(matt's comments in bold)

This page presents the continued investigation into the timestream PSD features for the April 2009 XMMJ1230 data. It was previously noted that there is a spike in the timestream PSD's around 20 Hz, which could be affecting the filtering.

We're concerned that the relative amplitude of this signal across the array may be different than the relative amplitude of sky signals. This would skew the flat-fielding, and mean that the spatial template is not subtracting signals from neighboring channels with the correct relative gains.

(Which relative gain did you use for the plots below? Sky cal on planet, or flat-field? obviously its important!)

---

First we look at the timestream PSDs across the array for a given scan. In this example we look at scan 13355 (no one will know what the scan number means - is this december during bad weather?). Figure 1 shows the PSD where the spike at 20 Hz is clear. However, not all channels see this prominent feature, and some channels find additional spikes. Figure 2 shows the timestream for channel 155 with no spike at 20 Hz and figure 3 shows the timestream psd for channel 96 which has a spike at 20 Hz and at 36 Hz.

(These two frequencies aren't very worrying, I guess, since we can filter them away without really affecting our signal band (1-9 Hz, about). People will want to know if there are ever signficant spikes in the 1-9 Hz regime. If there are, is it as bad as 20 Hz? What fraction of the array has spikes visible above 3 sigma in this regime?)

Figure 1.

Figure 2.

Figure 3.

We now compare the amplitude of the PSD peak at 20 Hz to the rms of the PSD between 5 - 30 Hz. Figure 4 shows the distribution of this ratio for scan 13355. Figure 5 shows the distribution of channels with > 3-sigma peak across the array

Figure 4.

Figure 5.

(state a conclusion: obviously this is a big and frightening thing. Next we'll try to evaluate how important it is. People will wonder about the relative badness of this peak from old data that was used successfully (the bullet) and "new" data that we're having trouble with. Did it get worse? If not, why wasn't this noticed before? )

Now we look at all the 78 XMMJ1230 scans and find the percentage of channels with a 3-sigma or greater spike at 20 Hz per scan. The distribution of these percentages is shown in figure 6.

Figure 6.

(State a conclusion)

Note that, for the standard BOA pipeline, the PSD is zeroed below 0.2 Hz and above 20 Hz, so they would not have seen or been affected by these spikes.

---

Effects on filtering

We first compare the effect of the 20 Hz PSD spike on the annular spatial template. We take the channel 9 timestream from the XMMJ1230 scan 14207. We first apply the regular flagging (optically dead channels, glitches, noisy channels) then remove a 20th order poly-secant. We remove an annular template fitted between 2' and 5' and compare the resulting PSD to the same filter but with an initial notch filter at 20 Hz (applied to all channels in the ring, right?). The result is shown in figure 7.

(Which relative gain was used for the angular ring template? If its the planet scan, we don't expect a big effect). How do you know that the 20 Hz spike doesn't screw up the 20th order poly secant? (I don't think it will, but would be good to check).

Figure 7.

The 20 Hz spike does not appear to affect the annular spatial template at other frequencies.

We also apply a 5-pole butterworth low-pass filter with a 10 Hz cut-off to all the XMMJ1230 timestreams and fully process the maps. The result is shown in figure 8.

Figure 8.

To finish (quick look at earlier runs. Same features?)
Final coadd with lowpass filter
correlation with SQUID

*James - there were three things on the list from last week. I don't have my paper with me that lists them, but I remember that one of them was to try the OLD technique of subtracting the array or wedge-based mean (or median) for the spatial template AFTER filtering away the 20 Hz spike. Maybe that method (which logically seems like the best one) really is best (or good), but was being fooled by this 20 Hz signal?

Also, I had thought the signal was above 20 Hz? but your plots show it is right at 20 Hz. It seems like the BOA strategy of zeroing everything above 20 Hz would still leave a lot of signal in place. Do you understand why they don't still have a 20 Hz problem? Were you able to get info on the PSD plots Kautsuv sent?- were the distributions cumulative? you may want to discuss them in your wiki.*

-- JamesKennedy - 11 Dec 2009

---

This topic: APEX_SZ > WebHome > XMMJ1230 > TimestreamPSDSAcrossTheArray Topic revision: r2 - 2009-12-11 - MattDobbs

---