MATLINK walkthrough

The toolbox opens OpenDPD Studio itself as its GUI. The MATLINK tab contains MATLAB signal exchange and report delivery. Use Studio's existing Datasets, Experiments, Results and plots throughout the rest of the workflow.

1. Connect from MATLAB

link = opendpd.studio("/path/to/my-pa-workspace", ...
    PythonExecutable="/path/to/python");

opendpd.studio() also works after Python has been configured. You can pass an existing Project:

p = opendpd.openProject("my-pa-workspace");
link = opendpd.studio(p);

Studio opens at MATLINK in the system browser. Selecting the same workspace as an existing Studio session attaches to the same service. Otherwise the SDK starts a local service. The connection remains active without keeping link in the workspace. Supplying p leaves that Project available to your scripts.

MATLINK displays session presence and the variables MATLAB has made available. If multiple MATLAB sessions are connected, choose the session that holds your signals before requesting a transfer. Its connection updates depend on MATLAB processing timer callbacks. A long blocking MATLAB operation can temporarily make the session appear offline; the heartbeat resumes when MATLAB can process callbacks again.

2. Choose a signal source

MATLINK offers two source cards. Offline connection history is hidden. With one online MATLAB session, that session is selected automatically. With several, choose the intended session. A selected session becoming unavailable never silently redirects a transfer to another MATLAB process.

OpenDPD Signal Generator

Select OpenDPD Signal Generator, then Open Signal Generator.
Use Studio's existing waveform presets, bandwidth, sample rate, modulation, sample count and advanced settings. Select one or several captures.
Choose the Virtual PA model and its parameters. The generator produces PA input x; the virtual PA creates output y, which PA training requires. The default is the library's solid-state AM/AM + AM/PM model (rapp-am-pm). Both signals are explicitly synthetic.
Click Generate & prepare experiment. Studio generates the inputs, creates the paired dataset, and asks the connected MATLAB session to save every capture. It waits for MATLAB's acknowledgment before navigating.
The experiment page opens with the generated dataset and raw-v1 selected. Sample rate, bandwidth and spectral segment length are inherited from the dataset. Training defaults come from the existing recipe and detected device. Review the normal experiment steps and start training.

Generated data is saved as an ordinary base-workspace struct named opendpdSignals, or a name with a suffix if that variable already exists:

x = opendpdSignals.captures(1).x;       % complex single column vector
y = opendpdSignals.captures(1).y;      % simulated PA output
fs = opendpdSignals.captures(1).signal.sample_rate_hz;
plot(abs(x), abs(y), '.');

Each capture keeps its own sampling metadata, split boundaries and simulation provenance. Full continuous arrays are preserved, including split guards. MATLINK verifies the stored I/Q hashes before loading. For a batch, every capture is saved; the experiment starts with the first capture. Choose a different dataset in experiment setup to train another capture. Nothing starts training automatically. The normal Signal Generator outside MATLINK retains its existing preview and Virtual PA Library workflow.

If MATLAB is busy, the page waits. If delivery fails, Retry transfer retries saving the existing paired dataset. It does not regenerate its signals. The paired dataset remains registered in Studio even if MATLAB disconnects.

MATLAB workspace

Select dense, nonempty single or double PA input and output vectors from the base workspace. They must be different variables with equal sample counts. Complex samples represent I + 1j*Q; real vectors are I-only. Enter sample rate and occupied bandwidth in MHz, plus measured/synthetic/unknown provenance. The script API uses Hz. An optional unique dataset name can be supplied.

Click Use these variables in an experiment. After MATLAB validates and imports the pair, Studio opens experiment setup with the imported data selected. Import preserves sample order and amplitude, stores float32 I/Q, and performs no normalization or alignment. You enter the PSD segment length (nperseg): the Welch segment of every spectral metric and the interval at which evaluation restarts a model's state. It has no default because it depends on your signal; Studio's generated signals use 512-4096. This path uses 256-sample split guards and one subchannel; use opendpd.importIQ for other settings. Invalid values, unequal lengths and duplicate IDs are reported.

3. Train in Studio

Use Studio's normal experiment flow. PA training creates a surrogate; DPD training references a successful PA experiment. Model choices, configuration, progress and comparisons remain in the existing Studio pages. MATLAB and the browser share one queue. Defaults are starting points, not tuned hyperparameters or a guarantee of model quality.

4. Review and save results

Return to MATLINK and choose an experiment from the searchable list. The preview shows the stored metrics, their evidence type, metric profile and available spectrum. Unavailable metrics retain their status instead of showing zero. Use Open experiment for the full Studio analysis.

Save to MATLAB workspace provides a suggested variable name, which you can edit. Click Save to MATLAB to deliver a struct containing:

The saved primary evaluation report, including metrics and evidence.
configuration: the resolved experiment settings.
plots: available spectrum, time-window and AM–AM/AM–PM plot arrays.
Dataset provenance and software metadata from the report.

Plot arrays retain the stored preview resolution. They are not full capture exports or native MATLAB model weights. For full predistorted waveforms use opendpd.apply or opendpd.runDPD in the script workflow.

report = opendpd_train_pa_gru;  % use the variable name shown by MATLINK
struct2table(report.metrics)
report.configuration.training
report.plots.spectrum

The actual saved variable name is shown after MATLAB confirms delivery. Open in MATLAB opens it in the variable editor. Existing variables are preserved by adding a suffix. Repeating the same completed transfer reuses its variable when it still holds the same report; if it has been cleared or replaced, Send again creates a safe new variable.

For a running experiment, Send when ready queues delivery after successful completion. MATLAB must stay connected and process callbacks. Disconnecting ends pending delivery; the experiment can still finish and be saved after reconnecting. Failed and cancelled experiments do not produce a successful result delivery. Run IDs never need to be copied or typed in this workflow.

Documentation and lifecycle

opendpd.help();         % bundled guide, available offline
opendpd.help("gui");    % this walkthrough
opendpd.disconnect();   % detach MATLAB; training continues

Closing the browser does not disconnect MATLAB. Reopen Studio to return to the same workspace and MATLINK session. Closing MATLAB or calling disconnect ends the session and its pending exchanges. Neither action cancels training. If you passed an existing Project to studio, disconnecting MATLINK leaves that Project usable.

To stop a service started by the SDK explicitly:

p = opendpd.openProject("my-pa-workspace");
opendpd.closeProject(p, StopService=true);

This stops the workspace's service and jobs. Stop a service launched through Studio's own launcher using that launcher.