Signal Processing · arXiv:2611.16786 v1 · submitted 12 Nov 2026 · 13 references
RadarML: A Multi-Channel Over-the-Air Dataset of Common Radar Waveforms for ES Evaluation
The paper says
“This paper introduces "RadarML", a publicly available dataset of modulated radar pulses recorded over-the-air across multiple semi-independent receiver channels.”
more of the abstract
The dataset was generated using the UCL ARESTOR platform which is a Radio Frequency System on a Chip (RFSoC) based Software Defined Radio (SDR). The dataset contains a library of around two million waveforms with varying centre frequencies, bandwidths, pulse widths and modulations, with 1,000 examples of each signal variation per receiver channel. The goal of this dataset is to enable comparative analysis of radar modulation classification techniques on real data and stimulate research into multi-channel signal detection and modulation classification methods.
from the abstract
Pith says
Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person
It can be wrong. Challenge this review or reply to it.
Holds up. RadarML is a solid, well-documented dataset that fills a real gap in radar ES. The paper is a data descriptor, not a scientific claim, and it does what it sets out to do. No load-bearing flaws, but the scope is narrow and the generalizability to operational environments is limited. This line is wrong
7 minor
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 13 references say.
7 minor objections
The paper says, at Abstract & §II
Pith says: minor objection 1 of 7
The abstract states 'around two million waveforms' but the dataset description does not give the exact number of parameter combinations. Adding a brief calculation or a table with the total number of pulses per modulation type would improve clarity.
The paper says, at §II, Table 1
Pith says: minor objection 2 of 7
The table lists '4 CP-FSK' as a modulation type. The meaning of '4' is clear from context (four frequencies), but it would be helpful to spell out '4-ary CP-FSK' or add a footnote.
show 5 more
The paper says, at §II, Eq. (4)
Pith says: minor objection 3 of 7
The equation for the tangential chirp appears to have a formatting glitch: 'T an f(t)' should likely be 'f(t)' or 'f_{tan}(t)'. Please check the LaTeX rendering.
The paper says, at §II
Pith says: minor objection 4 of 7
The paper says 'In half of the frequency-modulated examples, the start (f0) and stop (f1) frequencies were swapped to generate the corresponding "down" chirps.' This is good, but it would be useful to state explicitly whether the dataset includes both up and down chirps for all FM types or only for a subset.
The paper says, at §III, Table 3
Pith says: minor objection 5 of 7
The table lists attenuations for ADC2 and ADC4 but not for ADC0. It says 'N/A' for DAC0 and ADC0, which is clear, but consider adding a note that ADC0 has no additional attenuation.
The paper says, at §IV
Pith says: minor objection 6 of 7
The dataset DOI is provided. It would be helpful to also mention the approximate total size of the dataset (e.g., in GB) so users can plan downloads.
The paper says, at General
Pith says: minor objection 7 of 7
The paper would benefit from a brief discussion of limitations, such as the controlled anechoic chamber environment, the use of a single transmitter, and the absence of multipath or interference. This would help users assess the dataset's applicability to operational scenarios.
What would settle it
Download the dataset from the provided DOI and verify that a randomly selected subset of files has the stated dimensions [3, 1000, 33600] and np.int16 dtype. Run the provided dataloader_demo.ipynb to confirm that the pulses load and display as described. This would confirm the dataset's basic integrity.
from the skeptic's reading
How the review went
4 readings by deepseek-v4-flash · 13 Nov 2026
- Reader holds up
- Skeptic agrees
- Editor wrote the verdict above
- Referee accept: 7 minor
Not run on this paper: the simulated authors' defence.
Pith did not check this review's objections against the paper.
The review in full
The review in brief, machine-written
Hey,
RadarML is exactly what it says on the tin: a public, multi-channel, over-the-air dataset of ~2 million radar pulses with a wide parameter sweep. That alone is a genuine contribution to the radar ES community, where public datasets are scarce and most work is on synthetic data. The hardware documentation is thorough, the data is raw (no synthetic noise baked in), and they ship code for loading and a baseline classifier. Credit where due: this is reproducible, well-scoped, and fills a gap.
The paper is a data descriptor, so there is no scientific claim to falsify. The baseline VGG13 results are just a demo, not a claim of state-of-the-art performance. That is fine for this kind of paper.
Soft spots: the dataset was recorded in an anechoic chamber at short range with co-located receivers. The authors do not discuss limitations of generalizability to real-world ES scenarios (multipath, interference, non-cooperative emitters). The stress-test note says this is a limitation of scope, not a flaw, and I agree—but it is worth noting because a naive user might over-interpret the dataset as representative of operational environments. Also, the paper leans heavily on previous ARESTOR publications for hardware details; a reader new to the platform may need to chase references. Minor point: the waveform definitions are given but the CP-FSK frequency selection is described as 'chosen at random' without a seed or reproducibility guarantee—this could matter if someone wants to exactly replicate the dataset.
