Curation and Labeling

Curator Workflow

workflow

Capture to Screens

The tested RIA Hub path from upload or synthesis through curation, training, packaging, and a running Screens app.

1. Capture or synthesize Start from a local recording, uploaded RF sample, or synthetic signal problem.
2. Curate and label Select source files, slice useful intervals, and attach labels or qualifiers.
3. Inspect readiness Review balance, metadata, empty states, and recommendations before training.
4. Train model Choose a repository, runner, dataset, template, and mode to launch a workflow run.
5. Package app Connect model artifacts into an application pipeline and validate target runtime settings.
6. Run in Screens Start, stop, inspect logs, and view manifest panels in the Screens runtime.

curate

Curate and label

Select source files, slice useful intervals, and attach labels or qualifiers.

Where this appears in RIA Hub

Dataset Manager Curator.

What you need next

Enough labeled slices for the target class or scenario.

Open guide

Curator selects source repositories, branches, and directories, filters candidate recordings, slices useful intervals, applies labels or qualifiers, and tracks task progress.

Curator wizard steps

Open all Collapse all

Select source data

Pick the repository, branch, directory, and candidate recordings.

Filter candidates

Use search and advanced filters to remove irrelevant or incomplete recordings.

Create slices

Mark signal intervals that should become training examples.

Apply labels

Attach the target label and any useful qualifiers.

Review task progress

Wait for the curation task to finish before inspecting the dataset.

Labeling decisions

search Filters Narrow candidate recordings by repository, branch, directory, and metadata.intervals Slices Select useful signal intervals and avoid empty or ambiguous regions.classes Labels Apply modulation, device, scenario, or other task labels.checks Validation Confirm enough examples per class and consistent metadata.

search

Filters

Filtering keeps the wizard focused on examples that can actually support the target label.

Curation Checks

Class balance preview

Example modulation classes before and after curation checks.

• Before
• After

Data table

Series	BPSK	QPSK	8PSK	16QAM
Before	32 slices	18 slices	12 slices	9 slices
After	28 slices	27 slices	25 slices	24 slices

Before training, inspect class counts, empty slices, metadata consistency, and the target label. For a simple modulation classifier, every slice should map to a modulation class that the selected template can consume.

Next steps

• Check training prerequisites — Basic Training Materials explains what a trainable dataset needs before you launch a run.
• Inspect the dataset — Inspect a Dataset walks through the Inspector workflow for verifying class balance and signal statistics.