Author Topic: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread  (Read 947465 times)

Offline moralec

Hi guys,
Just a note to let you know that with the help of Princess, Arnezami and wronkiew, I have started to work on a new "progress report". The idea is to come up with a long write up describing all the different aspects of the video recovery effort (MMB sequences, bitfilps, clock recovery, etc). I'm also planning to include a section for video interpretation.  I'm thinking this could eventually be used as a base for a NSF article. 

The latest version is the wiki (http://spacexlanding.wikispaces.com/Latest+Progress+Report). Any comments or additions are more than appreciated, please feel free to add new paragraphs or to improve what is already there.



« Last Edit: 06/15/2014 05:53 PM by moralec »

Offline Jester

  • Administrator
  • Senior Member
  • *****
  • Posts: 6991
  • SpaceShip Earth
  • Liked: 3252
  • Likes Given: 114
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1601 on: 06/15/2014 06:27 PM »
Hi guys,
Just a note to let you know that with the help of Princess, Arnezami and wronkiew, I have started to work on a new "progress report". The idea is to come up with a long write up describing all the different aspects of the video recovery effort (MMB sequences, bitfilps, clock recovery, etc). I'm also planning to include a section for video interpretation.  I'm thinking this could eventually be used as a base for a NSF article. 

The latest version is the wiki (http://spacexlanding.wikispaces.com/Latest+Progress+Report). Any comments or additions are more than appreciated, please feel free to add new paragraphs or to improve what is already there.


I'm cleaning it up a bit, did we ever get a picture of the famous pizza-dish antenna ? it would be nice to add it .

Offline Quialiss

  • Member
  • Posts: 75
  • Liked: 82
  • Likes Given: 29
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1602 on: 06/15/2014 07:33 PM »
<snip>
Generated mmb files:
http://adama.nocdirect.com/~wronkiew/spx_crs3/mmbs_alt/

There are a few minor differences in the generated mmbs between the wiki and the spreadsheets. Please take a look through the files and let me know if you see any problems. Otherwise it will go live tomorrow night.

I ran a diff on those and and the ones generated from the wiki.. someone may have gotten to them before me as I couldn't see any differences anymore.  So all good to go :) 

Offline Lourens

  • Full Member
  • *
  • Posts: 156
  • The Netherlands
  • Liked: 206
  • Likes Given: 304
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1603 on: 06/15/2014 08:01 PM »
Well, I'm still working on automated bit flipping, and a frame that no human could or would do sounds like just what we would like to have an automatic system for. I think I'm getting a little bit closer to something workable, although I doubt it will ever be fully automated. But that's okay. And even if we never get I-frame 1 back, it may still be useful for filling in some of the gaps in the more interesting I-frames.

<snip>

The code is an ugly hack in Bash, but if anyone wants it, just let me know. Meanwhile I'll keep fiddling with it :).

I didn't realize you were doing them automatically!  I wouldn't mind seeing your script. 

Have you tested it on any of the more intact frames?  That should help figuring out the 'correctness' thresholds, giving you a reference frame to look at to find any places where your script is making incorrect flips.

The flips I posted before were done by hand, and I'm now trying to automate that process. I agree that it would be good to try it on one of the frames that we have some ground truth for. I'll try that next time I have time to work on this, hopefully somewhere in the coming week.

I've attached the script, the original extension is .sh, but that doesn't fly with the forum software, which is a good thing. Like I said it's a bit of a mess, and it's getting a bit too complex for bash really, but my Python is rusty, C++ (my usual weapon of choice) even less suitable, and Octave isn't really it either, so bash it is.

Hi guys,
Just a note to let you know that with the help of Princess, Arnezami and wronkiew, I have started to work on a new "progress report". The idea is to come up with a long write up describing all the different aspects of the video recovery effort (MMB sequences, bitfilps, clock recovery, etc). I'm also planning to include a section for video interpretation.  I'm thinking this could eventually be used as a base for a NSF article.

