Correlating space object detections
Beam park radar observations are commonly used to determine how much space debris is out there. One points the radar in one direction, and measures for 24 hours. Each space object observation is recorded, and thus one obtains a statistical sample of the debris in space.
I've been working on a new beam park observation, and improving the coherent integration length that can be achieved. After writing some C and CUDA code, I've got a fairly fast range-velocity-acceleration matched filter. I also now have a good clustering algorithm, which groups together individual detections. The last step is now to determine which objects are in the NORAD catalog and which are not. Most of the large objects are in the catalog, and their characteristics are relatively well known. Anything not in the catalog is likely to be space debris.
Either way, by removing all NORAD catalog objects from a beam park observation, one gets a better idea of what is the size distribution of small space debris. Even large objects can be detected as low signal to noise ratio targets. Large objects are far more likely to be detected in a side lobe of the radar beam, and they tend to dominate a beam park observation.
I've been working on the correlation algorithm, and it seems to work now. I require both the range and Doppler residuals to be very small. This seems to do the trick. It is far from perfect, but this allows us to remove all known objects from the beam park measurement, and continue with getting a better size distribution of small objects.
While we can correlate objects within our beam park measurement with the NORAD catalog, we cannot measure the orbital elements of objects not within the catalog. We only have a measurement of range and range-rate, which is not sufficient to determine orbital elements for the uncatalogued objects. However, if we use a well calibrated position for the radar location, we could potentially improve the orbital elements for known objects. We get about 2700 correlated range and range rate detections per day of observation!
One of the first things I discovered, is that the two low altitude clusters in my beam park measurement are flocks of CubeSats launched in 2017. I first thought that these might be new fragmentation events, but they are not -- unless you classify launching of hundreds of Cubesats as a fragmentation event. I'll probably write more about this later.
I've been working on a new beam park observation, and improving the coherent integration length that can be achieved. After writing some C and CUDA code, I've got a fairly fast range-velocity-acceleration matched filter. I also now have a good clustering algorithm, which groups together individual detections. The last step is now to determine which objects are in the NORAD catalog and which are not. Most of the large objects are in the catalog, and their characteristics are relatively well known. Anything not in the catalog is likely to be space debris.
Either way, by removing all NORAD catalog objects from a beam park observation, one gets a better idea of what is the size distribution of small space debris. Even large objects can be detected as low signal to noise ratio targets. Large objects are far more likely to be detected in a side lobe of the radar beam, and they tend to dominate a beam park observation.
I've been working on the correlation algorithm, and it seems to work now. I require both the range and Doppler residuals to be very small. This seems to do the trick. It is far from perfect, but this allows us to remove all known objects from the beam park measurement, and continue with getting a better size distribution of small objects.
While we can correlate objects within our beam park measurement with the NORAD catalog, we cannot measure the orbital elements of objects not within the catalog. We only have a measurement of range and range-rate, which is not sufficient to determine orbital elements for the uncatalogued objects. However, if we use a well calibrated position for the radar location, we could potentially improve the orbital elements for known objects. We get about 2700 correlated range and range rate detections per day of observation!
 |
| Signal to noise ratio as a function to range and time, and annotations of all NORAD objects passing the beam less than 25 degrees off axis. Correlated detections are marked with red, and uncorrelated (objects not in the catalog) are marked with green. Because objects can be detected so far off axis, a large fraction of the beam park observations are actually NORAD cataloged objects larger than 10 cm. |
 |
| The same figure, but without annotations. |
One of the first things I discovered, is that the two low altitude clusters in my beam park measurement are flocks of CubeSats launched in 2017. I first thought that these might be new fragmentation events, but they are not -- unless you classify launching of hundreds of Cubesats as a fragmentation event. I'll probably write more about this later.
 |
| Flocks of CubeSats in the beampark observation marked with circles. |
Nice
ReplyDeleteGreat!
ReplyDelete