Support multiple pulses in chopper cascade

[nvaytet]

Example:

from scippneutron.tof import chopper_cascade
from ess.odin.beamline import choppers
import scipp as sc

source_position = sc.vector([0, 0, 0], unit='m')
disk_choppers = choppers(source_position)

pulse_frequency = sc.scalar(14.0, unit='Hz')
pulse_period = 1 / pulse_frequency

chops = {
    key: chopper_cascade.Chopper(
        distance=(ch.axle_position - source_position).fields.z,
        time_open=ch.time_offset_open(pulse_frequency=pulse_frequency / 4),
        time_close=ch.time_offset_close(pulse_frequency=pulse_frequency / 4),
    )
    for key, ch in disk_choppers.items()
}

frames = chopper_cascade.FrameSequence.from_source_pulse(
    time_min=sc.scalar(0.0, unit='ms'),
    time_max=sc.scalar(4.0, unit='ms'),
    wavelength_min=sc.scalar(0.0, unit='angstrom'),
    wavelength_max=sc.scalar(10.0, unit='angstrom'),
    npulses=3,
    pulse_period=pulse_period,
)
frames = frames.chop(chops.values())

at_sample = frames.propagate_to(sc.scalar(65.0, unit="m"))
at_sample.draw()

[SimonHeybrock]

Should there be source pulses also before the "pulse of interest? Or is that handled by some wrapping you do afterwards?

[nvaytet]

I do some wrapping manually afterwards in essreduce. I thought I would keep it simple here, so that it behaves a little like tof. But I don't mind either way.

[SimonHeybrock]

Its fine as it is, probably. But I do wonder if one could formulate this on a cyclic coord system instead?

[nvaytet]

Maybe something like making the period boundary act like a chopper and chop the polygons as well?

[SimonHeybrock]

What I mean: After every time propagation, do a mod 71 ms or (142 ms if pulse-skipping, etc.). But the polygon math might not be easy to port.

[nvaytet]

Yes I got that. You'd have to do the mod, but as you say you would run into cases where some polygons span across the period boundary.
I was thinking one easy way to split those polygons would be to use the chop mechanism?
You would then of course have to duplicate the pulses to the frame before the current frame.

[SimonHeybrock]

I don't think pulse duplication would be necessary any more, since periodicity takes care of that? Splitting the polygons that cross the boundary is a good suggestion, provided it does not interfere with other code that tries to look at subframes?

[SimonHeybrock]

Not a fan of repeating the implementation-math in the test, but not sure what a better test would be.

[nvaytet]

I could check that the min of frame.bounds().time is zero, and the max is 2 * period + 1ms?
And remove the looping over the frames. Not sure it's better, but it's different.