Data analysis is a big topic! There are a lot of resources on the web, but it might be tough to know where to start. Here are a few suggestions of our favorite entry points:
The GWOSC Learning Paths walk gently from intro videos to more advanced tutorials. This is great for high school or undergraduate students.
The GW introductory tutorials are a good place to start if you want some code, but are new to python and/or signal processing
Iām a GravitySpy glitch classifier that would like to make some Q-scans of glitches. I believe Omicron does this 24/7 but I would just like to find out how itās done.
I did the GWOSC #5 course (up to q-scans) but it doesnt go into the algorithm. The python code tells us to specify a āQ-rangeā in the Q-scan, 4 - 12 works wonders , but it was confusing that āconstant Qā has a range.
Various sources say Q is frequency divided by its bandwidth. i.e a sampling rate/ window size.
Following your suggestion to someone else on this forum I found the formula for Q in prof Keith Chatterjiās PhD paper.
There are two parameters ( on page 82 equation 3.25 Q(min) and N(Q) equation 3.24c )
Are these the two parameters that go in Q-range in the Q-transform algorithm ?
Hi @David . Nice to meet you! Thank you for your work with GravitySpy, and for your question.
Iāll give you some resources, and hope that some of these help. Please let me know if something more would help.
The equations you point to in Chatterjiās thesis are one method for selecting the lower and upper values of the Q-range. In the gwpy method, instead, the user selects the lower and upper range of Q-values.
Typically, we would select a range of Q-values that captures as much GW signal power as possible in a small number of pixels.
For a nice picture of what Q means, see Figure 3.1 in the same thesis. The figure shows a sine-Gaussian with a Q of 10, which corresponds roughly to the number of cycles you can see (I count about 7). If we doubled the Q to 20, weād see around 14 cycles instead.
I put some more notes about how to select a good Q-range in this introductory tutorial. Additional notes can be found by clicking āSee Notesā under the Q-scan plot in the Quickview Web App.
In fig 3.1 the central frequency is 1 Hz. So tuning Q from 10 to 20 is basically asking the sineGaussian to, do whatever it does, to resolve frequencies down to 1/20 Hz i.e 2 times better?
And also the waveform fits about 2 times more cycles in its packet.
While still constrained to produce a tiling with poorer time resolution.
Anyway I appreciate you replying and I will look at that tutorial.
regards
David
Hi @David Yes - thatās right. Adjusting Q is a trade-off between better resolution in time and better resolution in frequency.
Typically, high mass signals (around 50 solar masses) are in the LIGO band for a short time, so high time resolution (low Q) is important. On the other hand, lower mass signals (say, a binary neutron star merger) are in the LIGO band for longer, so using a higher Q can be a better choice.
