Look into scaling/lowpass-highpass/channel reference

[mscheltienne]

As the last addition to this reader, I wanted to look into the different possible channel references (#5) and different channel units (#19). So, I went into the eego software and exported a file applying the following montage:

• CA_208, 64 channels all filtered between 0.3 and 30 Hz.
• All channels referenced to CPz except:
  • Fp1 referenced to TP7
  • Fpz referenced to F3
  • Fp2 and F7 referenced to TP8
• All "Scale" set to 70 [uV] exceèt:
  • Fz: 70 [V]
  • F4: 10000 [V]
  • F8: 2000 [V]
  • FC5: 2000 [uV]

The dataset was recorded with no cap connected (floating pin) thus the data itself is garbage.

---

• [x] For the filters, I can't find a way to retrieve the information, it seems absent from the CNT file.
• [x] For the references, it seems to work correctly, I will add tests with a similar dataset next week.
• [ ] For the channel unit and scale, the eego software seems to have many options for the Scale and only uV or V for the units. But in the .cnt file, I can't retrieve either. The unit is always uV, and I added the pyeep_get_channel_scale function but which returns the same float number for all channels, number which does not correspond to anything in eego.

@proloyd Could you have a look through this and confirm that (1) I did not mess up the pyeep_get_channel_scale function and (2) that you also can't retrieve the unit/scale information from the attached dataset? Better to have 2 pairs of eyes on this before calling it quit. Thanks!

[proloyd]

@mscheltienne Thanks for bringing the scaling up. I looked into this earlier and reached to the conclusion that we don't need to do anything on our end. libeep C library takes care of the scaling in libeep_get_samples function that we use in pyeep_get_samples_as_buffer or pyeep_get_samples functions. CNT files do save all relevant scale and unit information, and libeep C library reads all those channel specific information as eegchan_s structure. See the definition here. Basically the actual sample values are raw_data_buffer * chan_info.iscale * chan_info.rscale in chan_info.runit. Function libeep_get_samples reads the raw buffer, and then applies the scaling chan_info.iscale * chan_info.rscale, so that the scaled buffer is in unit chan_info.runit. Refer to the buffer_scaled and buffer_scaled variables here. So, we no longer need to track the scaling explicitly in mne.io.rawANT objects, i.e., don't expose the pyeep_get_channel_scale method.

I second you on the absence of filtering information from the CNT file.

[mscheltienne]

Great, thanks for the explanation! Monday, I'll export an actual EEG recording while applying a montage with weird scaling, let's see how it loads back in MNE ;)

[mscheltienne]

OK, I can now confirm that scaling is a non-issue! I exported an actual EEG recording with a couple of surprising scaling in the montage, yielding 'flat' lines in the review mode of the eego software (e.g. 1000uV, 10000 uV, 1000V, 10000V). When loaded in MNE, the signals are correctly scaled ;)