Background
Field report from a user comparing PiFinder's SQM against a Unihedron SQM-L: the
final values agreed closely (21.44 vs 21.43), but panning from a bright patch of
sky to a dark one showed ~19 at first and took roughly half a minute to settle on
21.44. Because the title bar now always shows an SQM number, there is no cue
that the displayed value is still converging — it reads as a finished
measurement the whole time.
Why it happens
This is the rolling window working as designed, not a photometry fault.
RadiometerAccumulator (python/PiFinder/sqm/radiometer.py:224) publishes the
median of up to 12 radiometer samples inside a 15-second window:
max_samples: int = 12
max_age_seconds: float = 15.0
...
value = float(np.median(values))
A median does not move until more than half the window has been replaced, so
after a slew the published value stays blended across two patches of sky for the
length of the window. At night-time frame cadence that is comfortably the ~30
seconds the user observed.
Proposal
Surface the convergence state in the UI. No new photometry is required —
RadiometerAccumulator.estimate() already computes and returns both signals in
its details dict:
radiometer_samples — how many samples in the window were acceptedradiometer_frame_scatter — np.std() across the window's values
Plumb these through to the UI and render a settling cue while the window is
mixed or scatter is high. Consumers to update:
python/PiFinder/ui/sqm.py — the SQM screen's main valuepython/PiFinder/ui/base.py:47 — RotatingInfoDisplay, the title-bar readout,
which already has a "---" path for the no-data case that a settling state
could reuse
Presentation is open — dimming the value, a small progress pip, or dashing it
until the window is coherent would all work. The requirement is only that a user
can tell "still settling" from "this is the reading".
Supersedes
This was chosen over a separate proposal to carry the measurement's RA/Dec in
state.SQM and blank the reading once the telescope moved more than 10–20°.
That approach needs a shared-state dataclass change across two processes, and it
degrades badly when there is no solve (SQM is deliberately solve-independent, so
pointing may be unknown or IMU-only). Showing convergence directly is a better
solution: it uses data the accumulator already produces, needs no pointing, and
covers every cause of an unsettled window — not just deliberate slews.
Background
Field report from a user comparing PiFinder's SQM against a Unihedron SQM-L: the
final values agreed closely (21.44 vs 21.43), but panning from a bright patch of
sky to a dark one showed ~19 at first and took roughly half a minute to settle on
21.44. Because the title bar now always shows an SQM number, there is no cue
that the displayed value is still converging — it reads as a finished
measurement the whole time.
Why it happens
This is the rolling window working as designed, not a photometry fault.
RadiometerAccumulator(python/PiFinder/sqm/radiometer.py:224) publishes themedian of up to 12 radiometer samples inside a 15-second window:
A median does not move until more than half the window has been replaced, so
after a slew the published value stays blended across two patches of sky for the
length of the window. At night-time frame cadence that is comfortably the ~30
seconds the user observed.
Proposal
Surface the convergence state in the UI. No new photometry is required —
RadiometerAccumulator.estimate()already computes and returns both signals inits details dict:
radiometer_samples— how many samples in the window were acceptedradiometer_frame_scatter—np.std()across the window's valuesPlumb these through to the UI and render a settling cue while the window is
mixed or scatter is high. Consumers to update:
python/PiFinder/ui/sqm.py— the SQM screen's main valuepython/PiFinder/ui/base.py:47—RotatingInfoDisplay, the title-bar readout,which already has a
"---"path for the no-data case that a settling statecould reuse
Presentation is open — dimming the value, a small progress pip, or dashing it
until the window is coherent would all work. The requirement is only that a user
can tell "still settling" from "this is the reading".
Supersedes
This was chosen over a separate proposal to carry the measurement's RA/Dec in
state.SQMand blank the reading once the telescope moved more than 10–20°.That approach needs a shared-state dataclass change across two processes, and it
degrades badly when there is no solve (SQM is deliberately solve-independent, so
pointing may be unknown or IMU-only). Showing convergence directly is a better
solution: it uses data the accumulator already produces, needs no pointing, and
covers every cause of an unsettled window — not just deliberate slews.