troosh / AIS-catcher

Dual channel AIS receiver for RTL SDR dongles and Airspy HF+. Output is send as NMEA messages to screen or broadcasted via a UDP network connection.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

AIS-catcher - An AIS receiver for RTL-SDR dongles and Airspy HF+

This package will add the AIS-catcher command - a dual channel AIS receiver for RTL-SDR dongles and the Airspy HF+. Output is send in the form of NMEA messages to either screen or broadcasted over UDP. The idea behind AIS-catcher is that it should be easy to run various decoding model in parallel and read input from file to create an environment to test and benchmark different decoding models in a live setting. The program provides the option to read and decode the raw discriminator output of a VHF receiver as well.

AIS-catcher is created for research and educational purposes under the MIT license. It is a hobby project from an unqualified amateur and not tested and designed for reliability and correctness. You can play with the software but it is the user's responsibility to use it prudently. So, in summary, DO NOT rely upon this software in any way including for navigation and/or safety of life or property purposes.

Developments

Release version 0.12: Introduction of a basic container image for AIS-catcher for use with Docker, Podman and the like.

Release version 0.09: Some updates to improve stability when continuously running on a Rapberry Pi:

  • added threading with buffer overrun and timeout detection, faster downsampling, some stability improvements
  • v-switch allows for bespoke update frequency to keep logs maneagable
  • small improvement in reception

Purpose

The aim of AIS-catcher is to provide a platform to facilitate continuous improvement of receiver models. Any suggestions, observation or sharing of recordings for setups where the current models are struggling is highly appreciated! The algorithm behind the default receiver model was created in this way by investigating signals and trying different ways to get a coherent model running whilst keeping it simple at the same time. If I have some more free time I will try to expand the documentation and implement some improvement ideas.

Usage

use: AIS-catcher [options]

        [-h display this message and terminate (default: false)]
        [-s xxx sample rate in Hz (default: based on SDR device)]
        [-v [option: xx] enable verbose mode, optional to provide update frequency in seconds (default: false)]
        [-q surpress NMEA messages to screen (default: false)]
        [-u address port - UDP address and port (default: off)]

        [-r filename - read IQ data from raw 'unsigned char' file]
        [-r cu8 filename - read IQ data from raw 'unsigned char' file]
        [-r cs16 filename - read IQ data from raw 'signed 16 bit integer' file]
        [-r cf32 filename - read IQ data from WAV file in 'float' format]
        [-w filename - read IQ data from WAV file in 'float' format]

        [-l list available devices and terminate (default: off)]
        [-d:x select device based on index (default: 0)]
        [-d xxxx select device based on serial number]

        [-p xx frequency correction for RTL SDR]

        [-m xx run specific decoding model (default: 2)]
        [       0: Standard (non-coherent), 1: Base (non-coherent), 2: Default, 3: FM discrimator output]
        [-b benchmark demodulation models - for development purposes (default: off)]

Examples

To test a proper installation and/or compilation (see below), we can first try to run the program on a RAW audio file as in this tutorial:

wget "https://github.com/freerange/ais-on-sdr/wiki/example-data/helsinki-210-messages.raw"
AIS-catcher  -m 3 -v -s 48000 -r cs16 helsinki-210-messages.raw

AIS-catcher on this file should extract roughly 360 AIVDM lines. Notice that if the sample rate is set at 48 KHz with switch -m 3, AIS-catcher runs a decoding model that assumes the input is the output of an FM discriminator. In this case the program is similar to the following usage of GNUAIS:

gnuais -l helsinki-210-messages.raw

which produces:

INFO: A: Received correctly: 153 packets, wrong CRC: 49 packets, wrong size: 4 packets
INFO: B: Received correctly: 52 packets, wrong CRC: 65 packets, wrong size: 10 packets

To list the devices available for AIS reception:

AIS-catcher -l

The output depends on the available devices but will look something like:

Found 1 devices:
0: Realtek, RTL2838UHIDIR, SN: 00000001

A specific device can be selected with the d-switch like -d:0 or -d 00000001.

To start AIS demodulation, print some occasional statistics (every 10 seconds) and broadcast AIS messages via UDP, we can use the following command:

AIS-catcher -v 10 -u 127.0.0.1 12345

If succesful, NMEA messages will start to come in, appear on the screen and send as UDP messages to 127.0.0.1 port 12345. These console messages can be surpressed with the option -q.

