Laman

Remove Your Ain Ligo Noise

I lead hold live on played amongst the raw information from the outset LIGO gravitational wave, GW150914. Everyone who tin code is invited to practice the same. It's a lot of fun.

Although I had to larn some Python for the Higgs Kaggle contest, my preferred tool is Wolfram Mathematica – whose full online version was only completed a calendar week ago.




What should yous practice to evaluate the LIGO raw information yourself? Download all the datasets as well as Python code from the official LIGO tutorial page. Or follow full online version was only completed a calendar week ago.




What should yous practice to evaluate the LIGO raw information yourself? Download all the datasets as well as Python code from the Nicholas Lincoln's blog – he is a information analyst who has no official physics background, equally far equally I understand.




Or yous tin maintain amongst me as well as I volition endeavor to brand the things genuinely concise.



The audio of the merger I tin take heed sounds almost only similar the heartbeat around 0:01 of this Lenka's song. BTW Lenka would celebrate nurture twenty-four hours today if she were all the same Czech. I am deeply aware of this fact because I had met a pretty canonical holder of that nurture on her nurture day, Feb 21st, but inwards 1995. She is German linguistic communication these days (much similar the vocalizer is Australian) but amongst the nurture similar Lenka, it's difficult to claim that yous are not Czech.

Spoiler: I got a beautiful "heartbeat" audio from both detectors, pretty much shifted past times 0.0073 seconds (30 sampling points, equally I volition explain), as well as amongst the nearly contrary phase. My approximate of the actual confidence degree is manner higher than v sigma. It would live interesting if yous tried to give me your ain persuasion nigh the significance of the signal.

For your convenience, I created simplified copies of the raw information from both detectors:
b-Louisiana-32sec-at-4096hz-in-1e-24.txt
b-Washington-32sec-at-4096hz-in-1e-24.txt
The files are posted at Dropbox. I recommend yous to create your ain 2 GB Dropbox cloud if yous don't lead hold one.

Each one-megabyte text file contains \(32\times 4,096=131,072\) lines amongst uncomplicated integers comparable to 1 i chiliad one thousand (with either sign). They are the "strain" information describing the "length difference" of the 2 LIGO arms during a 32-second window, sampled 4,096 times a minute (the sampling frequency is 4,096 Hz, nosotros say), as well as it's inwards the units of \(10^{-24}\) meters. (So to practice the files, I divided the master copy numbers past times \(10^{-24}\) as well as rounded them.)

So far, it looks simple, right? Among the 131,072 entries inwards each file, the dark hole merger is unopen to the middle, to a greater extent than exactly around 67,250. Let me tell yous inwards advance that the phases are to a greater extent than or less contrary to each other, the Washington signal appears nigh 0.0073 seconds (about thirty sampling points) after the Louisiana signal, as well as the Washington signal is a fleck stronger.

Now, if yous play the raw "sound" directly, it volition comprise likewise much noise. You won't take heed anything. You volition Fourier transform it as well as apply a filter – to alternative frequencies betwixt 50 Hz as well as 250 Hz (but the filter should live sort of shine equally a component of the frequencies). You endeavor to play it as well as it's all the same racket – well, yous volition take heed some resonant whistling at preferred frequencies.

The filter wasn't enough! You ask to take away this "predictable" whistling – which represents the component of the signal amongst a high total of autocorrelation. How practice yous practice that? Compute the Fourier transform of the signal \(|\hat S(\omega)|^2\), average this component of the frequency over nearby frequencies, accept the foursquare root again, split the master copy \(\hat S(\omega)\) past times this foursquare root, as well as multiply it past times \(1/\omega\) to acquire a realistic conduct at depression plenty \(\omega\) again.

This removal of the highly autocorrelated component of the information is known equally whitening.

Or practice whatever your mutual feel tells yous to do! At the end, yous should acquire rid of the racket – likewise high as well as likewise depression frequencies; the highly autocorrelated whistling etc. And when yous play this "clean" version of the strain data, yous should clearly take heed a heartbeat at the house that I indicated.

When yous lead hold it, yous should endeavor to combine the Louisiana as well as Washington information to maximize the describe of the signal as well as compare its intensity to the noise. Estimate the probability that such potent as well as highly similar heartbeats appear past times chance. I haven't quite estimated my ain value of the confidence degree but I am confident it volition live far higher than v sigma, fifty-fifty after the look-elsewhere correction is applied.

You know, the possible delay is betwixt minus 10 milliseconds as well as summation 10 milliseconds – betwixt –40 as well as +40 sampling points. Add a element of 2 for the possible flip of the relative phase. The look-elsewhere effect is increasing the tiny \(p\) from the \(p\)-level at most past times 160 which is negligible. But when I combined the 2 signals (with the shift as well as the inversion), as well as normalized it as well as then that this component has the commons criterion departure of one, I got a signal that exceeded six sigma at to the lowest degree at 3 independent plenty places. So I intend that the local significance degree is unopen to six times the foursquare root of 3 which is over 10 sigma. The probability is some \(10^{-24}\) or whatever it is, or \(10^{-22}\) when yous cut it past times the 2 orders of magnitude. So a signal of this magnitude appears i time inwards \(10^{-22}\) cycles or \(10^{-17}\) seconds or 10 billion years or so.

Maybe my calculation is naive but it qualitatively reveals my certainty that this can't live a statistical racket – such "bad luck" has occurred i time since the Big Bang. It's much to a greater extent than probable that it's a number of LIGO's hacked computers. Computers are hacked much to a greater extent than often than i time per 10 billion years.

I haven't created a Mathematica notebook that would live illustration but yous may hold back at i linked to inwards this sentence. Fourier[] as well as ListPlay[] are the most nontrivial Mathematica functions for this sort of game. And I am sure enough I could acquire a much cleaner signal producing a higher significance level, too.

My notebook volition exclusively live useful to an intelligent user. Influenza A virus subtype H5N1 component of it has to live run twice, i has to shout out back the Louisiana as well as Washington results inwards 2 variables, as well as they're used later. I guess that Mathematica volition live useless for a user who can't practice these things herself, anyway. (Update sun night: I made the notebook runnable without whatever manual adjustments.) Maybe such users may prefer a PDF printout of the notebook, anyway. Funny, yous ask to larn Farsi or Standard Arabic to read it. Is Stephen Wolfram promoting multiculturalism? ;-) I honestly did aught else than "save equally PDF". (Update: It's genuinely a glitch inwards the Dropbox PDF viewer, the PDF looks fine e.g. inwards Chrome when downloaded.)



After diverse adjustments of the filters as well as whitening parameters as well as combinations of the 2 detectors, this is in all probability the best reconstructed signal I could get. It's included inwards the updated PDF file linked to above. The strain signal is normalized as well as then that the criterion departure is "1" on the \(y\)-axis. Nevertheless, the signal locally reaches 8-10 sigma at 3 places! Also, yous may meet that the signal genuinely looks to a greater extent than (quasi-)periodic earlier the merger than after it. It genuinely looks similar I tin almost clearly meet a whopping number of 7 orbital periods earlier the peak of the merger! ;-)

No comments:

Post a Comment