Great! I've been working on an article with Chris. It's in the early draft stages, and we think we can make something good out of it. I don't have time right now, but I'll have a look next time I work on the article.

Offline Quialiss

  • Member
  • Posts: 75
  • Liked: 82
  • Likes Given: 29
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1604 on: 06/16/2014 06:21 AM »
I got around to calculating those statistics on triple flip patterns.  Without further ado, the top 15.  The top 2 were already known, but the top 3-5 may be useful to know, beyond that the statistical significance fades drastically.

10   100011000000001100101
10   10111000000000111001
10   11111000000000100001
11   10011000000000110101
13   10001000000000110011
13   1101000000000010111
14   10000001000000110000011
14   100001000000001100011
18   1011000000000011101
20   1111000000000010001
25   1001000000000011011
54   101000000000001111
87   111000000000001001
342 11000000000000101
754 1000000000000011

With reasonable completeness, searching for 10 flips within 10 bits of the start of the first flip, and 4 flips within 16 bits.  So the fact that the most common flips are within 4 bits is not a sampling bias, and may be helpful with the automated bit flip searches, reducing the space to search. 

The other way to go about calculating stats on relative occurrences of the flips would be to take a portion of the padding and attempt to cancel out all the differences between princess's two files with triple flips.  That assumes that all the flips are triple flips, if that's true it would give a more complete picture of the distribution of the flips in the heavily corrupted areas, as simple matching can't find a constantly overlapping sequence of flips.

Offline wronkiew

  • Full Member
  • *
  • Posts: 186
  • 34.502327, -116.971697
  • Liked: 105
  • Likes Given: 121
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1605 on: 06/16/2014 07:09 AM »
<snip>
Generated mmb files:
http://adama.nocdirect.com/~wronkiew/spx_crs3/mmbs_alt/

There are a few minor differences in the generated mmbs between the wiki and the spreadsheets. Please take a look through the files and let me know if you see any problems. Otherwise it will go live tomorrow night.

I ran a diff on those and and the ones generated from the wiki.. someone may have gotten to them before me as I couldn't see any differences anymore.  So all good to go :) 

Thanks for checking it over. I have enabled this by default. The newest version on YouTube was generated from the spreadsheets.

Offline mhenderson

  • Member
  • Posts: 69
  • USA
  • Liked: 101
  • Likes Given: 18
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1606 on: 06/16/2014 01:06 PM »
Is an evolution / linear programming approach to bitflipping workable?

Some optimization algorithms work by exploring a gradient to find the way down a sloping tableau to a minimum.  Example: Find the deepest part of an ocean by systematically dropping 1000 large lead spheres and determine where they settle. (Assume they roll downhill). From the best candidates, repeat the process in that neighborhood using smaller spheres. This works as long as the deepest part of the ocean isn't a well drilled in a flat plateau. Flat areas would be eliminated early in the search.

In our case, we are searching for good MBs. Do our spheres roll downhill?  i.e. Are improvements incremental? We have some evidence (triple flips), bands of bad MBs, etc. that these errors came in bunches.

We know that "good" MBs score well in certain criteria such a "MB has no quant error", "iFrame MB has a size that is comparable to other iFrame MBs in the same position", and "dependent neighbor MBs to the right and below do not have errors." You could even add "this bitflip is part of a triple flip that further improves the MB Score".

1) Develop a MB scoring metric: Start with good MBs, add random bitflips. If errors accumulate to continuously decrease the "MB score", then that is a good metric that might be applied to bad MBs.

2) Use the metric on bad MBs ... add random bitflips, keep the ones that improve the "MB score". Lather, rinse, repeat. 

Online cscott

  • Senior Member
  • *****
  • Posts: 2896
  • Liked: 2016
  • Likes Given: 664
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1607 on: 06/16/2014 03:09 PM »
1) Develop a MB scoring metric: Start with good MBs, add random bitflips. If errors accumulate to continuously decrease the "MB score", then that is a good metric that might be applied to bad MBs.