The following commands record a signal with rtl_sdr at a sampling rate of 288K Hz and then subsequently decodes the input with AIS-catcher:

rtl_sdr -s 288K -f 162M  test_288.raw
AIS-catcher -r test_288.raw -s 288000 -v

Multiple receiver models

In the current version 4 different receiver models are embedded:

  • Default model (-m 2): the default demodulation engine
  • Base model (non-coherent) (-m 1): similar to RTL-AIS (and GNUAIS/Aisdecoder) with some modifications to PLL and main receiver filter (taken from here).
  • Standard model (non-coherent) (-m 0): as the base model with brute force timing recovery.
  • FM discriminator model: (-m 3) as the 'standard' model but assumes input is output of a FM discriminator, hence no FM demodulation takes place which allows AIS-catcher to be used as GNUAIS and AISdecoder.

The default model is the most time and memory consuming but experiments suggest it to be the most effective. In my home station it improves message count by a factor 2 - 3. The reception quality of the standard model over the base model is more modest at the expense of roughly a 20% increase in computation time. Advice is to start with the default model, which should run fine on most modern hardware including a Raspberry 4B and then scale down to -m 0or even -m 1, if needed.

To get a sense of the performance of the different models, I have run a simple test in two different setups whereby AIS-catcher ran the main two models in parallel for 3 minutes and we counted the number of detected messages.

Location: The Hague residential area with RTL-SDR v3 dongle and Shakespeare antenna with quite some (perhaps fair to say a lot of) blockage from surrounding buildings and antenna placed within a window, we have the following message count for various models and sample rates (over 3 minute run):

Model Settings Run 1 Run 2 Run 3
AIS-catcher v0.07 Default @ 1536K 153 131 152
AIS-catcher v0.07 Standard (non-coherent) @ 1536K -m 0 69 44 51
AIS-catcher v0.07 Default @ 288K -s 288000 111 136 117
AIS-catcher v0.07 Standard (non-coherent) @ 288K -m 0 -s 288000 47 49 35
AISRec 2.2 (trial) Sampling: very high, super fast 132 140 121
rtl-ais v0.3 @ 1600K -n 15 17 6

For completeness I performed seperate runs with AISRec and RTL-AIS as well. AISRec has some excellent sensitivity and is one of the most user friendly packages out there. It is highly recommended. RTL-AIS is a very efficient and elegant open source AIS receiver with minimal hardware requirements and is a pioneer in the field of open source AIS software.

The first two rows are ran in parallel (i.e. on the same input signal) and therefore are comparable. Same for the two rows following. The other runs are provided for information purposes and cannot be compared as message density fluctuates over time and have different system and hardware requirements. Nevertheless, these non-scientifically conducted experiments suggest that 1) the Default model can perform better than the Standard model and 2) a higher sampling rate should be preferred over a lower rate where possible.

Finally some results with direct sight at the beach (Scheveningen) with a higher message density over 3 minutes (NESDR dongle with factory included antenna):

Model 288K Ubuntu 1536K Ubuntu
AIS-catcher v0.06 Default 489 499
AIS-catcher v0.06 Standard (non-coherent) 417 439
AIS-catcher v0.06 Base (non-coherent) 408 429

The results for each column are comparable as based on the same input signal. Notice that with a better reception there is less of a difference between the different models and sampling rates.

Running multiple models

The command line provides the -m option which allows for the selection of the specific receiver models (AIS-catcherhas 4 tested models included and one so-called Challenger model - a possible release candidate). Notice that you can execute multiple models in one run for benchmarking purposes but only the messages from the first model specified are displayed and forwarded. To benchmark different models specify -b for timing and/or -v to compare message count, e.g.

