Yin, Zhiyi, Wen, Zhizhao, Zhang, Ruoxin, Wang, Chao, Xu, Puyang, Jiang, Yifan, Gong, Yusi · Nov 4, 2026
Deep Meets Classical: A CPU-Efficient Decision-Fusion Model for Automatic Modulation Classification
Markdown · JSON · Original source
Zhiyi Yin and coauthors argue that two -feasible approaches to automatic modulation classification have complementary strengths: a compact convolutional is more robust at low signal-to-noise ratio, while a classical higher-order-cumulant feature classifier is more accurate at high signal-to-noise ratio. They combine the models’ class-posteriors with a logistic-regression meta-combiner, calling the resulting model FuseAMC. On RadioML 2016.10a, their held-out results show 55.8% overall accuracy, an improvement over either approach alone; the paper reports positive paired-bootstrap 95% lower bounds for both comparisons. The authors say FuseAMC matches the classical baseline in its strong high-SNR regime and the CNN in its strong low-SNR regime. Their practical argument is that combining the approaches can retain both strengths without requiring hardware acceleration: they report roughly 3 MFLOPs per frame and operation on one CPU core. These results concern the stated dataset and test setup; the abstract does not establish performance on other datasets or in deployed systems.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Nov 4, 2026 (2026-11-04).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-09T08:02:22.371Z. First observed: 2026-11-05T05:00:21.241Z. Retrieved: 2026-11-09T08:01:37.423Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The reported evaluation is on RadioML 2016.10a, with a 44k-frame held-out ; the abstract does not report results on other datasets or deployed systems.
Open question
How would FuseAMC perform on datasets or signal conditions beyond the RadioML 2016.10a held-out ?
Source-stated
The authors report complementary performance: the CNN is more robust at low SNR, while the cumulant classifier is more accurate at high SNR.
the CNN is markedly more robust at low SNR, where hand-crafted statistics collapse toward chance, while the cumulant classifier is more accurate at high SNR
Source-stated
FuseAMC combines the class-posteriors of a residual CNN and a gradient-boosted-tree cumulant classifier using a logistic-regression meta-combiner.
a lightweight decision-fusion model that trains a logistic-regression meta-combiner over the concatenated class-posteriors
Source-stated
On RadioML 2016.10a, the authors report 55.8% overall accuracy and improvements over both the cumulant-feature baseline and the CNN alone.
FuseAMC reaches 55.8% overall accuracy, improving on the best cumulant-feature baseline by 5.1 points and on the CNN alone by 3.2 points
Source-stated
The authors report that the pipeline runs on one CPU core without hardware acceleration.
the whole pipeline trains and runs on one CPU core, targeting general-purpose and embedded processors without hardware acceleration.
Source and contributor context
Yin, Zhiyi, Wen, Zhizhao, Zhang, Ruoxin, Wang, Chao, Xu, Puyang, Jiang, Yifan, Gong, Yusi · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
CPU
The central processing unit: a general-purpose processor that carries out a computer's instructions.
Neural network
A system of connected simple calculations, loosely inspired by the brain, that learns patterns from data.
Held-out set
Data kept aside during training and used only to test how well a model works on examples it has not seen.
Confidence interval
A range of values consistent with the data at a stated level of confidence.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 83928037-ee15-47dd-8a07-2e3d49bca098.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T08:02:22.371Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 03b2635170fc68eb048116afaa346e212062abe2381aa964b8a7a2e3899595b7. Source text hash: e490f249aa759aa8493ddb727d5d0601ce1b84137f54ad6a9f0a4c20ed1ae86c.
Revision and source-check history
Version 1 · 2026-11-09T08:02:22.371Z
Deep Meets Classical: A CPU-Efficient Decision-Fusion Model for Automatic Modulation Classification
Zhiyi Yin and coauthors argue that two -feasible approaches to automatic modulation classification have complementary strengths: a compact convolutional is more robust at low signal-to-noise ratio, while a classical higher-order-cumulant feature classifier is more accurate at high signal-to-noise ratio. They combine the models’ class-posteriors with a logistic-regression meta-combiner, calling the resulting model FuseAMC. On RadioML 2016.10a, their held-out results show 55.8% overall accuracy, an improvement over either approach alone; the paper reports positive paired-bootstrap 95% lower bounds for both comparisons. The authors say FuseAMC matches the classical baseline in its strong high-SNR regime and the CNN in its strong low-SNR regime. Their practical argument is that combining the approaches can retain both strengths without requiring hardware acceleration: they report roughly 3 MFLOPs per frame and operation on one CPU core. These results concern the stated dataset and test setup; the abstract does not establish performance on other datasets or in deployed systems.
Original source
cs.AI · Article
Source details
Original source date: Nov 4, 2026 (2026-11-04).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-09T08:02:22.371Z. First observed: 2026-11-05T05:00:21.241Z. Retrieved: 2026-11-09T08:01:37.423Z.
Limits
The reported evaluation is on RadioML 2016.10a, with a 44k-frame held-out ; the abstract does not report results on other datasets or deployed systems.
