arxiv: 2605.09770 · v1 · submitted 2026-05-10 · 💻 cs.NE · eess.SP· q-bio.NC

Recognition: 2 theorem links

· Lean Theorem

Encoding and Decoding Temporal Signals with Spiking Bandpass Wavelets

Jens Egholm Pedersen , Tony Lindeberg , Peter Gerstoft

Authors on Pith no claims yet

Pith reviewed 2026-05-12 02:41 UTC · model grok-4.3

classification 💻 cs.NE eess.SPq-bio.NC

keywords spiking waveletstime-causal waveletsneuromorphic hardwarewavelet framestemporal signal encodingbandpass filterssparsity preservationsignal reconstruction

0 comments

The pith

Spike encoders can be recast as time-causal wavelet frames with quantitative bandwidths and reconstruction bounds.

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

The paper establishes that spike-based encodings of temporal signals can be equivalently represented as time-causal wavelet frames. This connection supplies explicit bandwidth measures and error bounds while retaining the sparse and local character of the spikes. A sympathetic reader would care because it joins probabilistic spiking models to deterministic signal-processing methods, supporting both analysis and direct hardware use. The reformulation is tested on ECG and audio signals, where reconstruction error reaches levels comparable to standard continuous wavelet transforms.

Core claim

We recast spike encoders as time-causal wavelet frames with quantitative bandwidths and reconstruction error bounds. The proposed wavelets preserve the sparsity and locality of spiking representations, with reconstruction up to spike quantization and time discretization. We demonstrate reconstruction on ECG and audio datasets, achieving a normalized RMSE comparable to continuous wavelet transforms. The spiking wavelets map directly to neuromorphic hardware.

What carries the argument

Time-causal bandpass wavelet frames for encoding and decoding temporal signals through spikes.

If this is right

Reconstruction achieves normalized RMSE comparable to continuous wavelet transforms on ECG and audio data.
Sparsity and locality properties of the original spike encodings are retained.
The wavelets support direct implementation on neuromorphic hardware.
Explicit quantitative bandwidths and reconstruction error bounds are supplied.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The link may let frame theory supply stability proofs for spiking systems.
The same recasting could be tested on additional temporal data streams such as video or sensor streams.
Hybrid classical-plus-spiking pipelines become easier to analyze with shared wavelet tools.

Load-bearing premise

Spike-based encodings can be exactly recast as time-causal wavelet frames without extra signal-dependent assumptions or loss of the claimed sparsity and locality.

What would settle it

A temporal signal where spiking-wavelet reconstruction error substantially exceeds continuous-wavelet error after accounting for quantization and discretization.

Figures

Figures reproduced from arXiv: 2605.09770 by Jens Egholm Pedersen, Peter Gerstoft, Tony Lindeberg.

**Figure 2.** Figure 2: Scaled representations of a noisy signal smoothened by Gaussian [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Impulse and frequency responses to a single [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Reconstruction using identity weights (136) (baseline) or best-fit (30) of a boxcar signal using a lowpass filter (K = 0, left) and a lowpass combined with a single bandpass channel DoE wavelet (K = 1, right). We use dt = 0.1, µ = 2 dt and θthr = 0.01. 5 Spiking scale-covariant wavelets We define spiking wavelets by quantizing bandpass wavelet responses as streams of signed spikes, whose dual can be approx… view at source ↗

**Figure 5.** Figure 5: Covariance guarantees for scale-space kernels. Given some signal, [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗

**Figure 6.** Figure 6: Normalized Gram matrices (126) (top rows) and bandpass frequency responses (bottom rows) for K = 10 channels. Each column corresponds to a cascade order: DoE (n = 1), DoT (n = 3 and n = 7). 25 [PITH_FULL_IMAGE:figures/full_fig_p025_6.png] view at source ↗

**Figure 7.** Figure 7: The leaky integrate-and-fire system governed by [PITH_FULL_IMAGE:figures/full_fig_p026_7.png] view at source ↗

