pith. sign in

arxiv: 2605.24974 · v1 · pith:JGRDJ4LYnew · submitted 2026-05-24 · 📡 eess.SP

Lattice Modulo Sampling

Yhonatan Kvich , Yonina C. Eldar This is my paper

Pith reviewed 2026-06-30 00:03 UTC · model grok-4.3

classification 📡 eess.SP
keywords modulo samplinglattice theorybandlimited signalsanalog to digital conversionmultidimensional samplingquantizationsignal recoverynormalized second moment
0
0 comments X

The pith

Modulo sampling of bandlimited signals generalizes to arbitrary lattices at the same sampling rate, with improved lattices reducing reconstruction error through lower folded power and better quantization.

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

The authors introduce a lattice-based approach to modulo sampling for multidimensional bandlimited signals. Instead of folding each component separately into a square, the signal is folded according to the geometry of a chosen lattice. Recovery guarantees hold at the identical sampling rate as the conventional method, and algorithms extend accordingly. Lattices with smaller normalized second moments achieve lower mean squared error by reducing the power inside the folded region and by supporting more efficient quantization. Concrete gains appear in low and high dimensions, including a 16.7 percent MSE reduction in two dimensions and 57 percent noise reduction in eight dimensions.

Core claim

Modulo sampling can be performed by folding signals into the Voronoi cell of any lattice rather than the hypercube, with recovery possible at the Nyquist rate for bandlimited signals. The normalized second moment of the lattice controls reconstruction quality via two effects: reduced folded signal power at fixed SNR and lower quantization error with a matched quantizer. Higher-dimensional lattices provide progressively better performance than the hypercube.

What carries the argument

The general lattice modulo folding, which maps the signal into the fundamental domain of an arbitrary lattice instead of the unit hypercube.

Load-bearing premise

The recovery conditions and bandlimited assumptions from the component-wise square modulo case transfer directly to folding by an arbitrary lattice without additional sampling rate or invertibility demands arising from the lattice structure.

What would settle it

An experiment showing that recovery of a bandlimited signal fails when using a non-hypercube lattice at the standard Nyquist sampling rate, or that MSE does not decrease when switching to a lattice with smaller normalized second moment.

Figures

Figures reproduced from arXiv: 2605.24974 by Yhonatan Kvich, Yonina C. Eldar.

Figure 1
Figure 1. Figure 1: Square vs. hexagonal modulo sampling of a 2D BL signal. Original [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
read the original abstract

We propose a lattice-theoretic framework for modulo sampling of multidimensional bandlimited signals. Standard modulo analog-to-digital converters (ADCs) fold the signal component-wise into a square domain, reducing the recovery problem to independent one-dimensional problems. We extend the recovery guarantees to any lattice, requiring the same sampling rate as in the standard component-wise modulo setting. We also extend existing recovery algorithms to the general highdimensional lattice setting. Selecting a lattice with a smaller normalized second moment reduces the reconstruction mean squared error (MSE) through two complementary mechanisms: it lowers the folded signal power, which reduces the absolute noise energy at a fixed signal-to-noise ratio (SNR), and it reduces the quantization error when a matched lattice quantizer is applied. Higher-dimensional lattices offer better second moment compared to the hypercube lattice, with gains that grow substantially with dimension. Instantiating the framework in two dimensions with the hexagonal lattice reduces the MSE relative to the square at the same inradius by 16.7%. Furthermore, simulations on 8-dimensional signals using the E8 lattice to achive 57% in both additive and quantization noise. A topological interpretation connects each folding geometry to a surface whose genus reflects the lattice complexity, and reveals a natural hardware implementation via comparator circuits.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a lattice-theoretic framework for modulo sampling of multidimensional bandlimited signals. It extends recovery guarantees and algorithms from the standard component-wise (hypercube) modulo setting to arbitrary lattices while claiming the same sampling rate suffices, and shows that lattices with smaller normalized second moment reduce reconstruction MSE via lower folded-signal power (at fixed SNR) and lower quantization error with a matched quantizer. Specific gains are reported: 16.7% MSE reduction for the hexagonal lattice versus square in 2D at equal inradius, and 57% reduction in both additive and quantization noise for the E8 lattice in 8D. A topological interpretation relating each lattice to a surface of corresponding genus is sketched, together with a comparator-based hardware realization.

