Skip to content

Compute and publish results

Analysis belongs next to the measurements. CloudSprite does not replace scikit-rf or numpy; it stores the outputs as datasets so the rest of the team can plot them without rerunning your laptop session.

Pattern

  1. Filter the source datasets (parameters/tags).
  2. Load traces in the SDK.
  3. Compute (filter, mixed-mode, correlation, …).
  4. cs.publish(...) into the same project.
  5. Put source and result traces on one notebook.
  6. Opt in to cs.track_script(__file__) when this should be reproducible.

Mixed-mode (differential) from four 2-port files

A 4-port DUT measured as four 2-port S2Ps is a common bench workflow. Store each port-pair as its own dataset and record the pair in a parameter (mixed_mode=p13, p14, p23, p24) plus a shared serial (sn=6).

Before computing:

  • Confirm the port map (which physical connector is P/N in and out). If it is ambiguous, stop and ask — a swapped pair silently inverts mixed-mode signs.
  • Compute SDD, SCC, SDC, SCD (and CMRR if needed) with scikit-rf / the documented mixed-mode formulas.
  • Publish each mixed-mode parameter as its own traces on a result dataset.
  • Notebook: sources + SDD21 (and CMRR) side by side.

Theory lives in the RF-domain pages — start at Mixed-mode S-parameters.

Waveform correlation (lot QC)

For a batch of the same measurement (cable S21, a day’s production):

  1. Align x-axes (interpolate or trim to overlap).
  2. Pairwise Pearson correlation → per-trace similarity.
  3. Flag outliers (a common default is 2σ below the mean of those scores; make the threshold visible).
  4. Average the non-outlier traces.
  5. Publish the average as a dataset and a notebook that contains only the outliers plus the average.

Do not drop outliers from the project. Tag them (outlier) so they remain auditable.

Lineage

cs.track_script() is opt-in. When on, publish can attach the script file and a short run record (what was read/written, SDK version). It is meant for “how did this SDD21 dataset appear?”, not for shipping your entire home directory.

Permissions

Publishing creates a dataset. You need permission to create datasets in that project. Viewers can inspect results others published; they cannot publish.