**Figure 8.** Figure 8: nRMSE errors versus θthr, which is linear in θthr following (32). Now, given that the bandpass filters are self-similar over scales, a reasonable measure of the bandwidth Ωk of the bandpass filters must obey the following variability over scales Ωk = Ω1 c k−1 , (140) which when inserted into (141) by the sum of a geometric series gives the closed-form expression ∥f(t) − fe(t)∥∞⩽ C θthr Ω1 cK−1 + Ω1 [PITH_… view at source ↗

**Figure 9.** Figure 9: Sample signals from the MIT-BIH ECG and LibriSpeech datasets. The signals have been [PITH_FULL_IMAGE:figures/full_fig_p030_9.png] view at source ↗

read the original abstract

Spike-based encodings are sparse and energy-efficient, but have largely been formulated probabilistically, disconnected from most signal processing literature. We recast spike encoders as time-causal wavelet frames with quantitative bandwidths and reconstruction error bounds. The proposed wavelets preserve the sparsity and locality of spiking representations, with reconstruction up to spike quantization and time discretization. We demonstrate reconstruction on ECG and audio datasets, achieving a normalized RMSE comparable to continuous wavelet transforms. The spiking wavelets map directly to neuromorphic hardware.

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.

Desk Editor's Note private letter to a colleague

The paper recasts spike encoders as time-causal bandpass wavelets with claimed error bounds, but the exact mapping from nonlinear spiking to linear frames looks like it may rest on unstated approximations.

read the letter

The core move is to treat spike-based temporal encoding as a wavelet frame that stays time-causal and gives explicit bandwidths plus reconstruction bounds. That framing is not the usual probabilistic take on spikes, and the authors show it can reconstruct ECG and audio signals with normalized RMSE close to what a standard continuous wavelet transform gets. The neuromorphic hardware angle is a direct payoff: the same structure maps to event-driven chips without extra conversion steps. Those are the concrete pieces that stand out from the abstract and the reported experiments. The sparsity and locality claims are preserved at least in the examples they give, which is useful for anyone already working on sparse temporal codes. What is less clear is whether the recasting is exact or whether it quietly adds signal-dependent choices when picking the mother wavelet or handling the discretization. Spiking is threshold-driven and event-based; wavelets are linear operators. If the paper only achieves the mapping after some smoothing or rate approximation, then the error bounds and the “up to quantization and discretization” qualifier become conditional rather than general. The abstract states the bounds exist but does not sketch the derivation or report how sensitive they are to parameter choices, so a reader cannot yet judge tightness. No obvious circularity or invented entities in the abstract, and the citations appear to ground the wavelet side in established time-causal work. This is the kind of paper that would interest neuromorphic engineers and people who want classical signal-processing tools for spiking hardware. A reader who already knows both literatures will get the most out of it; someone looking for a fully rigorous equivalence proof may come away wanting more steps. The empirical results are straightforward enough that the work deserves a serious referee rather than a desk reject, mainly to check the derivation details and see whether the bounds survive when the assumptions are relaxed.

Referee Report

2 major / 3 minor

Summary. The paper recasts spike encoders as time-causal wavelet frames with quantitative bandwidths and reconstruction error bounds. The proposed wavelets preserve the sparsity and locality of spiking representations, with reconstruction up to spike quantization and time discretization. Reconstruction is demonstrated on ECG and audio datasets with normalized RMSE comparable to continuous wavelet transforms, and the approach maps directly to neuromorphic hardware.

