adsb2dd

Convert ADSB data to delay-Doppler truth - see a live instance at http://adsb2dd.30hours.dev.

Features

• Provides an API to input receiver/transmitter coordinates, radar center frequency and tar1090 server.
• A web front-end calculator is provided to generate a correct API endpoint.
• Outputs JSON data with a delay in km and Doppler in Hz.
• Use the JSON output to map truth onto a delay-Doppler map, for example in blah2.

Usage

• Install docker and docker-compose on the host machine.
• Clone this repository to some directory.
• Run the docker compose command.

sudo git clone http://github.com/30hours/blah2 /opt/adsb2dd
cd /opt/adsb2dd
sudo docker compose up -d

The API front-end is available at http://localhost:49155.

Method of Operation

The delay-Doppler data is computed as follows:

• The tar1090 server provides the latitude, longitude and altitude of aircraft at the endpoint /data/aircraft.json - an example from a live server is http://adsb.30hours.dev/data/aircraft.json. The default data update rate is once per second. A timestamp is provided to match coordinates with a time.
• The bistatic range of each aircraft is computed using distance_rx_to_target + distance_tx_to_target - distance_rx_to_rx. The latitude, longitude and altitude is converted to ECEF coordinates which means distances can be computed with a simple norm.
• The bistatic Doppler by definition is the rate-of-change of the bistatic range. Unfortunately it's well known that differentiation amplifies noise - as the bistatic range data has a small amount of noise, the Doppler values have even larger noise. We also require a causal solution (dependent only on previous values) which means we can't use a more accurate Savitzky Golay filter. The approach here is to use less accurate moving average filter to smooth the bistatic rangedata prior to differentation.
• Currently computing a smoothed derivative by finding the median on the last k samples of the bistatic range vector. This is by no means optimal - however it seems to work reasonably well and follow targets with k=10. Note this is causal and generally slightly lags the truth since we're using previous samples unweighted.
• Future work will be to try and extrapolate/guess future bistatic range values (assume a constant acceleration) and apply the Savitzky Golay filter - I will call this pseudo-causal since I'm guessing future samples. I expect this will be a more accurate source of truth.
• On second thoughts, may make more sense to run a Kalman filter smoother (which is inherently causal).

The system architecture is as follows:

• The first API call to a set of inputs will result in a blank response {}. This is fine - the first API call adds the set of inputs to the processing loop.
• This approach allows multiple sets of inputs to run simultaneously on the same server.
• Refresh and if there are moving aircraft in the server, the delay/Doppler coordinates will be computed.
• The API provides a JSON output in the format {"<hex-code>":{"timestamp":<timestamp>,"flight":<flight-number>,"delay":<delay>,"doppler":<doppler>}}.
• If no API calls are provided for a set of inputs after 10 minutes, that set will be dropped from the processing loop.

Future Work

• Add a 2D plot showing all aircraft in delay-Doppler space.
• Add a map showing aircraft in geographic space below the above plot.
• Investigate algorithms to accurately compute smooth Doppler values.
• Some functional tests to ensure key features are working.

License

MIT