ssp

✉️ Mailing List Archive (read-only)
1

I have a question about ssp. Participants in my study have a high rate
of artifacts (they are elderly), and we are exploring using ssp vectors
from a fun in which participants look at a dot of light for one minute
on the experimental condition data. Obviously there are tradeoffs here -
the more you apply vectors the more you probably also lose meaningful
data. My question is: When creating the ssp vectors, is it better to
create them from a longer segment or a shorter segment? Is there a
minimum amount of time that should be used?
Thank you,
Wiltrud Fassbinder

Wiltrud Fassbinder, Ph.D.
Adjunct Assistant Professor
Communication Sciences and Disorders
School of Health and Rehabilitation Science
University of Pittsburgh

Research Associate
Department of Audiology and Speech Pathology
VA Pittsburgh Healthcare Center

2

You want to use as much data as you can, and the data should be as accurate a representation of the noise as possible. SSP, as described in the 1997 Uusitalo paper, works by defining a projection matrix consisting of all the noise in your data and then projecting that noise out. The more accurately your noise is represented in the projection matrix, the better a job SSP will do in de-noising your data. In almost all cases, I think, you're better off with more data. As to the minimum, I don't know of a definitive answer to that, but in our lab we've used data snippets as short as 30 seconds. Granted, at 1kHz, that's still 30,000 data points, but its not a lot of data relative to a 45-minute study.

Hope that helps -

Elli