Significance. If the claimed exact equivalence holds generally, this bridges spiking neural encodings with classical wavelet theory, enabling quantitative bandwidth analysis and error bounds for sparse temporal representations. This could advance neuromorphic hardware design and signal processing applications by providing a principled, hardware-mappable framework. The empirical comparability on real datasets is a practical strength, though significance hinges on the generality of the reformulation without hidden assumptions.

major comments (2)

[Wavelet frame construction (main methods section)] The central claim of an exact recast of spike encoders (nonlinear and threshold-driven) as linear time-causal wavelet frames requires explicit derivation showing preservation of sparsity and locality without signal-dependent assumptions or approximations; the abstract's qualification 'up to spike quantization and time discretization' suggests this may be conditional rather than general, directly affecting the reconstruction error bounds.
[Experiments section] Table or figure reporting reconstruction results: normalized RMSE comparability to CWT is stated, but without error-bar reporting, data exclusion criteria, or explicit normalization details, the strength of the empirical support for the bounds cannot be fully assessed.

minor comments (3)

[Abstract] The abstract claims 'quantitative bandwidths' but does not preview their form or values; a brief indication would improve accessibility.
Ensure consistent numbering of all equations and explicit cross-references in the text to aid verification of the frame operator and bounds.
[Introduction] Add citations to standard references on time-causal wavelets and event-based signal processing to better situate the contribution relative to existing literature.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments on our manuscript. We address each major comment point by point below and indicate the revisions we will make.

read point-by-point responses

Referee: [Wavelet frame construction (main methods section)] The central claim of an exact recast of spike encoders (nonlinear and threshold-driven) as linear time-causal wavelet frames requires explicit derivation showing preservation of sparsity and locality without signal-dependent assumptions or approximations; the abstract's qualification 'up to spike quantization and time discretization' suggests this may be conditional rather than general, directly affecting the reconstruction error bounds.

Authors: We appreciate the referee highlighting the need for greater explicitness. The methods section reformulates the spike encoder by equating spike times to level crossings of the continuous wavelet transform using time-causal bandpass filters, yielding a linear frame operator prior to quantization. Sparsity is preserved by the threshold selecting only significant coefficients, and locality by the compact causal support. We agree the derivation can be made more prominent and have added a dedicated theorem in the revised methods that derives the frame bounds and error expressions directly from wavelet admissibility conditions, without further signal-dependent assumptions. The abstract qualification accounts for practical discretization and quantization effects in hardware but does not render the theoretical recast conditional; the bounds incorporate these as additive terms. revision: yes
Referee: [Experiments section] Table or figure reporting reconstruction results: normalized RMSE comparability to CWT is stated, but without error-bar reporting, data exclusion criteria, or explicit normalization details, the strength of the empirical support for the bounds cannot be fully assessed.

Authors: We agree these reporting details are required for full assessment. In the revised experiments section we will augment the reconstruction results with error bars (standard deviation across signal segments), state that all ECG and audio segments were included without exclusion, and specify the normalization as RMSE divided by the RMS amplitude of the original signal. These additions will allow direct evaluation of the reported comparability to continuous wavelet transforms. revision: yes

Circularity Check

0 steps flagged

Reformulation of spike encoders as time-causal wavelet frames shows no load-bearing circularity

full rationale

The central claim is a recasting of existing spike encoders into wavelet frames with claimed preservation of sparsity and locality. No equations or steps in the abstract reduce by construction to fitted inputs, self-citations, or renamed known results. The equivalence is presented as a theoretical mapping grounded in standard wavelet and spiking concepts rather than a self-referential definition. This is consistent with a minor self-citation score at most, with the derivation remaining externally grounded.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that spiking encodings admit an exact wavelet-frame representation; no free parameters or invented entities are introduced in the abstract.

axioms (1)

domain assumption Spike encoders admit an exact representation as time-causal bandpass wavelet frames that preserve sparsity and locality
This is the load-bearing premise invoked to derive quantitative bandwidths and reconstruction bounds.

