Sgp4 validation

[thkruz]

This fixes the accuracy issues raised in #107 and a second issue with satrec initialization.

• WGS constants have been changed from WGS-84 to WGS-72
• Missing parenthesis added to dspace.js
• a, alta, and altp moved from io.js to sgp4init.js to match C# version
• 1st and 2nd derivatives of mean motion no longer calculated
• Position/Velocity validated to 1mm accuracy for time of 0, 360, 720, 1080, and 1440 minutes for full catalog
  • Results of sgp4prop will not be uploaded because of ITAR banner (and it is 4GB+)
  • Snapshots of calculations saved to ensure future updates don't invalidate propagation
• Test creator scripts included
• Code Coverage currently 98.93%

Screenshot showing the sgp4prop tests running when the snapshots were made.

[ezze]

@thkruz What a huge work you did, mate! Awesome!!!

Could you please describe a little bit in details how you generated snapshot data for tests. I mean: where did you take TLE inputs and calculation outputs corresponding to them? I see it's based on original C++ library but I feel that we must provide more explainations for next generations so they could recreate these data if they need. :)

[thkruz]

@ezze something like this?

How I Tested Against Sgp4Prop

1. I pulled the latest catalog from https://keeptrack.space (out of convenience to me) but you could pull it from space-track.org or celestrak
2. I downloaded the sgp4prop library from space-track.org and then used the included python wrapper (more on how below)
3. I split the TLEs into 500 satellite chunks (1000 lines) and placed those chunks in the python wrappers input folder as TLEX.inp
4. I ran each input file through the python wrapper to generate a standard sgp4prop output for 0-1440 minutes
5. I parsed/converted those ouput files into json
6. I created 52 test files that read in the corresponding json files (line 5) and check if satellite.js gets the same answer at 0, 360, 720, 1080, and 1440 minutes to within +/-1mm

The probability of me parsing the sgp4prop output incorrectly and then calculating the same wrong value with satellite.js to 5 decimal places seemed low enough to ignore that possibility.

I could have tested more data points for each satellite but getting jest kept trying to do too much at once and ran out of memory. There is probably a way to get it to do the tests in order and garbage collect as needed but I couldn't figure it out. 6 data points felt sufficient.

I chose the python wrapper so that I didn't need to compile anything. I would recommend the same unless you have an environment already setup for the other languages included in sgp4prop.

How I Made the Snapshots

Now that I knew satellite,js got the right answer at those 6 times I created 52 more tests (test maker included in the PR) that tests if satellite.js still gets the same snapshot as it did the last time it ran. On the initial run Jest will generate a snapshot to compare against in the future - this is a built in feature - that initial snapshot is what I uploaded.

So by propagating the same TLE to the same 6 points in time that passed my sgp4prop test I know that my snapshots are accurate without uploading the actual values sgp4prop got. If the math gets altered, the snapshots will no longer match and the test will fail.

[ezze]

@thkruz Thanks, that's what I was asking for!

Let's try to optimize these 52 files with tests by "merging" them into a single file, and then, I guess, we are good to go. Probably, it would be better to include it in a major release.

@TSKelso @davidcalhoun @tikhonovits @DangoDev Do you also want to review it?

[thkruz]

@ezze I had split them on purpose so that Jest would run them in parallel, but will defer to your expertise on the best way to organize stuff. (Hard to believe it was only a few years ago that you informed me that I should get npm and catch up to the rest of the world).

Let me know if you want to do the merging or if you want me to. I will be away all of September, so I might be a bit slow.

[davidcalhoun]

Good stuff! I agree that this is a significant change that warrants a major release!

I didn't realize Jest needed tests in separate files for parallelization but I confirmed it, makes sense to me! Many thanks for documenting the testing development.

[ezze]

@thkruz Ok, if you want to split them for the sake of parallel run then at least you can create a single utility function including the repeating code. And each test will look something like:

sgp4-catalog-<n>.test.js

import { sgp4test } from './sgp4test';
sgp4test(<n>);

where <n> is the number of your test file.

Moreover, these little test files could be generated easily by some Node script so you don't have to write them by yourself.

[ezze]

@thkruz Let's look when you optimize test scripts, and then I can merge it. If you can't due to lack of time, let me know. I will do it by myself then in upcoming days. All the most difficult stuff is already done by you, mate, so all the rest is super easy. ;)

[thkruz]

Consolidated all tests into one test file and all snapshots into one file. Was actually comically easy to do.

Proof that the commit currently matches the official propagator when the snapshots were generated:

[vinifraga]

Incredible work @thkruz and @ezze! Thank you for the amazing work.

I'm using this branch on my project, any news when this will be merged?

[ezze]

Fixed in 5.0.0.

@vinifraga Let me know whether it works for you.