I think the best MB scoring metric would be "closeness to hand-tuned MB".  We've already done a lot of work on manual MB fixup, but probably missed some bit flips in LSBs which are invisible (but accumulate).   A triple flip search could use the work we've already done by hand to guide an automated search for triple flip patterns, potentially fixing up some LSBs and finding valid MB corrections where we are currently using alignment or DC level overrides of various kinds.

That is, use the PNG output for am i frame as the target, and see if we can improve on what we have.  iframe improvements should lead to p-frame improvements, etc.  Hopefully we can use the humans to bootstrap the automated search, by giving it a solid idea of what it should be looking for.

The results are in.

In a slight change of plan I want to get information from everyone who received at least 1 vote, this way if SpaceX are feeling extra generous then they have everything they need ;)

So in no particular order (actually alphabetically) here are the chosen 15:

arnezami
Lourens
mhenderson
michaelni
Mlindner
moralec
mvpel
Princess
Quialiss
Saliva_Sweet
Shanuson
SwissCheese
theshadow27
Untribium
wronkiew

And here are the other 15 who I still want to get addresses for :)

Adaptation
ajmartin
dgdpg
dorkmo
Exclavion
gospacex
grythumn
jakusb 
Jared
John_L
lgjy98d
morningdew76
pshrpd
seanpg71
Turix

I need the following information from people:

Name:
Gender: (for clothing purposes I'm not a stalker)
T shirt size: (S/M/L/XL/XXL)
Address: (this needs to be somewhere that a fairly large package can be taken and signed for e.g. a work address, don't forget to include the country ;))

I'm going to start chasing people pretty soon :) However I won't chase forever, if I haven't got the information by the end of the week then you will be removed from the list :o (Hopefully it doesn't come to that though)

Thank you to everyone that voted, and in particular to those who graciously removed themselves from the list.

Hi guys,
Just a note to let you know that with the help of Princess, Arnezami and wronkiew, I have started to work on a new "progress report". The idea is to come up with a long write up describing all the different aspects of the video recovery effort (MMB sequences, bitfilps, clock recovery, etc). I'm also planning to include a section for video interpretation.  I'm thinking this could eventually be used as a base for a NSF article. 

The latest version is the wiki (http://spacexlanding.wikispaces.com/Latest+Progress+Report). Any comments or additions are more than appreciated, please feel free to add new paragraphs or to improve what is already there.


I'm cleaning it up a bit, did we ever get a picture of the famous pizza-dish antenna ? it would be nice to add it .

I also corrected some things (spelling/grammar mainly) and left a couple of comments.

Offline catdlr

  • Member
  • Senior Member
  • *****
  • Posts: 5304
  • Viewed launches since the Redstones
  • Marina del Rey, California, USA
  • Liked: 1960
  • Likes Given: 1317
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1610 on: 06/16/2014 07:10 PM »
Perhaps a few seconds of credits could be added to the end of the video so that everyone involved has their member name (or real name) noted?
« Last Edit: 06/16/2014 11:11 PM by catdlr »
Tony De La Rosa

Offline moralec

Great! I've been working on an article with Chris. It's in the early draft stages, and we think we can make something good out of it. I don't have time right now, but I'll have a look next time I work on the article.

Awesome Lourens! Im happy to integrate both if that works for you.

Offline saliva_sweet

  • Full Member
  • ****
  • Posts: 552
  • Liked: 423
  • Likes Given: 1328
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1612 on: 06/16/2014 11:51 PM »
I'm able to convert I frames to P frames now, this enables using the -3 mmb command in iframes. Nobody is using -1 or -2 mbs in p frames because -3 is just much better for the video. I think there are iframes (particularly towards the end of the video) that may greatly benefit. However, mmb conversion isn't working yet and manual conversion is not an option at this point. So results look pretty crappy atm. I'm attaching the leg deploy sequence where iframe 81 is converted to pframe. dcs get thrown, all looks ugly, but I think there's potential.