pith-pipeline@v0.9.0 · 5379 in / 1093 out tokens · 39570 ms · 2026-05-12T02:41:57.409935+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost Jcost uniqueness (washburn_uniqueness_aczel) unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We recast spike encoders as time-causal wavelet frames... DoE (19) and DoT (18)... frame bounds (22) and closed-form reconstruction error bound (32)
IndisputableMonolith/Foundation/ArithmeticFromLogic LogicNat recovery unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

time-causal scale spaces... truncated exponential kernel (7)... LIF (11)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

46 extracted references · 46 canonical work pages · 2 internal anchors

[1]

20347--20351

Scale-covariant spiking wavelets , author=. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP 2026) , pages = "20347--20351", year=

work page 2026
[2]

IEEE Transactions on Signal Processing , volume=

Sampling and reconstruction of bandlimited signals with multi-channel time encoding , author=. IEEE Transactions on Signal Processing , volume=. 2020 , publisher=

work page 2020
[3]

IEEE Transactions on Circuits and Systems , volume=

Perfect recovery and sensitivity analysis of time encoded bandlimited signals , author=. IEEE Transactions on Circuits and Systems , volume=. 2004 , publisher=

work page 2004
[4]

Biccari, Umberto , year = 2025, month = sep, number =. Spiking. arXiv , doi =. 2509.21920 , primaryclass =

work page arXiv 2025
[5]

IEEE Transactions on Communications

Burt, P. J. and Adelson, E. H. , Journal = "IEEE Transactions on Communications", Pages =. The

work page
[6]

and Romberg, J

Candes, E.J. and Romberg, J. and Tao, T. , year = 2006, month = feb, journal =. Robust Uncertainty Principles: Exact Signal Reconstruction from Highly Incomplete Frequency Information , shorttitle =. doi:10.1109/TIT.2005.862083 , urldate =

work page doi:10.1109/tit.2005.862083 2006
[7]

IEEE Transactions on Information Theory , volume =

Decoding by Linear Programming , author =. IEEE Transactions on Information Theory , volume =. doi:10.1109/TIT.2005.858979 , urldate =

work page doi:10.1109/tit.2005.858979 2005
[8]

Model agnostic signal encoding by leaky integrate-and-fire, performance and uncertainty , volume =

Carbajal, Diana and Romero, José Luis , year =. Model agnostic signal encoding by leaky integrate-and-fire, performance and uncertainty , volume =. doi:10.1016/j.acha.2026.101856 , journal =

work page doi:10.1016/j.acha.2026.101856 2026
[9]

International Conference on Machine Learning , pages =

Group equivariant convolutional networks , author =. International Conference on Machine Learning , pages =. 2016 , organization =

work page 2016
[10]

Frontiers in neuroscience , volume =

Spiking autoencoders with temporal coding , author =. Frontiers in neuroscience , volume =. 2021 , publisher =

work page 2021
[11]

1992 , publisher =

Ten Lectures on Wavelets , author =. 1992 , publisher =

work page 1992
[12]

Communications on Pure and Applied Mathematics , author =

An optimal family of exponentially accurate one-bit Sigma-Delta quantization schemes , volume =. Communications on Pure and Applied Mathematics , author =. 2011 , pages =. doi:10.1002/cpa.20367 , number =

work page doi:10.1002/cpa.20367 2011
[13]

IEEE Transactions on Information Theory , volume =

Compressed Sensing , author =. IEEE Transactions on Information Theory , volume =

work page
[14]

2014 , publisher =

Neuronal Dynamics: From Single Neurons to Networks and Models of Cognition , author =. 2014 , publisher =

work page 2014
[15]

Goh, Say Song and Goodman, Tim N. T. and Lee, S. L. , year =. Singular integrals, scale-space and wavelet transforms , volume =. doi:10.1016/j.jat.2013.09.007 , journal =

work page doi:10.1016/j.jat.2013.09.007 2013
[16]

Applied and Computational Harmonic Analysis , volume =

Sampling based on timing: Time encoding machines on shift-invariant subspaces , author =. Applied and Computational Harmonic Analysis , volume =. 2014 , publisher =

work page 2014
[17]

Applied and Computational Harmonic Analysis , author =

Quantized Frame Expansions with Erasures , volume =. Applied and Computational Harmonic Analysis , author =. 2001 , month = may, pages =. doi:10.1006/acha.2000.0340 , number =

work page doi:10.1006/acha.2000.0340 2001
[18]

URLhttps://doi.org/10.1007/BF01456326

Zur. Mathematische Annalen , author =. 1910 , month =. doi:10.1007/BF01456326 , number =

work page doi:10.1007/bf01456326 1910
[19]

Computational Brain & Behavior , author =

Learning Temporal Relationships Between Symbols with. Computational Brain & Behavior , author =. 2025 , month =. doi:10.1007/s42113-024-00230-8 , number =

work page doi:10.1007/s42113-024-00230-8 2025
[20]

James Bradbury and Roy Frostig and Peter Hawkins and Matthew James Johnson and Yash Katariya and Chris Leary and Dougal Maclaurin and George Necula and Adam Paszke and Jake Vander

work page
[21]

Proceedings of the AAAI Conference on Artificial Intelligence , author =

Fully Spiking Variational Autoencoder , volume =. Proceedings of the AAAI Conference on Artificial Intelligence , author =. 2022 , month =. doi:10.1609/aaai.v36i6.20665 , number =

work page doi:10.1609/aaai.v36i6.20665 2022
[22]

Biological Cybernetics , volume =

The structure of images , author =. Biological Cybernetics , volume =. 1984 , publisher =

work page 1984
[23]

Biological Cybernetics , author =

Scale-time , volume =. Biological Cybernetics , author =. 1988 , month = feb, pages =. doi:10.1007/BF00364135 , number =

work page doi:10.1007/bf00364135 1988
[24]

and Gommers, Ralf and Waselewski, Filip and Wohlfahrt, Kai and O'Leary, Aaron , year = 2019, month = apr, journal =

Lee, Gregory R. and Gommers, Ralf and Waselewski, Filip and Wohlfahrt, Kai and O'Leary, Aaron , year = 2019, month = apr, journal =

work page 2019
[25]

Lindeberg, Tony , year = 1994, volume =. Scale-. doi:10.1007/978-1-4757-6465-9 , urldate =

work page doi:10.1007/978-1-4757-6465-9 1994
[26]

Journal of Mathematical Imaging and Vision , volume =

Image matching using generalized scale-space interest points , author =. Journal of Mathematical Imaging and Vision , volume =. doi:10.1007/s10851-014-0541-0 , urldate =

work page doi:10.1007/s10851-014-0541-0
[27]

Journal of Mathematical Imaging and Vision , volume =

Time-causal and time-recursive spatio-temporal receptive fields , author =. Journal of Mathematical Imaging and Vision , volume =

work page
[28]

Biological Cybernetics , volume =

A time-causal and time-recursive scale-covariant scale-space representation of temporal signals and past time , author =. Biological Cybernetics , volume =. 2023 , publisher =

work page 2023
[29]

Time-causal and time-recursive wavelets

Time-causal and time-recursive wavelets , author =. ar X iv:2510.05834

work page internal anchor Pith review Pith/arXiv arXiv
[30]

, number =

Distinctive image features from scale-invariant keypoints , author =. International Journal of Computer Vision , volume =. doi:10.1023/B:VISI.0000029664.99615.94 , urldate =

work page doi:10.1023/b:visi.0000029664.99615.94
[31]

, year = 1989, month = jul, journal =

Mallat, S.G. , year = 1989, month = jul, journal =. A Theory for Multiresolution Signal Decomposition: The Wavelet Representation , shorttitle =. doi:10.1109/34.192463 , urldate =

work page doi:10.1109/34.192463 1989
[32]

