Dear Ola Badarani- Zahalka

I will endeavor to provide the best possible response, with the understanding that this will be of assistance to you. Furthermore, I hope that I have accurately interpreted your question.

1. When referencing EEG data to the common average, it is essential to exclude channels that are not being recorded, such as P7, P8, TP7, and TP8, if these are affected by the cochlear implant processor. The inclusion of non-functional or noisy channels can result in the skewing of the common average and the introduction of artifacts. It is therefore recommended that these channels be excluded from the re-referencing process.

2. Interpolation may be conducted at various stages, but it is typically advised to interpolate the channels after re-referencing and before ICA. This facilitates the assurance that the ICA decomposition is based on a comprehensive set of channels, which can result in a more precise identification of artifacts. The following represents a typical pipeline:

Preprocessing (filtering, bad channel removal)

Re-reference to the common average (excluding bad channels)

Interpolate bad channels

Run ICA to identify and remove artifacts

3. The ICA process is then applied to identify and remove artifacts. ICA can be run on epoched and baseline-corrected data, which may be preferable for specific types of analyses, particularly when dealing with event-related potentials (ERPs). However, ICA is traditionally applied to continuous raw data, as this allows the algorithm to have more data with which to work, thereby improving its ability to identify and separate components. Should one elect to run ICA on epoched data, it is imperative that the epochs be sufficiently lengthy and encompass a diverse array of data, thus ensuring accurate ICA decomposition.

I must profess a certain lack of familiarity with Hamming window filters, and thus I am hopeful that another individual may be able to provide insight on this matter.

Sincerely,

Florian