Open question
How would FuseAMC perform on datasets or signal conditions beyond the RadioML 2016.10a held-out ?
Source-stated
The authors report complementary performance: the CNN is more robust at low SNR, while the cumulant classifier is more accurate at high SNR.
the CNN is markedly more robust at low SNR, where hand-crafted statistics collapse toward chance, while the cumulant classifier is more accurate at high SNR
Source-stated
FuseAMC combines the class-posteriors of a residual CNN and a gradient-boosted-tree cumulant classifier using a logistic-regression meta-combiner.
a lightweight decision-fusion model that trains a logistic-regression meta-combiner over the concatenated class-posteriors
Source-stated
On RadioML 2016.10a, the authors report 55.8% overall accuracy and improvements over both the cumulant-feature baseline and the CNN alone.
FuseAMC reaches 55.8% overall accuracy, improving on the best cumulant-feature baseline by 5.1 points and on the CNN alone by 3.2 points
Source-stated
The authors report that the pipeline runs on one CPU core without hardware acceleration.
the whole pipeline trains and runs on one CPU core, targeting general-purpose and embedded processors without hardware acceleration.
Source and contributor context
Yin, Zhiyi, Wen, Zhizhao, Zhang, Ruoxin, Wang, Chao, Xu, Puyang, Jiang, Yifan, Gong, Yusi · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
CPU
The central processing unit: a general-purpose processor that carries out a computer's instructions.
Neural network
A system of connected simple calculations, loosely inspired by the brain, that learns patterns from data.
Held-out set
Data kept aside during training and used only to test how well a model works on examples it has not seen.
Confidence interval
A range of values consistent with the data at a stated level of confidence.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 83928037-ee15-47dd-8a07-2e3d49bca098.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T08:02:22.371Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 03b2635170fc68eb048116afaa346e212062abe2381aa964b8a7a2e3899595b7. Source text hash: e490f249aa759aa8493ddb727d5d0601ce1b84137f54ad6a9f0a4c20ed1ae86c.
uncertain · 2026-11-09T08:02:22.371Z · anthropic/claude-opus-5.5
Publication: 83928037-ee15-47dd-8a07-2e3d49bca098
attribution — pass: The draft attributes the argument and results to Zhiyi Yin and coauthors, matching the arXiv author list. Findings stay framed as reported by the authors rather than as independently established facts.
Authors:Zhiyi Yin, Zhizhao Wen, Ruoxin Zhang, Chao Wang, Puyang Xu, Yifan Jiang, Yusi Gong
summary — uncertain: The summary is faithful, but a general reader is never told what the central task is. It does not explain that automatic modulation classification identifies how a radio signal is encoded, or that the abstract ties it to spectrum sensing and identifying signals in cognitive radio. Several key terms also go unexplained: higher-order-cumulant, class-posteriors, logistic-regression meta-combiner, SNR and MFLOPs. Smallest repair: add one plain sentence on what AMC is and the use the abstract gives it, and add a tooltip or short gloss for SNR and the fusion idea. Keep the technical terms themselves unchanged.
Automatic modulation classification (AMC) underpins spectrum sensing and non-cooperative signal identification in cognitive radio
date — pass: originalPublished 2026-11-04 matches the arXiv v1 submission date. The feed date of 2026-11-05 is correctly not used.
[v1] Wed, 4 Nov 2026 04:25:29 UTC (107 KB)
support — pass: The claims match the abstract: complementarity by SNR, the fusion architecture, 55.8% accuracy with gains over both baselines, positive confidence-interval lower bounds, matching each baseline in its strong regime, and single-CPU operation at about 3 MFLOPs per frame. The practical-benefit framing stays as the authors' argument, and the limitation on generalization beyond RadioML 2016.10a is appropriately noted. The question follows from that limitation and does not lead.
FuseAMC reaches 55.8% overall accuracy, improving on the best cumulant-feature baseline by 5.1 points and on the CNN alone by 3.2 points
media — pass: The selected media is the verified direct arXiv abstract page for this paper. It is the original source and the easiest access point.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
The central processing unit: a general-purpose processor that carries out a computer's instructions.
Definition referenceA system of connected simple calculations, loosely inspired by the brain, that learns patterns from data.
Definition referenceData kept aside during training and used only to test how well a model works on examples it has not seen.
Definition referenceA range of values consistent with the data at a stated level of confidence.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivData kept aside during training and used only to test how well a model works on examples it has not seen.
Definition referenceData kept aside during training and used only to test how well a model works on examples it has not seen.
Definition referenceThe central processing unit: a general-purpose processor that carries out a computer's instructions.
Definition referenceA system of connected simple calculations, loosely inspired by the brain, that learns patterns from data.
Definition referenceData kept aside during training and used only to test how well a model works on examples it has not seen.
Definition referenceA range of values consistent with the data at a stated level of confidence.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivData kept aside during training and used only to test how well a model works on examples it has not seen.
Definition referenceData kept aside during training and used only to test how well a model works on examples it has not seen.
Definition reference