Significance. If the extension of recovery guarantees at unchanged rate and the two MSE-reduction mechanisms are rigorously established, the work would offer a principled way to improve high-dimensional modulo ADCs by importing optimal lattices from coding theory, with gains that increase with dimension. The explicit separation of power-reduction and quantization-error effects, plus the topological/hardware suggestions, would constitute a substantive contribution to signal-processing hardware design.

major comments (2)
  1. [Abstract] Abstract (extension of guarantees paragraph): the central claim that recovery guarantees for bandlimited signals extend to an arbitrary lattice λ at exactly the same sampling rate as the component-wise Z^n case is stated without derivation, proof, or explicit statement of the recovery condition. The allowable inter-sample difference is bounded by the covering radius of λ, which is not invariant under volume-preserving lattice changes; no argument is given showing why the oversampling factor required to keep differences inside the fundamental cell remains constant.
  2. [Abstract] Abstract (simulation claims): the reported 16.7% MSE reduction (hexagonal vs. square, same inradius) and 57% reduction (E8, 8-D) are presented without any description of the signal model, bandwidth, noise variance, number of Monte-Carlo trials, or error bars, rendering the quantitative claims unverifiable from the given text.
minor comments (2)
  1. [Abstract] Typo: 'achive' should read 'achieve'.
  2. [Abstract] The phrase '57% in both additive and quantization noise' is ambiguous; clarify whether this is a relative reduction, an absolute error figure, or a combined metric.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful review and constructive comments. We address each major comment below and will incorporate revisions to improve clarity and verifiability.

read point-by-point responses

  1. Referee: [Abstract] Abstract (extension of guarantees paragraph): the central claim that recovery guarantees for bandlimited signals extend to an arbitrary lattice λ at exactly the same sampling rate as the component-wise Z^n case is stated without derivation, proof, or explicit statement of the recovery condition. The allowable inter-sample difference is bounded by the covering radius of λ, which is not invariant under volume-preserving lattice changes; no argument is given showing why the oversampling factor required to keep differences inside the fundamental cell remains constant.

    Authors: We agree the abstract states the claim concisely without the supporting argument. The full manuscript derives the result in Section 3 by showing that the critical sampling rate is fixed by the lattice volume (identical to the hypercube case) and that the bandlimited property bounds inter-sample differences such that they lie inside the fundamental cell at this rate; the covering radius enters the recovery condition but does not alter the required density because the maximum gradient is controlled by the bandwidth. To address the concern directly, we will revise the abstract to include an explicit statement of the recovery condition and a one-sentence outline of why the rate remains unchanged under volume-preserving changes. revision: yes

  2. Referee: [Abstract] Abstract (simulation claims): the reported 16.7% MSE reduction (hexagonal vs. square, same inradius) and 57% reduction (E8, 8-D) are presented without any description of the signal model, bandwidth, noise variance, number of Monte-Carlo trials, or error bars, rendering the quantitative claims unverifiable from the given text.

    Authors: We acknowledge that the abstract reports the numerical gains without the accompanying simulation parameters. These values are obtained from the Monte-Carlo experiments detailed in Section 5 of the manuscript. We will revise the abstract to add a brief qualifier such as “as verified by simulations of bandlimited signals at the critical rate” and will ensure the main text explicitly lists the signal model, bandwidth, noise level, trial count, and error bars so the claims are fully verifiable. revision: yes

Circularity Check

0 steps flagged

No circularity; extension claims rest on independent lattice-theoretic arguments

full rationale

The provided abstract states an extension of recovery guarantees to arbitrary lattices at the same sampling rate as the component-wise case, plus algorithmic extensions and MSE benefits from lower second-moment lattices. No equations, self-definitions, fitted inputs renamed as predictions, or load-bearing self-citations are exhibited in the text. The central claim is presented as a derived result from lattice geometry rather than presupposed by definition or prior author work that itself assumes the target. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework relies on standard lattice theory and signal processing assumptions without new fitted parameters or invented entities in the abstract.

