Adam Ginsburg's blog

Articles in the bgps category

• Recovery as a function of bolometer noise part 2

  In the Experiment 7 simulations I was running, I observed greater noise than expected, causing me to question the results of the previous post. I therefore ran Experiment 8, which is the same as Experiment 6 from that post with a larger (320x320) map with a larger step size. The noise recovery remains linear, but the scaling is quite different - a factor of ~4 instead of ~12. The step size is the most likely culprit, since an 8x larger step size should result in sqrt(8)~2.8 worse noise per pixel.

  

  There are some curious / worrisome artifacts that turn up and are evident in the recovery fraction plot. For the low-noise cases, the middle bolometers get totally flagged out because they are over-weighted (by orders of magnitude).

  

  So I'm forced to explore via pyflagger. I will almost certainly need to re-run all experiments after making a change to how weights are computed. Well, it turns out the problem is that those 28 bolos are scaled to zero, even though there is nothing obvious (or even suggestive) in their timestream plots. This is only true when varyrelscale is off. Apparently varying the relative scales leads to a different problem. AHA! The noise is so low that the relative scales are SO well correlated that the signal is enough to cause problems! A plausible solution is therefore no change to the pipeline, but to add minimal (nominal) noise to the relative scales to increase the MAD so that the others don't get flagged out. So I added a 1% variation, which prevented flagging at the scale stage, but there are still some disturbing artifacts in the map:

  

  Unfortunately, this problem requires further examination in detail. Exp 9/10 should probably be gaussians and airys on larger step-size maps, but the solution will require something else, possibly even a change in the pipeline. On the plus side, I think I can re-run experiment 7 with a factor of 4 instead of 12 scaling for the noise and expect it to work.

---

• Recovery as a function of bolometer noise

  One clear simulation result comes from varying the amplitude of the gaussian noise added to each bolometer's timestream for a fixed atmosphere amplitude. The atmosphere amplitude sets the mean value of the atmosphere timestream. Increased bolometer noise results in decreased recovery of the outer rings of the PSF. This is demonstrated in that the peak amplitudes remain constant (as long as they are still recovered) while the aperture sums (in 100" radius apertures) decrease. The simulations were run on a logarithmic spacing from 1e-3 to 1e0. The 1e0 point is missing because the peak flux wasn't recovered by the gaussian fitter. The model maps don't recover the source at all for this run, which is worrisome, since it means there is no iterative process There is a minor concern that some simulations over-recover the peak at high noise, but the effect is at a <1% level so not very worrisome.

  

  From the same set of simulations, I derive the pixel RMS of the map (the noise level) derived from a give individual bolometer noise. The theoretical expectation would be Measured Noise = Input Noise / N(bolometers) if the pixel sampling and the timestream sampling were the same (i.e. there were exactly sqrt(nbolos) hits per pixel). This is not exactly the case, and there are potential additional sources of noise. Nonetheless, the naive theory appears to be good enough in this simulation:

  

  You can ignore the green/blue points in this plot; they just show that the std. dev. around the source is dominated by the source. Additionally, there is a noise floor, probably set by an inability to model the point source when the S/N gets to be ~500, preventing convergence of the iterator at a level better than 0.2%. In short, though, I'm going to use 1/sqrt(nbolos) to determine the appropriate input noise level in the astro simulations.

---

• Astrophysical Signal Modeling

  Here we finally get into the meat of the simulations. The goal is to develop realistic - but arbitrary - astrophysical models to run through simulations. The first step is to figure out what a realistic sky looks like. To this end, I use the HiGal SDP fields, looking only at their power spectra. They are well represented by a power law with α = 3 (shown in the dashed black line below).

  

  Therefore, I've attempted to randomly sample from a similar power law distribution using the following IDL code:

  dimsize = 512    realpower = realpowers[ii]    imagpower = imagpowers[ii]    imagscale = imagscales[ii]    peakamp = 1.0    noise = 0.03    smoothscale = 2.0    smoothkernel = psf_gaussian(npix=512,ndim=2,fwhm=31.2/7.2/2.0,/normalize)    sigma_gp = 128.0 ; sigma-width of the Galactic Plane (can get more accurate value from Cara's paper)    xx = findgen(dimsize) #  replicate(1.0,dimsize)    yy = findgen(dimsize) ## replicate(1.0,dimsize)    rr = sqrt( (xx-255.5)^2 + (yy-255.5)^2 )    realpart = (rr^realpower) * randomn(seed1,[dimsize,dimsize])    imagpart = ((rr*imagscale)^imagpower) * randomn(seed2,[dimsize,dimsize])*complex(0,1)     fakesky = abs(fft(shift(realpart + imagpart,0,0),1))    expweight = exp(-(yy-255.5)^2/(2.0*sigma_gp^2)) ; most power is in the inner plane    fakesky *= peakamp/max(fakesky)    fakesky_sm = convolve(fakesky,smoothkernel)    fakesky_sm = fakesky_sm*expweight    fakesky_sm += randomn(seed3,[dimsize,dimsize]) * noise

