The definition of beta in Eq. (14) is presented with a set-builder notation that is syntactically confusing; please rewrite it as a clear scalar root-finding problem.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the algorithms optimize the reported WPSL objective against external baselines, and the contested λmax(L)=N² step is a correctness issue, not a circular reduction.

full rationale

The derivation chain is self-contained against external benchmarks and involves no fitted-input-as-prediction step. The WPSL objective in (1) is the target metric, and both algorithms optimize it: AM via the epigraph and biconvex reformulations (3)-(6), and ALaMM via the l_p surrogate (7) followed by MM-based majorization (12). Reporting the resulting WPSL is a standard optimization outcome, not a disguised prediction, because no parameter is fitted to a subset of the reported results and then presented as an independent forecast. The comparisons are made against external baselines, namely the chirp sequence [5] and the filtered random polyphase sequence [13], and the SDR/MM machinery is inherited from prior work by other groups [8], [9], [10], [12], not from self-citations. The self-cited references [2], [4], [14] provide context such as chirp PSL asymptotics and ambiguity-function bounds, but they are not load-bearing for the algorithmic claims. The questionable assertion in Section III-B that λmax(L)=N² follows from 'orthogonality of DTFT vectors' is a mathematical premise that appears invalid for the reported settings (N=128, Nf=50, Fstop=[0.1,0.2]), and it may undermine the majorization guarantee of ALaMM; however, an incorrect or unsupported premise is a correctness risk, not circularity, because the claim does not reduce to its own input by definition. No circular step can be exhibited from the paper's equations, so the circularity score is 0.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central derivation rests on several unproven or false assumptions, most importantly the DTFT orthogonality used to set lambda_max(L)=N^2 in ALaMM. The free parameters p, eta, rho, and the initialization are chosen by hand and not specified in enough detail to reproduce the exact results.

free parameters (4)
  • p (lp-norm order in ALaMM) = 22
    Chosen in Section IV 'for balanced performance and convergence time'; controls how accurately the lp surrogate in Eq. (7) approximates the WPSL max operator.
  • eta (penalty weight in AM)
    Appears in Algorithm 1 subproblems (5) and (6); no value is given, yet it controls the balance between WPSL reduction and rank-one enforcement.
  • rho (augmented Lagrangian step size in ALaMM)
    Appears in Eq. (9) and Algorithm 2; no value is given, yet it controls the speed and stability of spectral constraint enforcement.
  • Initial sequence x_init
    Both algorithms require x_init as input; the paper does not describe how it is generated, and the final WPSL depends on it because the problems are nonconvex.
assumptions (5)
  • ad hoc to paper The DTFT vectors f_s at the stopband frequencies are orthogonal, giving lambda_max(L)=N^2.
    Invoked in Section III-B to replace an eigenvalue decomposition. False for the reported Nf=50 samples over [0.1,0.2] with N=128, so the MM bound is not guaranteed.
  • ad hoc to paper The penalty reformulation in Algorithm 1 converges to a rank-one solution whose leading eigenvector solves problem (2).
    Algorithm 1 checks the eigenvalue ratio sigma_1/sigma_0 <= epsilon_r, but no theorem shows the biconvex alternating minimization reaches a rank-one or global solution.
  • domain assumption The quadratic surrogate x^H Phi(x(t)) x majorizes the augmented Lagrangian L_rho in problem (11).
    Inherited from [8]-[10], but in this paper the surrogate depends on the unproven lambda_max(L)=N^2, so the majorization property is not established.
  • standard math The alternating projection P(v) in Eq. (13) exactly projects onto the constraint set {||x||^2=N, |x_n|<=sqrt(gamma)}.
    Taken from [7]; this is a standard projection procedure for PAPR-constrained waveform design.
  • domain assumption With p=22, minimizing f(x) in Eq. (7) gives a sequence whose WPSL is close to the true maximum in problem (2).
    The paper chooses p=22 for balanced performance and convergence time, but provides no bound on the gap between the lp approximation and the l-infinity WPSL objective.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Local Ambiguity Shaping for Doppler-Resilient Sequences Under Spectral and PAPR Constraints." pith.science (2026). https://pith.science/paper/PBU6NEGR

@misc{pith2026250601637,
  author       = {Pith},
  title        = {Pith review of: Local Ambiguity Shaping for Doppler-Resilient Sequences Under Spectral and PAPR Constraints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PBU6NEGR}},
  note         = {Machine review of arXiv:2506.01637}
}
read the original abstract

This paper focuses on designing Doppler-resilient sequences with low local Ambiguity Function (AF) sidelobes, subject to certain spectral and Peak-to-Average Power Ratio (PAPR) constraints. To achieve this, we propose two distinctoptimization algorithms: (i) an Alternating Minimization (AM) algorithm for superior Weighted Peak Sidelobe Level (WPSL) minimization, and (ii) a low-complexity Augmented Lagrangian-assisted Majorization Minimization (ALaMM) algorithm with effective WPSL suppression. The proposed schemes hold great potential for sequence design in future 6G and integrated sensing and communication applications, supporting robust sensing under spectral coexistence constraints in high-mobility scenarios.

Figures

Figures reproduced from arXiv: 2506.01637 by the authors.