Proceedings of the Royal Society of London

Theory of Edge Detection , author =. Proceedings of the Royal Society of London. B. Biological Sciences , volume =. doi:10.1098/rspb.1980.0020 , urldate =

work page doi:10.1098/rspb.1980.0020 1980
[33]

and Werzi, Anna and Lunglmayr, Michael , year =

Moser, Bernhard A. and Werzi, Anna and Lunglmayr, Michael , year =. Integrate-and-Fire from a Mathematical and Signal Processing Perspective , rights =. doi:10.1109/IEEECONF60004.2024.10942672 , journal =

work page doi:10.1109/ieeeconf60004.2024.10942672 2024
[34]

Applied Sciences , author =

Audlet Filter Banks: A Versatile Analysis/Synthesis Framework Using Auditory Frequency Scales , volume =. Applied Sciences , author =. 2018 , month = jan, pages =. doi:10.3390/app8010096 , number =

work page doi:10.3390/app8010096 2018
[35]

Librispeech:

Panayotov, Vassil and Chen, Guoguo and Povey, Daniel and Khudanpur, Sanjeev , booktitle =. Librispeech:. 2015 , organization =

work page 2015
[36]

Nature Communications , volume =

Neuromorphic intermediate representation: A unified instruction set for interoperable brain-inspired computing , author =. Nature Communications , volume =. 2024 , publisher =

work page 2024
[37]

Nature Communications , volume =

Covariant spatio-temporal receptive fields for spiking neural networks , author =. Nature Communications , volume =. 2025 , publisher =

work page 2025
[38]

Proceedings of the National Academy of Sciences of the United States of America , volume =

On variation-diminishing integral operators of the convolution type , author =. Proceedings of the National Academy of Sciences of the United States of America , volume =

work page
[39]

and Mark, R.G

Moody, G.B. and Mark, R.G. , year =. The Impact of the. IEEE Engineering in Medicine and Biology Magazine , volume =. doi:10.1109/51.932724 , urldate =

work page doi:10.1109/51.932724
[40]

Warland, David and Rieke, Fred and van Steveninck, Rob de Ruyter and Bialek, William , year = 1996, month = dec, series =. Spikes:

work page 1996
[41]

Song, Yang and. Score-. doi:10.48550/arXiv.2011.13456 , urldate =. arXiv , keywords =:2011.13456 , primaryclass =

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2011.13456 2011
[42]

IEEE Transactions on Information Theory , volume =

De-Noising by Soft-Thresholding , author =. IEEE Transactions on Information Theory , volume =. doi:10.1109/18.382009 , urldate =

work page doi:10.1109/18.382009
[43]

IEEE Transactions on Circuits and Systems , volume =

Generalized Sampling Expansion , author =. IEEE Transactions on Circuits and Systems , volume =. doi:10.1109/TCS.1977.1084284 , urldate =

work page doi:10.1109/tcs.1977.1084284 1977
[44]

and Telfer, Brian A

Szu, Harold H. and Telfer, Brian A. and Lohmann, Adolf W. , year =. Causal analytical wavelet transform , volume =. Optical Engineering , publisher =. doi:10.1117/12.59911 , number =

work page doi:10.1117/12.59911
[45]

IEEE Transactions on Neural Networks and Learning Systems , author =

LIF and Simplified SRM Neurons Encode Signals Into Spikes via a Form of Asynchronous Pulse Sigma-Delta Modulation. IEEE Transactions on Neural Networks and Learning Systems , author =. 2017 , month = may, pages =. doi:10.1109/TNNLS.2016.2526029 , number =

work page doi:10.1109/tnnls.2016.2526029 2017
[46]

An Autoencoder-based Spiking Representation for Images , rights =

Zhan, Qiugang and Tao, Ran and Wei, Ru and Jiang, Anning , year =. An Autoencoder-based Spiking Representation for Images , rights =. doi:10.1109/ICICML60161.2023.10424792 , publisher =

work page doi:10.1109/icicml60161.2023.10424792 2023