axioms (1)
  • domain assumption Bandlimited signals can be recovered from samples at Nyquist rate in standard square modulo setting
    Assumed to extend directly to arbitrary lattices at identical rate.

pith-pipeline@v0.9.1-grok · 6388 in / 1278 out tokens · 104241 ms · 2026-06-30T00:03:21.317195+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 3 canonical work pages

  1. [1]

    Restoring lost samples from an oversampled band-limited signal,

    R. Marks, “Restoring lost samples from an oversampled band-limited signal,”IEEE transactions on acoustics, speech, and signal processing, vol. 31, no. 3, pp. 752–755, 1983

    1983

  2. [2]

    Restoring a clipped signal,

    J. S. Abel, “Restoring a clipped signal,” inAcoustics, Speech, and Signal Processing, IEEE International Conference On. IEEE Computer Society, 1991, pp. 1745–1748

    1991

  3. [3]

    J. P. A. P ´erez, S. C. Pueyo, and B. C. L ´opez,Automatic gain control. Springer, 2011

    2011

  4. [4]

    A wide dynamic-range cmos image sensor using self-reset technique,

    D. Park, J. Rhee, and Y . Joo, “A wide dynamic-range cmos image sensor using self-reset technique,”IEEE Electron Device Letters, vol. 28, no. 10, pp. 890–892, 2007

    2007

  5. [5]

    On unlimited sampling and reconstruction,

    A. Bhandari, F. Krahmer, and R. Raskar, “On unlimited sampling and reconstruction,”IEEE Transactions on Signal Processing, vol. 69, pp. 3827–3839, 2020

    2020

  6. [6]

    Modulo sampling of FRI signals,

    S. Mulleti and Y . C. Eldar, “Modulo sampling of FRI signals,”IEEE Access, 2024

    2024

  7. [7]

    Modulo sampling in shift-invariant spaces: Recovery and stability enhancement,

    Y . Kvich and Y . C. Eldar, “Modulo sampling in shift-invariant spaces: Recovery and stability enhancement,”arXiv preprint arXiv:2406.10929, 2024

    work page arXiv 2024

  8. [8]

    Modulo sampling and recovery in shift-invariant spaces,

    ——, “Modulo sampling and recovery in shift-invariant spaces,” in ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2024, pp. 11–15

    2024

  9. [9]

    Modulo sampling and recovery with unknown and time-varying folding parame- ter,

    Y . Kvich, A. Yasar, E. Tasci, R. T. Yazicigil, and Y . C. Eldar, “Modulo sampling and recovery with unknown and time-varying folding parame- ter,” inICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025, pp. 1–5

    2025

  10. [10]

    On the identifiability of sparse vectors from modulo compressed sensing measurements,

    D. Prasanna, C. Sriram, and C. R. Murthy, “On the identifiability of sparse vectors from modulo compressed sensing measurements,”IEEE Signal Processing Letters, vol. 28, pp. 131–134, 2020

    2020

  11. [11]

    Wavelet- based reconstruction for unlimited sampling,

    S. Rudresh, A. Adiga, B. A. Shenoy, and C. S. Seelamantula, “Wavelet- based reconstruction for unlimited sampling,” in2018 IEEE Inter- national Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2018, pp. 4584–4588

    2018

  12. [12]

    Line spectral estimation via unlimited sampling,

    Q. Zhang, J. Zhu, F. Quet al., “Line spectral estimation via unlimited sampling,”IEEE Transactions on Aerospace and Electronic Systems, 2024

    2024

  13. [13]

    Unlimited sampling from theory to practice: Fourier-prony recovery and prototype adc,

    A. Bhandari, F. Krahmer, and T. Poskitt, “Unlimited sampling from theory to practice: Fourier-prony recovery and prototype adc,”IEEE Transactions on Signal Processing, vol. 70, pp. 1131–1141, 2021

    2021

  14. [14]

    A hardware prototype of wideband high-dynamic range analog- to-digital converter,

