Correction of readout delay using bgmap

[satoru99]

Implement readout delay correction from bgmap data

sotodlib/site_pipeline/update_bgmap_delay.py

• Core function is run_bgmap_process
• Analyze bgmap data to fit readout delay from bias step pulse.
• Estimate sample delay (channel_delay) by interpolation among reliable channels.
• Save AxisManager into hdf5 and make ManifestDB of bgmaps
• Make another ManifestDB to map observation to bgmap
• It can be tested by adding the followings into the context yaml:

metadata:
  - db: '{manifestdir}/smurf_detsets/v0/smurf_detset_info_local.sqlite'
    label: smurf
    det_info: true
    on_missing: fail
  - db: '/pscratch/sd/t/takakura/sojp/bgmap_delay/satp1/manifests/bgmap_delay/satp1_bgmap_delay_20260525/bgmap_delay.sqlite'
    label: bgmap_delay
    unpack:
    - bgmap_delay

sotodlib/tod_ops/filters.py

• Modify timeshift filter to be able to get 'dt' by name from AxisManger
• Made fft_apply_filter version like timeconst_filter
• It will be used with the following preprocess config

process_pipe:
    ...
    - name: "fourier_filter"
      skip_on_sim: True
      process:
        filt_function: "iir_filter"
        filter_params:
          invert: True

    - name: "fourier_filter"
      skip_on_sim: True
      process:
        filt_function: "timeconst_filter"
        filter_params:
          timeconst: "det_cal.tau_eff"
          invert: True

    # delay correction!
    - name: "fourier_filter"
      process:
        filt_function: "timeshift_filter"
        filter_params:
          dt: "bgmap_delay.channel_delay"
          invert: true
      skip_on_sim: true

[ykyohei]

Thank you so much for making this PR! This looks great.

If I understand correctly, this pipeline can be run both on-site and off-site (i.e. nersc), so I suggest dropping the 'run_method' and modifying '_main', as well as renaming 'run_update_obs_site' and 'run_update_bgmap_site'.

Just to confirm, is it correct that this can only correct step delay, and that we cannot measure global delay using bgmap?

[ykyohei]

This docstring is wrong

[ykyohei]

bee -> be

[ykyohei]

I suggest to define a constant at the beginning and use it everywhere, for example
PRE_STEP_SAMPLES = 20 # samples before bias step at t=0

[ykyohei]

traceback and fail_msg looks same, and fail_msg is not defined in RunBgmapResult, I suggest to unify these.

[ykyohei]

traceback and fail_msg looks same, and fail_msg is not defined in RunObsResult, I suggest to unify these.

[ykyohei]

Suggested change

	def run_obs_process(cfg: BgmapDelayCfg, obs_id: str) -> RunBgmapResult:
	def run_obs_process(cfg: BgmapDelayCfg, obs_id: str) -> RunObsResult:

[ykyohei]

There are some unused imports
dataclasses.astuple, dataclasses.fields, typing.Any, concurrent.futures.Future

[ykyohei]

I think this can be simplified as follows

Suggested change

	if np.isscalar(dt):
	if invert:
	target *= np.exp(2j * np.pi * freqs * dt)
	else:
	target *= np.exp(-2j * np.pi * freqs * dt)
	return
	assert(len(dt) == len(target)) # safe zip.
	if invert:
	for chdt, dest in zip(dt, target):
	dest *= np.exp(2j * np.pi * freqs * chdt)
	else:
	for chdt, dest in zip(dt, target):
	dest *= np.exp(-2j * np.pi * freqs * chdt)
	if not np.isscalar(dt):
	assert(len(dt) == len(target)) # safe zip.
	dt = dt[:, None]
	if invert:
	target *= np.exp(2j * np.pi * freqs * dt)
	else:
	target *= np.exp(-2j * np.pi * freqs * dt)