Opened 16 years ago
Closed 15 years ago
#41 closed task (fixed)
FDR pixel correlation
Reported by: | MatthewWhiting | Owned by: | MatthewWhiting |
---|---|---|---|
Priority: | normal | Milestone: | |
Component: | Searching | Version: | 1.1.7 |
Severity: | normal | Keywords: | |
Cc: |
Description
Investigate whether the factor of two in determining the number of correlated pixels for the FDR method is appropriate.
Also, look at how to manage this number in the case of Faraday Depth cubes, where there is more correlation along the third axis.
Change History (4)
comment:1 Changed 16 years ago by
Status: | new → assigned |
---|
comment:2 Changed 15 years ago by
Version: | 1.1.5 → 1.1.7 |
---|
comment:3 Changed 15 years ago by
From talking with Maxim, yes, older-style correlators in particular produce data that have 2 or even 3 neighbouring channels correlated. It is less of an issue with newer digital correlators (eg. CABB), where the effect is down at the 60dB level.
So it might be worthwhile including a user-adjustable parameter to help this calculation.
comment:4 Changed 15 years ago by
Resolution: | → fixed |
---|---|
Status: | assigned → closed |
Have added the new parameter as described, in [543]. Seems to work appropriately, so closing this ticket.
A suggestion:
Make this factor an input parameter, rather than hard-coding in the factor of 2. This way Enno can enter the dimension of his cubes and it will take that into account. Thus the number of correlated pixels will be beamSize x numChannels.
Upshot: new input parameter FDRnumCorChan, with default of 2? Still need to check whether you expect neighbouring channels to be correlated.