Overall: this is a solid data paper for the radar ES subfield. It does not overclaim, it ships the goods, and it will be useful as a benchmark. I would send it to peer review without hesitation.
Best, [Your name]
The strongest objection
The paper presents a dataset, not a scientific claim that can be falsified. The central claim is that RadarML is a publicly available, multi-channel, over-the-air dataset of radar pulses. The paper provides the dataset DOI, code repository, detailed hardware description, and waveform definitions. The reader's weakest assumption—that the dataset may not generalize to real-world ES scenarios—is acknowledged by the authors implicitly through the controlled chamber setup, but this is a limitation of scope, not a flaw in the dataset's construction or documentation. No internal inconsistency, missing evidence, or unsupported assertion is present. The dataset is what it claims to be: a large, reproducible collection of over-the-air radar pulses with multiple channels. The paper does not make a strong empirical claim that would require statistical validation or comparison to other datasets; it is a data descriptor. Therefore, no load-bearing concern about the central claim can be identified.
The reader's verdict is ACCEPT with high confidence, and the weakest assumption identified is about generalizability to real-world scenarios. This is a reasonable limitation but not a load-bearing concern because the paper does not claim the dataset is representative of all operational environments; it is a controlled, reproducible benchmark. The stress-test finds no internal contradictions, missing evidence, or unsupported claims. The dataset is publicly available, the code is provided, and the hardware setup is thoroughly described. The paper fulfills its stated goal of providing a multi-channel over-the-air dataset. Therefore, no adjustment to the verdict is needed, and the agreement with the reader is complete.
The referee's report
If the dataset is as described, it fills a clear gap: existing public radar waveform datasets are either synthetic or single-channel, whereas RadarML provides multi-channel over-the-air measurements with controlled variation of key parameters. The use of three mesochronous ADC channels with different analogue front-ends introduces realistic channel diversity. The dataset is large (∼2M pulses), well-documented, and publicly released under a permissive license. The accompanying code and baseline classifier lower the barrier for entry. This resource has the potential to become a standard benchmark for electronic surveillance research, analogous to what RadioML did for communications modulation classification.
Pith's reading of the paper (a machine's summary, not the authors')
RadarML is a dataset of radar pulses that were transmitted and received over the air in an anechoic chamber, using a software-defined radio platform called ARESTOR. The dataset contains about two million pulses with different modulations (continuous wave, linear and nonlinear frequency modulation, phase-shift keying, Barker codes) and varying parameters like center frequency, bandwidth, and pulse width. The pulses were recorded simultaneously on three receiver channels with different amplifiers and antennas, providing semi-independent views of the same transmitted signals. The raw digitized samples are stored as NumPy arrays, and the authors provide PyTorch dataloaders, augmentation tools, and a baseline VGG13 classifier to help researchers get started. The goal is to enable reproducible comparison of radar modulation classification methods on real captured data, rather than simulated signals.
Core claim. RadarML is a publicly available dataset of around two million over-the-air radar pulses with varying modulations and parameters, recorded across three semi-independent receiver channels, designed to enable comparative analysis of radar modulation classification techniques.
The premise it rests on. The dataset is representative of real-world radar electronic surveillance scenarios. The pulses were recorded in an anechoic chamber at short range with a single transmitter and three co-located receivers, which may not capture the multipath, interference, and propagation effects of operational environments. The authors acknowledge this implicitly by describing the setup as 'over-the-air' in a controlled chamber, but do not discuss limitations of generalizability.
Check this review
Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.
How often reviews like this one are right
Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.
How often objections from reviews like this one are right: not yet measured
Reviews by people
This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.
The paper
Abstract
This paper introduces "RadarML", a publicly available dataset of modulated radar pulses recorded over-the-air across multiple semi-independent receiver channels. The dataset was generated using the UCL ARESTOR platform which is a Radio Frequency System on a Chip (RFSoC) based Software Defined Radio (SDR). The dataset contains a library of around two million waveforms with varying centre frequencies, bandwidths, pulse widths and modulations, with 1,000 examples of each signal variation per receiver channel. The goal of this dataset is to enable comparative analysis of radar modulation classification techniques on real data and stimulate research into multi-channel signal detection and modulation classification methods.
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.
abstract compiled · 2026-11-13
The authors' main claim, as the classifier picked it out 0.67
“The goal of this dataset is to enable comparative analysis of radar modulation classification techniques on real data and stimulate research into multi-channel signal detection and modulation classification methods.”
