Skip to content

OpenDPD 2.2.4

Studio now separates PSD charts by physical signal-chain position and provides an explicit workflow for generating PA input, simulating a Virtual PA output and creating a paired synthetic dataset.

Changes

  • Independent DPD Input, DPD Output / PA Input and PA Output PSD charts throughout inspection, live training/testing, result review, comparison and publication. Output targets and with/without-DPD baselines remain together at PA Output.
  • Compact per-chart legends, full run-ID keys for comparisons, matching initial dB scales, independent zoom and restored per-position figure ranges. Legacy mixed figures remain readable and export as separate positions.
  • Live DPD previews include the actual predistorted drive captured in the same bounded shadow forward pass as the PA output. PSD values and scoring protocols are unchanged.
  • Signal Generator creates a PA Input Dataset only. Separate input CSV and JSON metadata downloads clearly state that no PA output exists yet.
  • PA Library provides nine mathematical Virtual PAs: linear, Rapp, Rapp with AM/PM, Saleh, memory polynomial, lagging-envelope GMP, GaN trap/thermal, illustrative Doherty and envelope tracking. Linked sliders, numeric inputs and formulas explain parameter effects.
  • Explicit output simulation, separate output/paired CSV and metadata exports, frozen provenance and paired dataset creation before PA model training.
  • A compact expandable workflow diagram separates dataset making from PA/DPD training. Existing paired datasets go straight to PA Training.
  • Updated README, workflow/API/review/visualization guides and real GUI screenshots.

Scope

Virtual PAs are normalized behavioral simulations, not calibrated device models. Generated pairs remain synthetic throughout training and export. Signal presets remain uncoded engineering stimuli; experimental Wi-Fi 8 presets make no standards conformance claim. Independent RF acquisitions, hardware reports and external EVM cross-validation remain separate research acceptance items.

Public sessions retain tenant isolation, sample/size quotas and expiry. Dataset contribution requires an operator-configured identity, an explicit submission and human PR merge review. No email is sent automatically.

Install and use

python -m pip install --upgrade "opendpd[gui]==2.2.4"
opendpd gui

Hosted Studio · Signal Generator · PA Library · PSD guide

Verification and deployment evidence is recorded in the 2.2.4 validation record.