AIS-catcher -s 1536000 -r posterholt_1536_2.raw -m 2 -m 0 -m 1 -q -b -v

The program will run and summarize the performance (count and timing) of three decoding models:

[AIS engine v0.07]		: 36 msgs at 45 msg/s
[Standard (non-coherent)]	: 3 msgs at 3.8 msg/s
[Base (non-coherent)]		: 2 msgs at 2.5 msg/s
[AIS engine v0.07]		: 3.4e+02 ms
[Standard (non-coherent)]	: 1.9e+02 ms
[Base (non-coherent)]		: 1.7e+02 ms

In this example the default model performs quite well in contrast to the standard non-coherent engine with 36 messages identified versus 3 for the standard engine. This is typical when there are few messages with poor quality. However, it doubles the decoding time and has a higher memory usage (800 floats) so needs more powerful hardware. Please note that the improvements seen for this particular file are an exception.

Releases

A release in binary format for Windows 32 bit (including required libraries) can obtained at request. Please note that you will have to install drivers using Zadig (https://www.rtl-sdr.com/tag/zadig/). After that, simply unpack the ZIP file in one directory and start the executable. For Linux systems, compilation instructions are below.

Compilation process for Linux/Raspberry Pi

Make sure you have the following dependencies:

  • librtlsdr and/or libairspyhf

The steps to compile AIS-catcher for RTL-SDR dongles are fairly straightforward on a Raspberry Pi 4B and Ubuntu systems. First ensure you have the necessary dependencies installed. If not, the following commands can be used:

sudo apt-get update
sudo apt-get upgrade

sudo apt-get install git make gcc g++ -y
sudo apt-get install librtlsdr-dev -y

Next step is to download AIS-catcher source and install the program:

git clone https://github.com/jvde-github/AIS-catcher.git
cd AIS-catcher
make rtl-only
sudo make install

If you want to include Airspy HF+ functionality, ensure you install the required libraries as descibed on https://github.com/airspy/airspyhf. The process to install AIS-catcher then becomes:

make
sudo make install

Container images

Pre-built container images containing AIS-catcher are available from the GitHub Container Registry. Available container tags are documented on the package's page, with latest (the latest release) and edge (the bleeding edge of the main branch) being the two main ones.

The following docker run command provides an example of the usage of this container image, running the latest release of AIS-catcher interactively:

docker run --rm -it --pull always --device /dev/bus/usb ghcr.io/jvde-github/ais-catcher:latest <ais-catcher command line options>

Alternatively, the following docker-compose.yml configuration provides a good starting point should you wish to use Docker Compose:

services:
  ais-catcher:
    command: <ais-catcher command line options>
    container_name: ais-catcher
    devices:
      - "/dev/bus/usb:/dev/bus/usb"
    image: ghcr.io/jvde-github/ais-catcher:latest
    restart: always

Note that the container images only support RTL-SDR at this time, and do not yet support Airspy HF+.

Considerations

AIS-catcher tunes in on a frequency of 162 MHz. However, due to deviations in the internal oscillator of RTL-SDR devices, the actual frequency can be slightly off which will result in no or poor reception of AIS signals. It is therefore important to provide the program with the necessary correction in parts-per-million (ppm) to offset this deviation where needed. For most of our testing we have used the RTL-SDR v3 dongle where in principle no frequency correction is needed as deviations are guaranteed to be small. For optimal reception though ensure you determine the necessary correction, e.g. see and provide as input via the -p switch on the command line.

On some laptops we observed that Windows was struggling with high volume of data transferred from the RTL SDR dongle to the PC. I am not sure why (likely some driver issue as Ubuntu on the same machine worked fine) but it is wortwhile to check if your system supports transferring from the dongle at a sampling rate of 1.536 MHz with the following command which is part of the osmocom rtl-sdr package:

rtl_test -s 1536000

In case you observe a high number of lost data, the advice is to run AIS-catcher at a lower sampling rate for RTL SDR dongles:

AIS-catcher -s 288000

To do

  • Documenting and finetuning the default decoding model
  • Ongoing: further improvements to reception and testing (e.g. improve coherent demodulation, filters, etc)
  • Performance improvements (align USB buffer size with multiple of 16K)
  • More testing: in particular in an area with high message density
  • Access to SDR hardware specific functionality, e.g. gain control
  • Windows GUI
  • Windows driver improvements
  • ....
  • ....

About

Dual channel AIS receiver for RTL SDR dongles and Airspy HF+. Output is send as NMEA messages to screen or broadcasted via a UDP network connection.

License:MIT License


Languages

Language:C++ 98.6%Language:Makefile 0.8%Language:Dockerfile 0.6%