How the main claim is stated: asserted 0.95
What the abstract itself states, as probabilities: claims first 0.05 · states limitation 0.08 · human participants 0.02 · controlled comparison 0.03 · randomized assignment 0.01 · states quantitative result 0.90 · compares with prior results 0.04 · mentions code or data availability 0.95
The abstract, sentence by sentence
This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.
| Sentence | Role | The authors' own work | States the main conclusion | Strength |
|---|---|---|---|---|
| This paper introduces "RadarML", a publicly available dataset of modulated radar pulses recorded over-the-air across multiple semi-independent receiver channels. | contribution 1.00 | 0.90 | 0.20 | asserted 1.00 |
| The dataset was generated using the UCL ARESTOR platform which is a Radio Frequency System on a Chip (RFSoC) based Software Defined Radio (SDR). | method 1.00 | 0.62 | 0.02 | asserted 1.00 |
| The dataset contains a library of around two million waveforms with varying centre frequencies, bandwidths, pulse widths and modulations, with 1,000 examples of each signal variation per receiver channel. | contribution 0.67 | 0.89 | 0.05 | asserted 1.00 |
| The goal of this dataset is to enable comparative analysis of radar modulation classification techniques on real data and stimulate research into multi-channel signal detection and modulation classification methods. | future work 0.57 | 0.23 | 0.17 | asserted 0.95 |
What kind of paper the classifier reports this is: dataset or benchmark 1.00 · evidence from observation or experiment 0.99
What was examined, and what was not
Sources read
- arXiv listing abstract v1
Not examined
- full text
- figures and tables
- references and citing sentences
- relations to other papers
Levels run
- P abstract
Not run
- S sections
- T body sentences
- C citations
- R cross-paper relations
How the labels were made
Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 3,650 input tokens.
Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON
Figures (5)
Showing 4 of the paper's 5 figures. Figures are the authors', reproduced from the paper.
References (13)
- [1] Pace, Phillip E , publisher =. Detecting and classifying low probability of intercept radar , year = on Pith
- [2] Peters, Nial and Horne, Colin and Ritchie, Matthew A. , booktitle=. 2022 , volume= on Pith
- [3] Peters, Nial J. and Horne, Colin P. and Amiri, Amin D. and Beasley, Piers and Ritchie, Matthew A. , booktitle=. Modular Multi-Channel RFSoC System Expansion and Array Design , year= on Pith
- [4] Ritchie, M. and Peters, N. and Horne, C. , booktitle=. Joint Active Passive Sensing using a Radio Frequency System-on-a-Chip based sensor , year= on Pith
- [5] Temiz, Murat and Horne, Colin and Peters, Nial J. and Ritchie, Matthew A. and Masouros, Christos , journal=. An Experimental Study of Radar-Centric Transmission for Integrated Sensing and Communications , year= on Pith
- [6] Temiz, Murat and Peters, Nial J. and Horne, Colin and Ritchie, Matthew A. and Masouros, Christos , booktitle=. Radar-Centric ISAC Through Index Modulation: Over-the-air Experimentation and Trade-offs , year= on Pith
- [7] Winter, Robert S. C. and Catherall, Aled and Masouros, Christos and Ritchie, Matthew , booktitle=. Impact and Mitigation of Index Modulation on Radar Quality for a FMCW-Based Communication System , year= on Pith
- [8] Horne, Colin and Peters, Nial J. and Ritchie, Matthew A. , journal=. Classification of LoRa Signals With Real-Time Validation Using the Xilinx Radio Frequency System-on-Chip , year= on Pith
- [9] White, Ryan and Horne, Colin and Ritchie, Matthew A. , booktitle=. An Empirical Evaluation of PDW Extraction Techniques Against Weak LPI Pulsed Radar , year= on Pith
- [10] White, Ryan and Winter, Robert S. C. and Horne, Colin and Ritchie, Matthew A. , booktitle=. Evaluation of YOLO for Automatic Radar Waveform Detection and Classification in Congested EME , year= on Pith
- [11] Analysis of Non Linear Frequency Modulation (NLFM) Waveforms for Pulse Compression Radar , author=. Jurnal Elektronika dan Telekomunikasi , year= on Pith
- [12] Very Deep Convolutional Networks for Large-Scale Image Recognition , author=. 2015 , eprint= on Pith
- [13] 11em plus .33em minus .07em 4000 4000 100 4000 4000 500 `\.=1000 \@makecaption#1#2 \@captype\@IEEEtablestring \@IEEEtabletopskipstrut \@IEEEfigurecaptionsepspace \@tempboxa #1. #2 \@tempboxa > \@tempboxa #1. [t] \@tempboxa#2 to \@tempboxa to \@tempboxa \@captype\@IEEEtablestri... on Pith
Challenges and discussion
Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.
Come back by email
A short list of the week's newly reviewed papers. Nothing is sent until you confirm from your inbox, and every email has a link that stops it. No account.
No comments yet.
Sign in to comment