Multi-Bolometer Covariance Matrix
date:30/06/09
Comparison of Covariance Matrix with PSB when Q,U are or are not estimated.
Comment:
The correlation are more intense when Q an U are estimated too and also more spread.
Full Covariance Matrix without estimation of Q and U | Full Covariance Matrix with estimation of Q and U |
| |
date:30/06/09
Full Matrix 45000x45000 with PSB and CrossPol < 1
Comment:
- One can see an anticorrelation between offsets of a pair of PSB. (bolo a and b)
- The correlation is less important between two different pair of bolometer.
Full Covariance Matrix (resolution reduced by a factor 5) |
|
A part of the inverse of Covariance Matrix |
|
A part of the Covariance Matrix |
|
Comment:
One can see that the second block is anti-correlated with the others.
The PSB used for this covariance matrix of offset are: 217-5a, 217-5b, 217-7a, 217-7b. (HPR: /space/dmc/m2db/DATA:bench6_HPR_2048_GALACTIC/217-5a_ptg_CPVtag_noPS)
date:22/06/09
Matrix 45000x45000
- There is a block at a strange position at the bottom of each "main sub-diagonal" !! -> IT IS THE 1-YEAR DELAYED SUBDIAGONAL !!!!
date:02/06/09
Strange Pattern fixed
date:27/05/09
Preliminary results?
CONCLUSION:
- Problem in re-ordering the covariance matrix elements. Seems to be during or after the inversion step!!
- The size of Blocks is 64x64 pixels
- A part of the multi-bolo covariance matrix:
*A part of the inverse of the multi-bolo covariance matrix:
Comment: The square feature is not observable on this matrix. It must appears during reconstruction after inversion by SCalaPack.
*Size of the blocks:
zoom of a part of the covariance matrix |
|
Comment:
The big blocks have a size of roughly 161,5 index on the plot so it corresponds to 161.5*2=323 pixels. (2 is the factor used to decrease the resolution)
But these blocks seems to be constituated of 25 smaller blocks.
The elementary pattern is then 323/5~64 pixels long.
So the results seems to be blocks of 64x64 pixels which is the block's size given to ScalaPack for the matrix inversion.