Single pulses

[RebeccaMcF]

I saw in the README that you recommend this version of "fake" for FRB simulation - can this script be used to generate single pulses/FRBs? And if so, how? I am familiar with the original (and MK's) "fake" but I can't see how (for this script) I should specify single pulse instead of repeating. Thank you!

[evanocathain]

Hi Rebecca,

You can use this version of fake to generate single pulses. It doesn't have a specific flag for just doing one pulse. Basically when I've just wanted one pulse (often!) I just set the period to be >= the length of the file. I could add a flag for adding just one pulse, but it would amount to that really I think. Also, I'd probably more generically like to implement a pulse amplitude distribution. What do you reckon?

The main changes wrt MK's fake were changing the pulse width etc. to be absolute rather than percentage duty cycles. But more importantly to make the S/N flag mean something else that made more sense to me.

Evan

[RebeccaMcF]

Hi Evan,

Thanks for your reply. Now that you've said it, it seems so obvious!

A few comments:

1. I notice that the pulse is placed around phase = 0.5 (dedispersed) so I guess: period = file length will give you a centred pulse, and period > file length will start placing the pulse into the second half of your file. For some applications (ie when building a training set), it could be important to make sure this offset varies!

2. By pulse amplitude distribution, do you mean the functionality to produce, say Gaussians, instead of a top hat? If so, I think that would be really useful as sharp edges provide false hope for detection algorithms!

3. I can see why absolute pulse width is particularly important for single pulse!

4. I notice that in both scripts, the SNR is scaled by a factor of 1/sqrt(nchans), however, you are also reducing the SNR by a factor of (sqrt) the ratio of smeared_width to original_width (line 230). Is this what you mean?

So the SNR given at the input of your script will represent the peak SNR of the coherently dedispersed (or original undispersed) signal, while in MK's version (which only scales for the number of channels) the SNR given at the input will represent the peak SNR of the signal integrated over all channels but still reduced by dispersive smearing within each channel. Does that sound right?

Thanks again!

Rebecca

[evanocathain]

Hi Rebecca,

1. Yes, the 0.5 phase is hard-coded. This has not been of any consequence for the things I've used this for, but I agree that making a flag to make it random could be useful for sure. And it's easily done.

2. I just meant the distribution of the area under the box-cars to be honest. I also, separately think a Gaussian (and other shapes) are a good idea. If you search for a Gaussian using a boxcar you get ~0.868 of the S/N recovered. I have been meaning to do this to 'see' that effect. But, as an aside, it looks like that is seen in Molonglo data and injections"
   https://ui.adsabs.harvard.edu/abs/2019arXiv190707948K/abstract
   https://ui.adsabs.harvard.edu/abs/2019MNRAS.488.2989F/abstract

3. Yes, indeed.

4. The 1/sqrt(nchans) is just because you write out the value per channel but the input is an overall S/N. And yes I smear the signal preserving the area under the curve. That is the same 'specific fluence' (often just called fluence or even 'energy'). The point there is that 2 identical pulses (say 2 FRBs from same z, or even same DM as scatter smearing would be qualitatively the same) will have different smearing and so different S/N for the same fluence. I have been known to rant about that and 'completeness'. You see everything above some S/N limit but fluence is a more useful physical unit and you need to understand things in terms of it to (say) scale to other bands or telescopes.

Anyways, I think the upshot of this are 3 feature requests for fake.c:

1. random phase offset
2. pulse amplitude distributions
3. different pulse shapes

Do you agree? If so I might close this issue, leave this fake.c intact as is, for posterity, and raise this in my sigproc branch here: https://github.com/evanocathain/sigproc/blob/master/src/fake.c and just do the changes to that fake.c.

What do you think?

Evan

[RebeccaMcF]

Hi Evan,

Happy to move across to your sigproc branch. You have summarised well - I had no expectation that you would implement all of these changes, but it would be AMAZING if you can. Thank you.

Happy to have a good rant about SNR sometime! I have come across a few different definitions, which I think is important to understand when making comparisons.

Thanks for the discussion,

Rebecca