[Inconsistent Results With CS16 File Input] - Smaller Input Size Generates Larger Number Of AIS Messages.

[ekarseras]

This issue regards v0.62 and the test conditions are the following:

1. Use raw file input (cs16) of IQ data sampled at 1.535 MSPS.
2. The first file has 400000 IQ samples stored (test_ch0.cs16).
3. A second file is generated by removing the first 60000 IQ samples from the first file (test_ch0_chopped.cs16).
4. When AIS-catcher is ran on the first file (model -m 2)

./AIS-catcher -m 2 -v -o 2 -s 1536000 -r -ga FILE test_ch0.cs16 FORMAT CS16

this will generate 5 messages.

5. When AIS-catcher is ran on the second file (model -m 2)

./AIS-catcher -m 2 -v -o 2 -s 1536000 -r -ga FILE test_ch0_chopped.cs16 FORMAT CS16

this will generate 7 messages.

6. This is contradictory because the second file should contain less messages than the first one, being its truncated version.
7. When the same test is repeated with other models, -m 1 and -m 0; the results are consistent, i.e., a smaller file will produce less messages.

The test files are attached.

test_ch0.tar.gz

[jvde-github]

Will have a look, but note that decoding can be a statistical game. For very weak messages, there is only a certain chance you will find them. And if you cut off part of the file, the decoder will be in a slightly different state and can then pick up a message or not.

For that reason the AIS standard also says that for very weak messages, a receiver should detect them in 20% of the cases (and not 100% for example)

So, if you slightly perturb a file or cut off a part, it will impact the messages that you will find. Specially for very weak messages on the boundary of what can be detected.

Nevertheless, I will have a more detailed look at the files. But might be difficult to find.

Can you give a bit more background on the files? Seems to be 8 bit stored as 16 bit.

[ekarseras]

Hi,

Thanks for replying. I have read your response.

The samples are 12-bit stored as 16-bit.