  Since Power is the fourier-transform squared, I'm using a power-law of α=1.5 for the "real" part of the sampling. The imaginary part follows a shallower slope to reduce the amount of power in large structures, which didn't look quite right (but maybe I should leave both slopes the same?). With both the same, and without the imaginary part down-scaled, the structure appears too "cloudy" and not "clumpy" enough. But back to that later...

  The peak amplitude is set by re-scaling the map. Ideally, we'd like to see this set by a point source, since that is true in most fields. The noise level should not be included in simulations, but should be used to show the difference between pipeline-leftover noise and gaussian noise on the sky. i.e., what structures disappear when you just add noise, and what structures are removed by the pipeline.

  The PSF is simply to smooth out signals that are removed by the telescope beam. We can replace this with a "real" PSF if and when we've come up with a believable one.

  The noise is added after the smoothing because it should be on a pixel scale rather than a beam scale.

  Here are some example realizations with different power laws:

  

---

• PSF modeling

  There haven't been many posts recently because I've primarily been writing up old results into the v2 paper. The Airy and Gaussian simulations with and without atmosphere seem to have yielded their results. There is a ~5% loss when mapping Airy-disk point sources. This is fine as long as it's quantified; it just means that when calibrating we need to know whether extended sources are similarly truncated. If all sources lose 5% in the pipeline, there will be no net offset. However, the Airy is not necessarily representative of the CSO's PSF. In order to come up with a more reasonable representation of the PSF, I've attempted to fit the *measured* PSF (from James' paper) with an Airy disk. The first sidelobe has a lot more amplitude and is closer to the peak than in an Airy disk. It is also asymmetric, but I'm ignoring that for the moment. To better represent the first sidelobe, I've fitted a modified Airy function. The "modification" is to fit two Airy functions, one to the peak and one to the rest. This is accomplished by setting everything outside the first null to zero in the first function, and everything inside the first null to zero in the second. The centers are the same, but the amplitudes and widths are independent.

  

  The above image shows the best fit Airy and modified Airy, both in log scale on an identical grid. The modeled PSF was then put through the pipeline to see it recovers the pipeline-derived PSF. The radial profile results are below, but note that this is only one particular realization of the simulation + pipeline.

  

  Note that the model matches the first sidelobe in the "observed" PSF pretty well, but both before and after pipeline processing, overpredicts the second sidelobe. There are a few "action items" remaining for this task:

  1. Re-derive the "observed" PSF using V2. Does it vary with epoch? Source?
  2. Find a model that better recovers the observed PSF

  Any thoughts on better models to use? Does this seem like a good idea at all?

---

• Azimuthal Averages

  We noted in discussion yesterday that the aperture sum in the (nearly perfectly recovered) is lower than expected even for a gaussian fit of an airy. A Gaussian approximation to an Airy should recover 93.5% of the flux; 6.5% of the power is in the sidelobes of an Airy disk. We recover about 86-88% of the flux in the input point-source map, or about 5-7% short, despite matching the peak to within 2-3%. This discrepancy is partly - but not completely - explained by a slight negative bowl effect: the Gaussian fit has a higher total flux than the aperture sum. Using the gaussian fit value, the recovery is about 90%. A mere 3% missing flux is not too much of a problem - it could still be that the gaussian is uniformly reduced by bowl effects, or perhaps that some flux is not reincorporated into the map (at a very low level) because it is correlated enough to be PCA-cleaned. In the figure, the top panel is the radial profile of the ds1 data (black), the input data (red), and their gaussian fits (see legend). The circles on the labeled images show the FWHM of the fitted gaussian, indicating a good fit. The apertures used for summation are NOT displayed. The sum aperture has a radius of 90" (6.25 pixels) and the "Noise Aperture" has the same area as the sum aperture.

  

  If we redo the analysis with a smaller aperture, the recovery is better: 95% in the central 45". However, now the airy ring cannot account for any lost flux, so it still looks like there's a 5% loss intrinsic in the pipeline.

