Menu
▾
▴
#123 Period analysis
This tracker is a placeholder for (post phase 1) period analysis work.
It is also a place to accumulate notes and links.
Matthew Templeton has mentioned a C Analysis of Variance implementation at:
http://users.camk.edu.pl/alex/#software
The readme file has a copyright notice, so it may not be possible to use this in VStar given its GPL license.
Since the initial comment above, I have had further communications with Matthew Templeton and obtained Date Compensated Discrete Fourier Transform code written in Fortran by he and Grant Foster several years ago. I am in the process of translating this to Java and comparing its behaviour against the Fortran code (compiled with GNU Fortran).
This work was originally going to be done post phase 1, but it is important to at least have prototyped this before NACAA (http://www.nacaa.org.au/2010/workshops?page=1) so this is currently my highest priority.
See also https://sourceforge.net/apps/mediawiki/vstar/index.php?title=Fourier_Analysis and https://sourceforge.net/apps/mediawiki/vstar/index.php?title=Phase_III_VStar_Requirements
In r441 (http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=441): first semi-working version of date compensated discrete Fourier transform code (translated from Fortran). Unit test shows that same frequency and period values are yielded from Fortran and Java versions. The work for this is covered under tracker 2942299. See the header comments for more background information and references.
Much code clean-up required. But first, the power and amplitude values are not being computed correctly wrt the Fortran code.
We now have an implementation of date compensated DFT translated from Matthew Templeton's and Grant Foster's ts1201.f Fortran code that is validated against the delcep.vis input and delcep.ts output example supplied with the latter! This is captured by the DateCompensatedDiscreteFourierTransformTest.java.
This is in revision r442; http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=442
Next steps:
o Clean up the code.
o Plot the data; freq vs period/power/amplitude?
Code clean-up in r443: http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=443
More period analysis work; all charts in place and sending/receiving selection notifications in http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=448
Changed period scale to be logarithmic else most of the freq vs period curve is flat and hard to select via cross hairs. http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=450
Changed ordering of period analysis coord types to yield a different ordering in tabs. http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=451
Added period analysis table in http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=453
Data table cell format changes and refactoring in http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=456 and http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=457
Now only applying period analysis to series which is source of mean series. Added subtitle to show this series.
In http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=461
Changing back to Phase 2 since more work will be done in that phase on period analysis.
I think Period Search should be pulled out of the Analysis menu into a more general Search menu (or similar).
Further updates for DC DFT period analysis visualisation including a focus on amplitude and top hits table.
In http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=465
Period analysis for a star is now only recomputed if the dataset's raw data "Means Source" series is changed via the Change Series button in the Plot Observations and Means view.
In http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=466
This completes what I plan to do for period analysis before NACAA and the version 1.0 Beta release later today
The top hits dialog should update when the other views are selected, as does the all-data table. Having said this, updates across views depends upon whether a datapoint exists at the point on the line that has been selected. This can probably be achieved by moving from a line plot to a scatter plot, but that won't be as "pretty". Perhaps there is some kind of snap-to setting for line plots.
Added more info re: period analysis to help page in http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=467
That's it.
At the NACAA 2010 workshop, Tom Richards mentioned other aspects of period analysis, such as:
o searching for overtones
o light-time effects of a 3rd star
o pre-whitening
He also thought that the existing roadmap (requirements etc) CLEANEST, Slick were worthwhile.
It was suggested at the NACAA 2010 workshop that the ability to sort on any data pane column in period analysis would obviate the need for the top-hits pane.
Excerpt from an email from Sara to Aaron, Arne, Matt, myself: "...make it easier for the user to change some of the parameters on the DCDFT plot (or make the procedure clearer). For example, this may be accomplished simply by giving clearer instructions on how to zoom in on the Amplitude vs Frequency plot so one can see the spike near zero better. Also, it might be nice to allow the user to adjust the frequency spacing on the FT by a factor of 2-4 above the Nyquist rate." Sara suggested talking with Matt further about this.
New phase plot can now be initiated from period analysis window. In http://vstar.svn.sourceforge.net/viewvc/vstar?view=rev&revision=488
Further to email excerpt below re: zoom, it would be worth adding zoom in/out buttons to period analysis plot panes, as we've done for light curve and phase plot panes.