DEEPDSP-AMC v1.0.1

DEEPDSP-AMC v1.0.1

This patch release corrects the right-hand label layout in the repository banner.

Visual correction

• Centers all three architecture labels within their cards.
• Reduces the label size from 18 px to 16 px to preserve safe horizontal padding under GitHub font fallbacks.
• Keeps the existing restrained color animation and ASCII-only English text.
• Adds a repository validation check for label centering and minimum horizontal padding.

Academic artifacts

The report, executed notebook, trained models, evaluation results, demo app, and video are unchanged from v1.0.0. The attached artifacts remain the complete audited project package.

DEEPDSP-AMC v1.0.0

DEEPDSP-AMC v1.0.0

This first public release packages the complete, evidence-backed academic project for automatic six-class digital modulation classification.

Included

• executed 33-cell notebook with 22 executed code cells and 13 embedded figure outputs;
• 44-page A4 academic report with Typst source;
• English Streamlit workbench and compressed trained models;
• 18-feature Random Forest, 81,030-parameter CompactIQCNN, and validation-selected probability fusion;
• fixed metrics, training history, and 1,320 test predictions;
• 15 original report figures plus English README figures and app capture;
• demonstration video, YouTube metadata, and five-minute script;
• deterministic repository validator and GitHub Actions workflow.

Locked test results

Method	Accuracy	Macro F1
Random Forest	58.48%	58.17%
CompactIQCNN	71.82%	72.01%
Hybrid	73.48%	73.40%

The hybrid fusion weight is 0.55. Test accuracy reaches 90.00% at 9 dB.

Scope

Results apply to the documented controlled synthetic dataset and fixed split. This release does not claim evaluation on SDR captures or live radio hardware.

Verification

The source workspace release audit passed notebook execution, report pagination, equations, algorithms, figures, video-script timing, real app-inference markers, package integrity, and Moodle 20 MB limits. The public repository validator independently checks metric drift, prediction-row agreement, PDF extraction, visuals, models, documentation, SVG safety, and repository hygiene.