Debian Codegraph @

doris_core/TODO

raw
1 This file describes things that still need to be done to improve
2 the Doris software. It may not be complete, but it focuses on the
3 most important general aspects.
4 TUDelft, July 2001-2010.
5
6
7 Mahmut ideas:
8
9 1) report precise orbital model in .res file like dgm-0e4
10 Example entry in .res: Start_precise: (model name)
11
12 2) ERS put info in .res if it is nominal replica: CEOS reader modification required
13
14 remove wrong error msgs for ALOS
15
16 3) put fdc to coarse orbits results.
17
18 4) cpxfiddle: logo for [processed by Doris]
19
20 5) ellipsodial parameters revisit. i.e. WGS84, GRS80
21
22 6) Revist OVS code
23
24 7) Resampling: in .res put width and height of the rsmp slave
25
26 8) Finalize Radarsat-2 reader
27
28 9) Plot scripts to be updated for GMT and internal consistancy: plotoffsets, plotcpm etc.
29
30 10) mtiming: Get # of cpu cores + distribute nwindows
31
32 11) log getorb call and Warnings to .log file using scratchlogfile("scratchlogcallsandwarnings")
33
34
35
36
37
38 ==================================================
39
40
41 0) check accuracy of geocoding.
42 compute wavelength from radar frequency, do not use annotated wavelength.
43 compute prf again from (t_end-t_start)/#bins, do not use annotated prf.
44
45
46 1) add method for readfiles: "asar" done 200X [BK]
47 2) add method for crop: "asar" done 200X [BK]
48 3) add method for unwrap: "snaphu" done 200X [BK]
49
50
51
52 A) Phase Unwrapping module.
53 *Delft users:
54 integrate the DLR MCF algorithm with Doris. (mostly file conversions,
55 system call from within Doris, input cards, output to resultfiles.)
56 *For users outside Delft.
57 This should either be an integration with an existing software,
58 or implemented by someone. If we implement it ourselves it cannot
59 be as good as other software.
60 *Curtis Cheng's "snaphu" statistical MCF algorithm was released to the
61 public domain. Make an interface with this software.
62
63 done 200X [BK]
64
65
66
67 B) Improve external DEM subtraction module.
68 With the shuttle mission data, and general for eg. 'permanent scatterers' method.
69 Interpolation to a regular grid in WGS84 (GRS80) orbit system in a sufficient
70 spatial resolution with Matlab or GMT.
71 Coregistration of the radarcoded DEM to the complex IFG.
72
73 still requires some work
74 #%// Bert Kampes, 10-Mar-2005
75
76
77
78 C) Usage of ground control points (GCP).
79 -geolocation, bias, trend correction
80 -baseline parameter improvement (Massonet's algorithm).
81
82 still needs to be done
83 #%// Bert Kampes, 10-Mar-2005
84
85
86
87 D) Improve usersmanual.
88
89 add a FAQ with "best" answers from yahoo group.
90
91
92 E) ellipsoid card for other orbit system/extern DEM. (+transformation routines,
93 or use proj (program see d4 disk).
94
95
96
97 ---- #%// Bert Kampes, 10-Mar-2005 ---
98 --------------------------------------
99 <> GEOID subtraction of model used for SRTM-C band dem
100 <> speed improvement: fDC as constant per image;
101 implement this in member function slcimage.get_fdc(pixel)
102 if max diff in fdc < 0.5*(PRF-ABW)
103 This will speed up the resampling step probably quite a bit.
104 <> class for input handling
105 <> class for parameter pool handling (read/write, no update)
106 <> class for products: add bperp ? think about classes:
107 image::master
108 image::slave
109 product::interferogram (etc.)
110 is function of images
111 orbit::masterorbit <--- should inherit from image ? be part of image?
112 orbit::slaveorbit
113
114 then interferogram.getbperp() <--- images <--- orbits
115
116 implement smarter way of getbperp(), e.g., use a polynomial;
117
118 <> algorithm eineder for dem2phase
119
120 <> ps processing: convert DEM to master grid in meters.
121
122 <> class polynomial (2d,1d, etc.) (is in matrix class?, but why is
123 fdc_coefficients not part of it)
124
125 <> add method for interpolation of scatterered data to regular grid
126 use this to, e.g., interpolate the COMPREFDEM phase in line,pixel
127 buffers (alternatively implement quickly an linear interpolation)
128 also use this to map other DEMs to regular grid in WGS84 etc.
129 <> this can be easily implemented at line
130 "~/DEVELOP/DORIS/doris/src/referencephase.cc" [readonly] line 1115 of 1384 --80%-- col 7
131 sqrdistance: closest_one, next_closest; third_closest;
132 and to use those three (unique points) to interpolate to integer l,p;
133
134 this is a small change that really will improve results and speed,
135 since also extradense is not important than anymore. still, of course, the input
136 dem should contain all peaks of the data.
137
138 <> add UTM class, conversions to WGS84, DEM etc.
139 <> add RD class, conversions to WGS84, DEM etc.
140
141 <> check concept of memory mapping; this could save time when relatively small overlapping
142 buffers need to be loaded from disk;
143 A memory map is a virtual access (IDL associate?) Matlab (template???)
144 to a file. it does not cost memory, but you can treat a file as a single matrix.
145 Only when you access the data it is actually read from disk.
146
147 <> make a class for fasttrig (a lookup table for sin/cos of 2K length or so)
148 expected to be factor 10 faster or so cmp to std:sin()
149
150 <> BASELINE class:: add functions (3d polynomials) for
151 phi=f(lat,lon,hei)
152 line=f(lat,lon,hei)
153 pixel=f(lat,lon,hei)
154 ---> use these in comprefdem directly as alternative method.
155
156 hei=f(phi,l,p)
157 ---> use this in slant2h my_method (cmp schwaebisch)
158
159 lat=f(hei,l,p)
160 lon=f(hei,l,p)
161 ---> use these in geocode (however, geocode is already fast and now "exact")
162
163 <> use strings instead of strstr in bk_message class, to parse inputfile.
164
165 <> use class to read records in ceos format
166
167
168 cpxfiddle: option invert/negative: seems amplitude image is quite nice if negative
169 ovs: check BW is not changed.
170
171
172 matrixclass: reorganize, good testprogram;
173 oversample: avoid copying of data to oversample a real4 matrix.
174 fft: avoid changing plan if fft size changes
175
176
177 check matrix class:
178 -what is faster A[i][j] or A(i,j) or is it the same.
179 currently in matmult, etc. A(i,j) is used, so if Veclib
180 is not available, it is important. furthermore, i changed
181 it to be operator() (const uint &l, const uint &p)
182 but maybe better to inline somehow.
183 -use BLAS library for matmult, particularly for resampling this may be faster.
184
185