Figure 2
Figure 2. ESDs of different schemes. normalized stopband interval is Fstop = [0.1, 0.2] with desired attenuation A = 20 dB and Nf = 50 bins in it. The PAPR values are set as 1 and 3 respectively, and p = 22 is chosen for the ALaMM algorithm for balanced performance and convergence time. In [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 1
Figure 1. WPSL of different schemes versus computational time [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. 3D plot of AF over the delay-Doppler plane for (a) filte [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

17 extracted references · 17 canonical work pages

  1. [12]

    Waveform design for optimal PSL under spectral and unimodular constraints via alternating minimization,

    C.-W. Huang, L.-F. Chen, and B. Su, “Waveform design for optimal PSL under spectral and unimodular constraints via alternating minimization,” IEEE Trans. Signal Process. , Jul. 2023

    work page 2023

  2. [1]

    Integrated sensing and communication signals tow ard 5G-A and 6G: A survey,

    Z. Wei, H. Qu, Y . Wang, X. Y uan, H. Wu, Y . Du, K. Han, N. Zhang , and Z. Feng, “Integrated sensing and communication signals tow ard 5G-A and 6G: A survey,” IEEE Internet Things J. , vol. 10, no. 13, pp. 11 068– 11 092, Jan. 2023

    work page 2023

  3. [2]

    Asymptotically optim al sequence sets with low/zero ambiguity zone properties,

    L. Tian, X. Song, Z. Liu, and Y . Li, “Asymptotically optim al sequence sets with low/zero ambiguity zone properties,” IEEE Trans. Inf. Theory , vol. 71, no. 6, pp. 4785–4796, Mar. 2025

    work page 2025

  4. [3]

    A unifying view of OTFS and its many variants,

    Q. Deng, Y . Ge, and Z. Ding, “A unifying view of OTFS and its many variants,” IEEE Commun. Surv. Tuts. , pp. 1–26, Feb. 2025

    work page 2025

  5. [4]

    Flag sequence set de sign for low-complexity delay-Doppler estimation,

    L. Meng, Y . L. Guan, Y . Ge, and Z. Liu, “Flag sequence set de sign for low-complexity delay-Doppler estimation,” IEEE Trans. V eh. Technol., pp. 1–17, 2025

    work page 2025

  6. [5]

    New chirp sequence radar wav eform,

    M. Kronauge and H. Rohling, “New chirp sequence radar wav eform,” IEEE Trans. Aerosp. Electron. Syst. , vol. 50, no. 4, pp. 2870–2877, Dec. 2014

    work page 2014

  7. [6]

    Low ambi guity zone: Theoretical bounds and Doppler-resilient sequence d esign in integrated sensing and communication systems,

    Z. Y e, Z. Zhou, P . Fan, Z. Liu, X. Lei, and X. Tang, “Low ambi guity zone: Theoretical bounds and Doppler-resilient sequence d esign in integrated sensing and communication systems,” IEEE J. Sel. Areas Commun., vol. 40, no. 6, pp. 1809–1822, Mar. 2022

    work page 2022

  8. [7]

    Transmit waveform/rec eive filter design for MIMO radar with multiple waveform constraints,

    L. Wu, P . Babu, and D. P . Palomar, “Transmit waveform/rec eive filter design for MIMO radar with multiple waveform constraints,” IEEE Trans. Signal Process. , vol. 66, no. 6, pp. 1526–1540, Dec. 2017

    work page 2017

Show all 17 references
  1. [8]

    Sequence design to mi nimize the weighted integrated and peak sidelobe levels,

    J. Song, P . Babu, and D. P . Palomar, “Sequence design to mi nimize the weighted integrated and peak sidelobe levels,” IEEE Trans. Signal Process., vol. 64, no. 8, pp. 2051–2064, Dec. 2015

    2015

  2. [9]

    J oint design methods of unimodular sequences and receiving filter s with good correlation properties and Doppler tolerance,

    F. Wang, X.-G. Xia, C. Pang, X. Cheng, Y . Li, and X. Wang, “J oint design methods of unimodular sequences and receiving filter s with good correlation properties and Doppler tolerance,” IEEE Trans. Geosci. Remote Sens. , vol. 61, pp. 1–14, Dec. 2022

    2022

  3. [10]

    Local ambiguity function shaping via unimodular sequence design,

    G. Cui, Y . Fu, X. Y u, and J. Li, “Local ambiguity function shaping via unimodular sequence design,” IEEE Signal Process. Lett. , vol. 24, no. 7, pp. 977–981, May 2017

    2017

  4. [11]

    Minimum PSL s equence set design for MIMO radars via manifold-based optimization,

    H. Zebardast, M. Farhang, and A. Sheikhi, “Minimum PSL s equence set design for MIMO radars via manifold-based optimization,” IEEE Sens. J., vol. 24, no. 13, pp. 20 981–20 988, May 2024

    2024

  5. [13]

    A review: FIR filter impleme ntation,

    M. B. Trimale and Chilveri, “A review: FIR filter impleme ntation,” in Proc. 2017 2nd IEEE Int. Conf. Recent Trends Electron., Inf. Commun. Technol. (RTEICT), May 2017, pp. 137–141

    2017

  6. [14]

    Generalize d Arlery- Tan-Rabaste-Levenshtein lower bounds on ambiguity functi on and their asymptotic achievability,

    L. Meng, Y . L. Guan, Y . Ge, Z. Liu, and P . Fan, “Generalize d Arlery- Tan-Rabaste-Levenshtein lower bounds on ambiguity functi on and their asymptotic achievability,” IEEE Trans. Inf. Theory , pp. 1–12, May 2025

    2025

  7. [15]

    D. A. Harville, Kronecker Products and the V ec and V ech Operators . New Y ork, NY: Springer New Y ork, 2001, pp. 139–159

    2001

  8. [16]

    Simple and globally converg ent methods for accelerating the convergence of any EM algorithm,

    R. V aradhan and C. Roland, “Simple and globally converg ent methods for accelerating the convergence of any EM algorithm,” Scand. J. Stat. , vol. 35, no. 2, pp. 335–353, Feb. 2008

    2008

  9. [17]

    Optimized sequences via proposed algoritms,

    “Optimized sequences via proposed algoritms,” https://github.com/he0001hi/Optimized Sequences

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.