1 | \secA{Notes and hints on the use of \duchamp} |
---|
2 | \label{sec-notes} |
---|
3 | |
---|
4 | In using \duchamp, the user has to make a number of decisions about |
---|
5 | the way the program runs. This section is designed to give the user |
---|
6 | some idea about what to choose. |
---|
7 | |
---|
8 | The main choice is whether to alter the cube to try and enhance the |
---|
9 | detectability of objects, by either smoothing or reconstructing via |
---|
10 | the \atrous method. The main benefits of both methods are the marked |
---|
11 | reduction in the noise level, leading to regularly-shaped detections, |
---|
12 | and good reliability for faint sources. |
---|
13 | |
---|
14 | The main drawback with the \atrous method is the long execution time: |
---|
15 | to reconstruct a $170\times160\times1024$ (\hipass) cube often |
---|
16 | requires three iterations and takes about 20-25 minutes to run |
---|
17 | completely. Note that this is for the more complete three-dimensional |
---|
18 | reconstruction: using \texttt{reconDim = 1} makes the reconstruction |
---|
19 | quicker (the full program then takes less than 5 minutes), but it is |
---|
20 | still the largest part of the time. |
---|
21 | |
---|
22 | The smoothing procedure is computationally simpler, and thus quicker, |
---|
23 | than the reconstruction. The spectral Hanning method adds only a very |
---|
24 | small overhead on the execution, and the spatial Gaussian method, |
---|
25 | while taking longer, will be done (for the above example) in less than |
---|
26 | 2 minutes. Note that these times will depend on the size of the |
---|
27 | filter/kernel used: a larger filter means more calculations. |
---|
28 | |
---|
29 | The searching part of the procedure is much quicker: searching an |
---|
30 | un-reconstructed cube leads to execution times of less than a |
---|
31 | minute. Alternatively, using the ability to read in previously-saved |
---|
32 | reconstructed arrays makes running the reconstruction more than once a |
---|
33 | more feasible prospect. |
---|
34 | |
---|
35 | On the positive side, the shape of the detections in a cube that has |
---|
36 | been reconstructed or smoothed will be much more regular and smooth -- |
---|
37 | the ragged edges that objects in the raw cube possess are smoothed by |
---|
38 | the removal of most of the noise. This enables better determination of |
---|
39 | the shapes and characteristics of objects. |
---|
40 | |
---|
41 | While the time overhead is larger for the reconstruction case, it will |
---|
42 | potentially provide a better recovery of real sources than the |
---|
43 | smoothing case. This is because it probes the full range of scales |
---|
44 | present in the cube (or spectral domain), rather than the specific |
---|
45 | scale determined by the Hanning filter or Gaussian kernel used in the |
---|
46 | smoothing. |
---|
47 | |
---|
48 | When considering the reconstruction method, note that the 2D |
---|
49 | reconstruction (\texttt{reconDim = 2}) can be susceptible to edge |
---|
50 | effects. If the valid area in the cube (\ie the part that is not |
---|
51 | BLANK) has non-rectangular edges, the convolution can produce |
---|
52 | artefacts in the reconstruction that mimic the edges and can lead |
---|
53 | (depending on the selection threshold) to some spurious |
---|
54 | sources. Caution is advised with such data -- the user is advised to |
---|
55 | check carefully the reconstructed cube for the presence of such |
---|
56 | artefacts. Note, however, that the 1- and 3-dimensional |
---|
57 | reconstructions are \emph{not} susceptible in the same way, since the |
---|
58 | spectral direction does not generally exhibit these BLANK edges, and |
---|
59 | so we recommend the use of either of these. |
---|
60 | |
---|
61 | If one chooses the reconstruction method, a further decision is |
---|
62 | required on the signal-to-noise cutoff used in determining acceptable |
---|
63 | wavelet coefficients. A larger value will remove more noise from the |
---|
64 | cube, at the expense of losing fainter sources, while a smaller value |
---|
65 | will include more noise, which may produce spurious detections, but |
---|
66 | will be more sensitive to faint sources. Values of less than about |
---|
67 | $3\sigma$ tend to not reduce the noise a great deal and can lead to |
---|
68 | many spurious sources (this depends, of course on the cube itself). |
---|
69 | |
---|
70 | The smoothing options have less parameters to consider: basically just |
---|
71 | the size of the smoothing function or kernel. Spectrally smoothing |
---|
72 | with a Hanning filter of width 3 (the smallest possible) is very |
---|
73 | efficient at removing spurious one-channel objects that may result |
---|
74 | just from statistical fluctuations of the noise. One may want to use |
---|
75 | larger widths or kernels of larger size to look for features of a |
---|
76 | particular scale in the cube. |
---|
77 | |
---|
78 | When it comes to searching, the FDR method produces more reliable |
---|
79 | results than simple sigma-clipping, particularly in the absence of |
---|
80 | reconstruction. However, it does not work in exactly the way one |
---|
81 | would expect for a given value of \texttt{alpha}. For instance, |
---|
82 | setting fairly liberal values of \texttt{alpha} (say, 0.1) will often |
---|
83 | lead to a much smaller fraction of false detections (\ie much less |
---|
84 | than 10\%). This is the effect of the merging algorithms, that combine |
---|
85 | the sources after the detection stage, and reject detections not |
---|
86 | meeting the minimum pixel or channel requirements. It is thus better |
---|
87 | to aim for larger \texttt{alpha} values than those derived from a |
---|
88 | straight conversion of the desired false detection rate. |
---|
89 | |
---|
90 | If the FDR method is not used, caution is required when choosing the |
---|
91 | S/N cutoff. Typical cubes have very large numbers of pixels, so even |
---|
92 | an apparently large cutoff will still result in a not-insignificant |
---|
93 | number of detections simply due to random fluctuations of the noise |
---|
94 | background. For instance, a $4\sigma$ threshold on a cube of Gaussian |
---|
95 | noise of size $100\times100\times1024$ will result in $\sim340$ |
---|
96 | detections. This is where the minimum channel and pixel requirements |
---|
97 | are important in rejecting spurious detections. |
---|
98 | |
---|
99 | Finally, as \duchamp is still undergoing development, there are some |
---|
100 | elements that are not fully developed. In particular, it is not as |
---|
101 | clever as I would like at avoiding interference. The ability to place |
---|
102 | requirements on the minimum number of channels and pixels partially |
---|
103 | circumvents this problem, but work is being done to make \duchamp |
---|
104 | smarter at rejecting signals that are clearly (to a human eye at |
---|
105 | least) interference. See the following section for further |
---|
106 | improvements that are planned. |
---|