[Feature Request] expose original units for each channel in BaseRaw

[sappelhoff]

Upon reading a raw data file, MNE-Python converts the units to e.g., Volts (and others, see here).

It would be good to expose the channel wise units as they are in the original data file within the BaseRaw object.

Concretely, we need this in MNE-BIDS, see: mne-tools/mne-bids#81

[agramfort]

you need this for what file format?

[sappelhoff]

for EEG:

• BrainVision (.vhdr / .vmrk / .eeg)
• EDF / BDF
• Neuroscan (.cnt)
• EEGLAB (.set / .fdt)

for MEG it depends a bit. In BIDS, all MEG file formats are accepted (i.e., no restrictions like we have in EEG). So for MEG, we need the requested feature for all file formats, that natively save data in a unit that is different from the one that MNE-Python converts to.

[agramfort]

you need to have this availble in the raw object or just a public function like read_xxx_header than you can use in mne-bids?

[sappelhoff]

both would be fine, but a read_xxx_header public function could potentially return a lot more information and could be helpful for other issues as well.

On the other hand, this would mean many new functions (one per data format), while making the information available in BaseRaw would only need adding a few private lines of code within each data reader in the IO module.

Are you more inclined to having a read_xxx_header function for each data format?

[larsoner]

read_xxx_header would probably require refactoring raw readers and, probably more annoyingly, introducing a bunch of otherwise unnecessary public functions. After all, read_xxx_raw(..., preload=False) is already very close to what the read_xxx_header would do and raw.info is very close to what it would return (though also annoyingly, not quite identical because it would also need to give the original channel types). So I don't see the point of a separate function, it seems cleaner to expose a raw._orig_units or something that is used only by BIDS.

[agramfort]

ok then. Why not for _orig_units attributes. Note that there is also the cals array that scales the data to V so maybe this can he used too.

[sappelhoff]

Listing the lines of the respective data readers, from where the original units could be forwarded to some _orig_units attribute.

BrainVision

mne-python/mne/io/brainvision/brainvision.py

Lines 690 to 697 in f4096ad

	if len(props) < 4:
	props += (u'µV',)
	name, _, resolution, unit = props[:4]
	ch_dict[chan] = name
	ch_names[n] = name
	if resolution == "":
	if not(unit): # For truncated vhdrs (e.g. EEGLAB export)
	resolution = 0.000001

EDF

mne-python/mne/io/edf/edf.py

Line 639 in f4096ad

units = [fid.read(8).strip().decode() for ch in channels]

Neuroscan

... seems to be a bit more complicated than BrainVision and EDF ... no units, but "sensitivity" and "cal" (=calibration?):

mne-python/mne/io/cnt/cnt.py

Lines 207 to 225 in f7e2a95

	for ch_idx in range(n_channels): # ELECTLOC fields
	fid.seek(data_offset + 75 * ch_idx)
	ch_name = read_str(fid, 10)
	ch_names.append(ch_name)
	fid.seek(data_offset + 75 * ch_idx + 4)
	if np.fromfile(fid, dtype='u1', count=1)[0]:
	bads.append(ch_name)
	fid.seek(data_offset + 75 * ch_idx + 19)
	xy = np.fromfile(fid, dtype='f4', count=2)
	xy[1] *= -1 # invert y-axis
	pos.append(xy)
	fid.seek(data_offset + 75 * ch_idx + 47)
	# Baselines are subtracted before scaling the data.
	baselines.append(np.fromfile(fid, dtype='i2', count=1)[0])
	fid.seek(data_offset + 75 * ch_idx + 59)
	sensitivity = np.fromfile(fid, dtype='f4', count=1)[0]
	fid.seek(data_offset + 75 * ch_idx + 71)
	cal = np.fromfile(fid, dtype='f4', count=1)
	cals.append(cal * sensitivity * 1e-6 / 204.8)

EEGLAB

... I can't find a "unit" field in the EEGLAB file structure ... it might be that it's not given and that the units are implicitly thought to all be in microVolts, see: https://sccn.ucsd.edu/pipermail/eeglablist/2014/008004.html

[agramfort]

can you give a try to a PR?