Congruence np1e Offset

[RobertoDF]

Hi,

I have a doubt abt the offset to apply to the recordings with np1e.
in https://open-ephys.github.io/bonsai-onix1-docs/articles/getting-started/reference.html#fn:1 offset is fixed at 512, however in the load data script it is set dynamically depending on the value of the scaling (itself dependent on the Gain applied):

ap_scalar = 1.2e6 / (2 ** bit_depth) / ap_gain
ap_offset = (2 ** (bit_depth - 1)) * ap_scalar

Which one is right?

Thanks!
R

[RobertoDF]

I also notice now that for np2 here scaling is 2.44141 and on the table 3.0517578125

[cjsha]

Hey @RobertoDF, good question.

The spikeinterface docs scales data in the following manner:

gain = 0.1
offset = -(2 ** (10 - 1)) * gain
recording.set_channel_gains(gain)
recording.set_channel_offsets(offset)

In the example script we provide, we scale data using read_binary which seems to require pre-scaling the offset as it internally executes the same functions set_channel_gains(gain_to_uV) and set_channel_offsets(offset_to_uV) in the same sequence.

        if gain_to_uV is not None:
            self.set_channel_gains(gain_to_uV)


        if offset_to_uV is not None:
            self.set_channel_offsets(offset_to_uV)

Also, if I'm interpreting spikeinterface's code correctly, this line suggests that offset subtraction seems to be applied after gain multiplication (Python follows PEMDAS order of operations).

Have you plotted the traces? I ran this on some test data and the traces seem centered around zero. I want to make sure it does on your end as well in case I'm overlooking something. Also, you can change num_channels = 384 to num_channels = 1 to inspect the one trace's plot more easily imo, this might be helpful.

[cjsha]

It should be 3.05176. Thanks for the heads up.

[RobertoDF]

And np1e ? 1.2e6 / (2 ** bit_depth) / ap_gain or 512?

[cjsha]

The first one. I laid out my reasoning here.

In any case, please plot your data, confirm it's centered around zero, and let me know how it goes.

[RobertoDF]

Here is a problem! It isnt centered even after the proposed scaling and offset.
ap_gain=1000, bit_depth=10

[RobertoDF]

mean seems to be at 553 somehow not 600

[RobertoDF]

I am still a bit puzzled that after the offset the mean is not at 0 but my original question regarding which was the correct answer was answered.

[cjsha]

The offset could be the result of a real voltage difference between the ref and the electrode in question. What are you using for your GND? And are your reference and your GND tied together on your probe like in the attached photo?

First try reducing the offset by altering these kinds of parameters

[RobertoDF]

Ground is connected to Ref like in the pic 👍. I was in external REF