Skip to content

Interactive Experiment

Scenario

A user interactively modifies station layout, observation parameters, and instrument models and immediately sees how the final image products change. The primary goal is conceptual exploration and algorithm intuition, not production-scale processing.

Scope

  • Simulation-based
  • Small datasets
  • Fast feedback (seconds, not minutes)
  • Panel app with sliders, toggles, and dropdown widgets

Out of scope:

  • Calibration (including self-calibration)
  • Large datasets

Layered Event Storming View

This workflow has three layers:

  1. Commands (user intent)
  2. Domain events (facts that happened)
  3. Views (what is displayed)

1. Commands (User Intent)

These map to notebook widgets.

Array configuration

  • Select array layout (example or custom)
  • Add station
  • Remove station
  • Modify station positions

Observation setup

  • Set integration time
  • Set total observation duration
  • Set frequency setup
  • Set phase centre

Sky & instrument

  • Select sky model
  • Select primary beam model
  • Enable / disable corruption models
  • Set noise level
  • Set gain / phase error parameters

Imaging

  • Select weighting scheme (natural / uniform / robust)
  • Set image size and cell size
  • Set deconvolution parameters

2. Domain Events (What Happens)

Configuration

  • Site location specified
  • Array configuration defined
  • Observation setup defined
  • Sky model selected
  • Instrument model defined
  • Primary beam model selected
  • Visibility averaging policy defined

Simulation

  • UVW sampling computed
  • Sky model visibilities predicted
  • Instrument corruptions applied
  • Visibilities averaged

Imaging

  • Visibilities gridded
  • Dirty image generated
  • PSF generated
  • Deconvolution performed
  • Model image generated
  • Residual image generated

Iteration

  • Parameters updated
  • Affected intermediate products invalidated
  • Required products recomputed

3. Views (Projections)

These are derived, read-only views of the domain state.

  • UV coverage displayed
  • Dirty image displayed
  • PSF displayed
  • Model image displayed
  • Residual image displayed
  • Restored image displayed

Views do not trigger domain changes by themselves.

Core Aggregates

Experiment

Represents a single interactive experiment.

Responsibilities:

  • Owns current configuration
  • Tracks derived products and caches
  • Decides what must be recomputed when parameters change

Telescope

Includes site location, station positions, and derived baselines.

Invariants:

  • Station identifiers are unique
  • Positions share a common coordinate frame
  • Baselines are derived solely from station positions

Observation

Includes time and frequency configuration and pointing.

Invariants:

  • Integration time divides observation duration
  • Frequency grid is consistent with channelization
  • Phase centre is well-defined

SkyModel

Defines the true sky used for simulation.

Invariants:

  • Flux units are consistent
  • Spectral models are defined over the observation band

MeasurementModel

Defines corruptions and beam effects.

Invariants:

  • Beam model matches frequency range
  • Corruption parameterization is internally consistent

VisibilitySet

Contains predicted, corrupted, and averaged visibilities.

Invariants:

  • Array shapes are consistent across visibilities, flags, and weights
  • UVW coordinates align with time/baseline axes
  • Units are fixed (e.g. meters, Hz)

DataProducts

Contains image-plane products.

Invariants:

  • Dirty image, PSF, model, and residual share a common grid
  • Residual = dirty − (model ⊗ PSF)

Dependency & Recompute Map

This dependency graph determines cache invalidation when parameters change.

mermaid flowchart LR A[Telescope] --> U[UVW] O[Observation] --> U S[SkyModel] --> Vp[Predicted visibilities] U --> Vp I[Corruptions] --> Vc[Corrupted visibilities] Vp --> Vc O --> Va[Averaged visibilities] Vc --> Va Va --> G[Gridded visibilities] O --> G G --> D[Dirty image] G --> P[PSF] D --> C[Deconvolution] P --> C C --> M[Model image] C --> R[Residual image]

Examples:

  • Changing integration time invalidates averaged visibilities and downstream products.
  • Changing array layout invalidates UVW sampling and everything downstream.
  • Changing beam model invalidates corrupted visibilities and downstream products.
  • Changing imaging weighting invalidates gridded visibilities and image products only.

Notes for Implementation

  • Widgets issue commands; they do not directly run algorithms.
  • The Experiment aggregate coordinates recomputation.
  • Algorithms remain stateless and reusable.
  • The domain model is independent of notebooks and plotting.

This document serves as the architectural anchor for interactive experiments.