  

---

• ds1-ds5 comparisons

  I'm comparing simulated ds1-ds5 to real ds1-ds5 comparison tests. In the simulated tests, I compare the recovered map after 20 iterations with 13 pca components subtracted to the input map. There are figures showing this comparison for ds1 and ds5 images in addition to one showing the comparison between ds1 and ds5. The agreement is pretty much as good as you could ask for. These simulations are the most realistic run yet. They include a simulated atmosphere that is perfectly correlated between all bolometers excepting gaussian noise, but the relative sensitivity of the bolometers is varied.

  

  This is what a 'real' ds1-ds5 comparison looks like. The image shown is a "cross-linked" observation of Uranus with downsampling off and on. Note that downsampling clearly and blatantly smears the source flux.

  

  The same image with "beam location correction" looks no better.

  

  The problem is essentially the same with the individual scan directions:

  

  What is causing this difference?

  • higher-order corrections to the atmosphere calculation?
  • inadequate sampling of the model?
  • "pointing" offsets between the model and the data (note that these are NOT pointing offsets, but they may be "distortion map" offsets)?
  • Other?

  Examining the weights and scales for two individual (real) observations, ds1 followed by ds5, is not particularly telling; there is one additional outlier bolometer flagged out in the ds1 observation, but there is nothing obviously wrong with that bolometer (it may have much lower high-frequency noise than others).

  

  The simulations actually have more discrepant weights, but that doesn't seem to cause any problems:

  

  The timestreams both have similar artifacts:

  

  while the simulated versions really don't:

  

  This is true even when the relative strength of the atmosphere is higher:

  

  I think the most viable candidate is the 'pointing offset' idea, which will take a little work to simulate properly...

---

• It is somewhat clear that sampling isn&#39;t causi...

  It is somewhat clear that sampling isn't causing huge issues in the sense I was imagining... the highest pixel ranges from 0.98 to 1.04 depending on how the PSF is made anyway, so 5% errors are totally unavoidable.

---

• Weighting and Scaling

  The "simple" relative sensitivity calibration was causing serious problems. The assumed model for a "gain" $G$, timestream $S$, and reference timestream $R$ is: $S = G R$ Naively, one would assume that something like $G = median(S/R)$ would work. However, it doesn't. The distribution of $G$ values looks like:

---

• Sampling Causes Problems: Rd 2

  Proof that a well-sampled timestream / gaussian is recovered better than a poorly sampled one:

  

  These aren't really the most convincing, since the flux loss is only 1-8% depending on how you count. The atmosphere-included ones have larger flux loss, which is important to understand... WHY does the added noise INCREASE the flux loss? Ugh, unfortunately, there is a perfect counter-example. The first image shows a timestream in which the input map is sampled by 7.2" pixels, which is just shy of nyquist sampling the 1-sigma width of the gaussian (but is fine for the FWHM of the gaussian). The second shows 3.6" sampling of the same. No improvement. The third, very surprisingly, shows significant improvement - but this was an experiment in which the relative scales were allowed to vary.

  

  The first two each had one high-weight bolometer rejected, the last had 12 high-weight bolos rejected. So that's problably the problem....

---

• Sampling causes problems

  One of the simplest experiments that can be run is a point-source observed in point-scan mode, i.e. with smaller step-sizes. I've done this for an Airy disk with noise added in the image plan but not atmosphere (some noise is necessary to avoid bizarre artifacts with weighting when you have truly zero signal and zero noise). At a S/N of 500, the noise is pretty minimal, though. It turns out for the 'noiseless' images, the PCA cleaning is the issue... curiously, it varies significantly with iteration. Is it worth trying to debug the PCA cleaning for noiseless timestreams? I mean... they really shouldn't have any correlated information anyway. There is still an outstanding issue where one bolometer gets scaled to be higher than the rest without any apparent reason for doing so. It is also clear that we do not nyquist-sample the pixels with the timestream. However, that doesn't explain a deficiency observed in the ds1 images. Also note that the sidelobes are not picked up at this S/N, but that's not too surprising. And here is the explanation: The peak is missed by about 10% because of finite sampling? Somehow the sampled gaussian nearly uniformly underestimates the gaussian... I think this violates theory a bit.....

  

  Here's the problem shown again:

  

  This is a comparison between the input image and a ds1-sampled image with perfectly correlated atmospheric noise. So there is something in the pipeline that is preventing the peaks from achieving the necessary heights.... I wonder if resampling the deconvolved image onto a higher-resolution grid, then downsampling afterwards, would fix this? Also note that the flux loss increases from 7% to 20% from ds1 to ds5:

  

---

« Page 5 / 20 »