Offline Quialiss

  • Member
  • Posts: 75
  • Liked: 82
  • Likes Given: 29
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1613 on: 06/17/2014 12:16 AM »
I'm able to convert I frames to P frames now, this enables using the -3 mmb command in iframes. Nobody is using -1 or -2 mbs in p frames because -3 is just much better for the video. I think there are iframes (particularly towards the end of the video) that may greatly benefit. However, mmb conversion isn't working yet and manual conversion is not an option at this point. So results look pretty crappy atm. I'm attaching the leg deploy sequence where iframe 81 is converted to pframe. dcs get thrown, all looks ugly, but I think there's potential.

It looks like the issues with the i frame converted to p frame are caused where we're using -1 or -2 with DC adjustments to make the next block look better?  That shouldn't be too much of a headache to work around, and you're right, it could seriously improve the look of parts 12-15, with their large gaps.  Nicely done.   :)

Also, I do sometimes use -1 in p frames.  Not often, but it has it's uses on frames where it's better to have the DC data on the missing mbs rather than leaving them blank. 

Offline Quialiss

  • Member
  • Posts: 75
  • Liked: 82
  • Likes Given: 29
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1614 on: 06/17/2014 02:35 AM »
Progress report!

Current
 


10 days ago   



I'm reaching the end of what I can do with MMB work.  Parts 3 and 4 are the only ones left that I know need working on, SwissCheese's alignments, and then my work on the DC values.  More eyes on the frames to spot things that have been missed are always appreciated!
Spreadsheet Review
Download all frames (I think that's the right one.  wronkiew?)

Promising avenues that I know of are saliva_sweet's work with converting i frames to p frames to effectively fill gaps in the i frames, Lourens's work on automatically finding bitflips in i frames to recover more data, and wronkiew's work restoring the clock.  Is there anything ongoing I've missed? 
« Last Edit: 06/17/2014 03:33 AM by Quialiss »

Offline wronkiew

  • Full Member
  • *
  • Posts: 186
  • 34.502327, -116.971697
  • Liked: 105
  • Likes Given: 121
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1615 on: 06/17/2014 04:58 AM »
Download all frames (I think that's the right one.  wronkiew?)

Yes, that is the hourly image build.

Offline mhenderson

  • Member
  • Posts: 69
  • USA
  • Liked: 101
  • Likes Given: 18
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1616 on: 06/17/2014 08:22 AM »
Side by side comparison in youtubedoubler of the ten day progress videos posted by @Quialiss. The newer video is on the right.

http://youtubedoubler.com/cID9
« Last Edit: 06/17/2014 08:49 AM by mhenderson »

Offline mhenderson

  • Member
  • Posts: 69
  • USA
  • Liked: 101
  • Likes Given: 18
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1617 on: 06/17/2014 08:38 AM »
Side by side comparison of current video (right) to original raw (left).

http://youtubedoubler.com/cIDx

The videos repeat but have different lengths. Click the Double Up button or reload the page to re-start both.

Offline zodiacchris

  • Full Member
  • *
  • Posts: 144
  • Port Macquarie, Australia
  • Liked: 165
  • Likes Given: 302
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1618 on: 06/17/2014 09:51 AM »
Guys,
What you have achieved leaves me speechless, brilliant work!! I would never have thought this quality would be possible, well done!  :D

Offline Paul Adams

  • Veteran
  • Full Member
  • ****
  • Posts: 494
  • United Kingdom and USA
  • Liked: 36
  • Likes Given: 24
Re: SpaceX Falcon 9 v1.1 CRS-3 Splashdown Video Repair Task Thread
« Reply #1619 on: 06/17/2014 10:27 AM »
Superb work everyone, really well done.

I would have thought that deserves a SpaceX t-shirt for the main contributors!

Paul
It's all in the data.

Tags: