Understand the Exposure Meter timing for mid-point determination

[awhoward]

The point of this Issue is to determine if we are accurately reporting the start and stop times of Exposure Meter (EM) exposures, which are used to determine the flux-weighted midpoint as a function of wavelength.

During a single exposure with the main spectrometer, the EM makes a series of exposures of its focal plane that are extracted into a time series of coarse spectra in the appropriate KPF L0 file. Each EM exposure (and the resulting extracted spectrum) has an associated start and stop time. There is a ~0.8 sec gap between the end of one exposure and the beginning of the next. Of this 0.8 sec, about 0.25 is used to read a windowed portion of the CCD, which does not have a shutter and is not cleared between exposures.

We have been assuming the flux recorded in an EM exposure arrives between the start and stop times. This is not exactly true because light continues to fall on the Exposure Meter CCD during the 0.8 sec gap. Some of charge is smeared out during the 0.25 sec read time and the rest probably accumulates prior to the official start of the next exposure. The goal of this Issue is to find a better determination of the start and stop times if the reported values are no accurrate. This may involve doing experiments with the exposure meter observing a constant source (perhaps the flat lamp with an ND filter for attenuation) at a range of exposure times. Recorded flux should be proportional to the exposure time.

The above text captures some Slack discussion involving Steve Gibson, Andrew Howard, and Will Deich. Below is a message that Will posted with many important details.

I'm tagging @lgbouma, @bjfultn, @howardisaacson, and @shalverson as interested parties.

---

Expmeter timing: the kpf_expmeter dispatcher runs in a loop in which it sleeps for 4ms, then polls the camera state. It takes about 4ms for the camera to return state, so the actual loop time is roughly 8ms. An SBIG exposure works like so:

• When commanded to start an exposure, the camera passes through some uninteresting-to-us states, such as ‘PreShutter’ and ‘DoShutterOpen’.
• Eventually it reaches ‘Starting’, which kpf_expmeter assigns as the start time, and then ‘Exposing’. Note that if the camera is in the ‘Starting’ state for < 4ms, kpf_expmeter may miss that transition, so it uses the rule that the last poll time prior to seeing ‘Exposing’ is the start time. That will be accurate to within 8ms, and it becomes the FITS keyword DATE-BEG.
• When the camera state transitions to ‘Exposing’, the difference between that timestamp and the start time is logged with an entry, “exposing at x.xxxxx s after T_Starting”. Curiously, this transition delay is usually 16ms, not the expected 8ms. I don’t know where that comes from. The distribution of times is: 15% of transition delays are ~8ms, 69% are 16ms, and 16% are 19-20ms (perhaps due to other activity on kpfserver delaying kpf_expmeter?). There are a tiny handful of exposures (totalling under 1% of all exposures) that are at other delays.
• The time of the first poll in which the camera state transitions out of ‘Exposing’ is recorded as the end time, and becomes DATE-END.
• The difference between DATE-BEG and DATE-END is FITS keyword ELAPSED.
• The requested exposure time is recorded as FITS keyword EXPOSURE. Generally, ELAPSED seems to be about 0.14s larger than EXPOSURE. I do not know where the effective start of exposure actually lies.
• I have previously proposed one way to calibrate the camera would be with an accurately-timed pulsed light source. By adjusting its start phase and pulse width, one could use the observed brightness in expmeter images to infer when the image exposure started, and the effective duration of the exposure. (It would be tedious, certainly, and there is surely a better scheme out there.)

[bjfultn]

@jonzink @bpholden Could you please update this issue with the knowledge you have gained from your various tests on this topic?

[bpholden]

There is a Confluence page that summarizes the tests here:

https://exoplanets.atlassian.net/wiki/spaces/shrek/pages/2695004325/Exposure+Meter+Issues+Calibration+Data+Processing+and+Operational#Exposure-Meter-Photon-Arrival-Times

I am guessing that the difference between the ELAPSED and the EXPOSURE values comes from an erase. The detector is clearly erased before each exposure (the above link gives evidence of that.) We can ask SBIG for what happens during an exposure.

On Tue, Feb 7, 2023 at 6:44 PM BJ Fulton @.***> wrote:

@jonzink https://github.com/jonzink @bpholden https://github.com/bpholden Could you please update this issue with the knowledge you have gained from your various tests on this topic?

— Reply to this email directly, view it on GitHub https://github.com/California-Planet-Search/KPF-Pipeline/issues/346#issuecomment-1421656436, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAJJHGS5KDUYSXG2LRBEDNLWWLM4XANCNFSM6AAAAAARXFIMVE . You are receiving this because you were mentioned.Message ID: @.***>