Originally foreseen before the ejection of BEESAT-4, the commissioning of AVANTI could only take place after the separation. This delay was in a sense better for the experiment, since it allowed us doing the preliminary functional tests with a real target in the field of view of the camera. The first attempt to observe the picosatellite was performed only twelve days after the separation. At that time, the cubesat had already escaped to a distance of several dozen kilometers. Would it be possible to see such a tiny object at such a big distance?
The picosatellite had fortunately been tracked by NORAD since the separation. As a result, it had still been possible for the flight dynamics team to maintain a loose formation based on Two-Line Elements (TLE), at a separation of about 30 to 60km. However, the poor accuracy of the TLEs did not play in favor of a close formation. As a result, it had been decided to stay at such a safe distance of BEESAT-4 in order to avoid any risk of collision.
On September 21st, the satellite was finally ready to point one of its star cameras in flight direction to search for the picosatellite. At such a distance, it is very difficult to find such a small object at the first glance, because, if visible at all, BEESAT-4 is going to be very faint. The early radiometric investigations had foreseen a visibility up to 20km separation. However, the results of these analyses were likely to be affected by a large error, because this estimated maximal distance depends on the characteristics on the surface of the picosatellite (not precisely known), on the sensitivity of the star camera (not precisely known) and on the incidence angle of the Sun and the observer to the illuminated surface (not known).
Detection of BEESAT-4 at 40km distance
In order to find the target object, a good search algorithm is needed. In fact, is not enough to simply recognize the stars present in the image to detect the cubesat, because the picture may contain many unrecognized items (like the stars not available in the star catalog or other satellites). The approach retained for AVANTI is based on a kinematical analysis of the relative motion based on several pictures: since BEESAT-4 is flying on a similar orbit, its apparent motion is likely to be very peculiar and thus easily detectable.
First good surprise of the experiment: it is possible to see the picosatellite at more than 40 km distance!
Finding optically the target is only part of the problem. The other part consists in estimating the relative state of the formation based on the optical measurements. And this can be tricky, because the problem of angle-only orbit determination is only weakly observable. One of the most important activities of the commissioning is to get acquainted with our ground-based precise relative orbit determination, which will accompany us throughout the whole experiment. Contrary to an onboard real-time algorithm, the main advantage on ground is the possibility to consider the whole history of observations to ensure the consistency of our solution and to used maneuver calibrated with GPS for a better accuracy. Another important aspect is the possibility to perform several iterations and have the user in the loop for better data screening. As a result, the ground-based products are considered as more robust and more accurate than the onboard navigation and will in fact serve as reference to assess the performance of the onboard algorithms.
First bad surprise of the experiment: only few measurements are available. As seen on the figure, the camera cannot see the picosatellite during eclipses and is blinded by the Sun during another large part of the orbit. The simulations had predicted two short time intervals during which the picosatellite would be observable. The reality is even worse: BEESAT-4 is only visible at one single location of the orbit during only 11 minutes. Nevertheless, this should not prevent us from trying to perform an orbit determination based only on optical measurements.
The first vision-based relative orbit determination has been run over five days (September 21st to 25th). During this time, the formation had been drifting slowly between 40km and 46km, experiencing only one formation maintenance maneuver. Of course we do not expect great precision of the relative positioning products at such a distance with so few maneuvers (since the maneuvers are helping improving the observability). Consistent line-of-sight residuals of about half a pixel give us however good hope that the relative orbit determination is working properly. Based on the formal covariance matrix, our tool claims to be accurate to (14m, 700m, 17m) in the (Radial, Tangential, Normal) orbital frame, which is already a great improvement with respect to the TLEs. The next step would be to practice some approaches, during which a significant enhancement of the accuracy of the relative positioning products is expected. As part of the commission, a ground-based radar campaign is also planned to help us assessing independently the accuracy of our tools. But this is another topic. Keep posted for more news!
Line-of-sight residuals over the 5 day data arc