    S. Mulleti, E. Reznitskiy, S. Savariego, M. Namer, N. Glazer, and Y . C. Eldar, “A hardware prototype of wideband high-dynamic range analog- to-digital converter,”IET Circuits, Devices & Systems, vol. 17, no. 4, pp. 181–192, 2023

    2023

  15. [15]

    Practical mod- ulo sampling: Mitigating high-frequency components,

    Y . Kvich, S. Savariego, M. Namer, and Y . C. Eldar, “Practical mod- ulo sampling: Mitigating high-frequency components,”arXiv preprint arXiv:2501.11330, 2025

    work page arXiv 2025

  16. [16]

    Residual recovery algorithm for modulo sampling,

    E. Azar, S. Mulleti, and Y . C. Eldar, “Residual recovery algorithm for modulo sampling,” inProc. IEEE Int. Conf. Acoust., Speech, Signal Process., 2022, pp. 5722–5726

    2022

  17. [17]

    Lasso-based fast residual re- covery for modulo sampling,

    S. B. Shah, S. Mulleti, and Y . C. Eldar, “Lasso-based fast residual re- covery for modulo sampling,” inICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023, pp. 1–5

    2023

  18. [18]

    mSQUID: Model-based leanred modulo recovery at low sampling rates,

    Y . Kvich, R. Arie, H. Hasan, S. B. Shah, and Y . C. Eldar, “mSQUID: Model-based leanred modulo recovery at low sampling rates,”arXiv preprint arXiv:2510.18729, 2025

    work page arXiv 2025

  19. [19]

    Multidimensional unlimited sampling: A geometrical perspective,

    V . Bouis, F. Krahmer, and A. Bhandari, “Multidimensional unlimited sampling: A geometrical perspective,” in2020 28th European Signal Processing Conference (EUSIPCO). IEEE, 2021, pp. 2314–2318

    2021

  20. [20]

    The surprising benefits of hysteresis in unlimited sampling: Theory, algorithms and experiments,

    D. Florescu, F. Krahmer, and A. Bhandari, “The surprising benefits of hysteresis in unlimited sampling: Theory, algorithms and experiments,” IEEE Transactions on Signal Processing, vol. 70, pp. 616–630, 2022

    2022

  21. [21]

    Unlimited sampling beyond modulo,

    E. Azar, S. Mulleti, and Y . C. Eldar, “Unlimited sampling beyond modulo,”Applied and Computational Harmonic Analysis, vol. 74, p. 101715, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1063520324000927

    2025

  22. [22]

    On unlimited sampling,

    A. Bhandari, F. Krahmer, and R. Raskar, “On unlimited sampling,” in Proc. Int. Conf. Sampling Theory Appl., 2017, pp. 31–35

    2017

  23. [23]

    Fast quantizing and decoding algorithms for lattice quantizers and codes,

    J. H. Conway and N. J. A. Sloane, “Fast quantizing and decoding algorithms for lattice quantizers and codes,”IEEE Transactions on Information Theory, vol. 28, no. 2, pp. 227–232, March 1982

    1982

  24. [24]

    On the voronoi regions of certain lattices,

    ——, “On the voronoi regions of certain lattices,”SIAM Journal on Algebraic Discrete Methods, vol. 5, no. 3, pp. 294–305, 1984

    1984

  25. [25]

    V oronoi regions of lattices, second moments of polytopes, and quantization,

    J. Conway and N. Sloane, “V oronoi regions of lattices, second moments of polytopes, and quantization,”IEEE transactions on information theory, vol. 28, no. 2, pp. 211–226, 2003

    2003

  26. [26]

    J. H. Conway and N. J. A. Sloane,Sphere packings, lattices and groups. Springer Science & Business Media, 2013, vol. 290

    2013

  27. [27]

    Optimization of lattices for quantization,

    E. Agrell and T. Eriksson, “Optimization of lattices for quantization,” IEEE Transactions on Information Theory, vol. 44, no. 5, pp. 1814– 1828, 2002

    2002

  28. [28]

    A lower bound on the average error of vector quantizers (corresp.),

    J. Conway and N. Sloane, “A lower bound on the average error of vector quantizers (corresp.),”IEEE Transactions on Information Theory, vol. 31, no. 1, pp. 106–109, 1985

    1985