Studio 2.2.11: preset matrices and automatic PA datasets

[gaochangw]

Signal Generator now offers a compact bandwidth × modulation preset matrix, multi-preset generation, and a direct path from Virtual PA simulation to a complete dataset. Each capture keeps its own sample rate and length; dataset details switch between captures and download a CSV or a collection ZIP containing metadata and a standalone PA replay script.

• Add 1,186 engineering waveform presets across NR, Wi-Fi 6/7 and custom signals; remove Wi-Fi 8 generation. Document the supported allocations and limits without claiming full air-interface conformance.
• Put Signal setup and Use this signal directly below family selection. Remove the separate pairing step and workflow node. Retain advanced controls and formula highlighting.
• Enable an optional ideal input filter by default; preserve nonlinear PA output spectra. Bound batches, protect authenticated downloads, and preserve tenant isolation and storage quotas.
• Update English and translated UI, documentation, screenshots, API contracts and package versions to 2.2.11.

Validation: 2,184 Python tests passed (2,181 regression tests in the isolated current-version runtime and 3 fresh-wheel packaging tests in the standard dependency runtime). All 256 frontend tests, types/lint/build and strict documentation build passed. Real-backend browser journeys passed at 1366×768, 1920×1080 and 390×844, including unequal-length presets, automatic dataset navigation, subdataset selection and CSV/ZIP downloads. Replay exports reproduce all nine PA models sample-for-sample. Axe reported zero WCAG A/AA violations on the three affected pages. GitHub CI and current-head review gate merge.

[gaochangw]

Automated current-head code review completed for this release; this is not an independent human approval.

Reviewed head: 9706c7d

Checked the new batch schema/rate/storage limits, CSRF and tenant routing, dataset membership and version selection, rollback ownership, download source/hash validation and temporary-file cleanup. Checked frozen PA equations, safe metadata embedding in the standalone export, and preservation of independent sample clocks. The default filter is input-only; nonlinear output and metric evidence remain unfiltered. The preset guide explicitly documents continuous uncoded stimuli and standards-access limits.

Validation: 2,181 regression tests plus 3 fresh-wheel installation/real training tests passed (2,184 total); 256 frontend tests and types/lint/build passed; strict documentation build passed. Real API/browser journeys passed at 1366, 1920 and 390 pixels with unequal capture lengths, CSV/ZIP downloads and automatic navigation. All nine exported PA models reproduced float32 outputs exactly. Axe reported zero WCAG A/AA violations on Generator, PA Library and dataset details; direct child URLs retain the collection selector and ZIP download. Chromium journeys passed under the CI snapshot policy; existing unrelated gallery pixel baselines differ locally and were not rewritten.

No blocking findings remain. Merge is contingent on the full GitHub CI matrix passing; no branch rules are being changed.

[gaochangw]

Automated current-head code review completed for this release; this is not an independent human approval.

Reviewed head: 4208dfd

Checked the new batch schema/rate/storage limits, CSRF and tenant routing, dataset membership and version selection, rollback ownership, download source/hash validation and temporary-file cleanup. Checked frozen PA equations, safe metadata embedding in the standalone export, and preservation of independent sample clocks. The default filter is input-only; nonlinear output and metric evidence remain unfiltered. The preset guide explicitly documents continuous uncoded stimuli and standards-access limits.

Validation: 2,181 regression tests plus 3 fresh-wheel installation/real training tests passed (2,184 total); 256 frontend tests and types/lint/build passed; strict documentation build passed. Real API/browser journeys passed at 1366, 1920 and 390 pixels with unequal capture lengths, CSV/ZIP downloads and automatic navigation. All nine exported PA models reproduced float32 outputs exactly. Axe reported zero WCAG A/AA violations on Generator, PA Library and dataset details; direct child URLs retain the collection selector and ZIP download. Chromium journeys passed under the CI snapshot policy; existing unrelated gallery pixel baselines differ locally and were not rewritten.

Also reviewed the final direct-remove controls on selected preset chips, keyboard deletion of the last preset, stale-export invalidation and stable capture-selection effect dependencies. The affected tests (51), types/lint/build, three real-browser sizes and Axe scans passed again. No blocking findings remain. Merge is contingent on the full GitHub CI matrix passing; no branch rules are being changed.

[gaochangw]

2.2.11 is published and online.

• Source: merged commit 4992442f71d3281e8d1176af8f5f656d629ae458; website PR Publish OpenDPD Studio 2.2.11 opendpd-site#9 pins the same source.
• Final CI: all checks passed. Each Python 3.10–3.13 job passed 2,159 tests (25 conditional skips); local validation passed 2,184. Frontend: 256 tests, 48 browser journeys, web transport and mobile touch checks passed.
• Live HTTPS: multi-preset generation, automatic dataset creation/navigation, subdataset selection, CSV/ZIP downloads and separate-workspace download denial passed.
• Actual CUDA: PA training, DPD training, PA evaluation, DPD inference and ILC/ILA passed on the selected Wi-Fi 7 subdataset. Its sample rate and 4,814 test samples remained correct.
• Both downloaded captures replayed bit-identically with the exported standalone Python script. All nine PA models also passed local replay tests.
• Official PyPI wheel: installed into an isolated environment, bundled GUI became ready, and collection export/replay passed. Three-platform CI separately verified default dependency installation.
• Trial QA workspaces were ended; no queued/running jobs or task GPU containers remain. Prior source/image backups are retained.

Open Studio · Release · PyPI