[Eeglablist] questions about ICA decomposition and component

[Jia Wu]

Hi, I used eeglab toolbox 4.41 to decompose some ERP data, 128 channel,
25 trials, each trials containing 250 frames. So I get 128 component. 

My first question is about component discrimination. When I inspected
the component map, I found difficulty in deciding which was noise and
which was not. According to my knowledge, some components are noise like
eye blink, eye movement, so on the ERP imaging picture presenting all
the 25 trials, some  area would stand out because they had different
color with their neiboughers. That means among the 25 trials, a small
part of trials have peaks while others don't. So I think those peaks are
made by movement or blink. Some components are background noise, so they
don't have an apparent peak or they have similar small peaks throughout
the time duration. My problem is, each component seems to be either
movement noise or background noise. Could that happen? What should I do?

My second question is, the original data is collected using a low-pass
30HZ filter. But after ICA, the rebuilt data contain a lot of singals
higher than 30HZ. Will ICA decomposition induce high frequency noise?

My third question is, will ICA don't work if the original data is
heavily containminated?

Thanks for any comment!
Jia Wu