tl;dr:

• You could say we automatically tracked GOMX-3 cubesats while it tracked airplanes from space!
• found cubesats in total: GOMX-1, GOMX-3 and PRISM
• strong signals with wideband modulation are detectable easily
• weak small band "beacon" signals are easy to detect too, if no other wideband signal is transmitted on that channel. But stupig people gave me good ideas to solve it.
• asking the operators of the found cubesats for confirmation is pending... :)

long:

After monitoring the 2m-band for weather satellites, we told you will will check what is going on in the 70cm band. So we hacked in some frequencies from this CubeSats list to see, what will happen. The result you see in the video above. Within a 13 hours window on a cloudless day/night, you can clearly see the signals in the waterfall diagram (center) and in their zoomed windows (above).

The automatic detection algo processes the information shown in the zoomed windows. Inside, it applies an edge detection and the sum of all edges within the zoomed area is taken. With this, the signal level graph (on the bottom) is calculated. You see high peaks for strong signas and wideband modulation, and lower peaks for small band beacon signals. That is good news, because it means it is working as intended. The bad news (for now) is, that there is still some work to do for two cases.

1. If there is too much noise, there are also high peaks and the algo cannot distinguish between noise and cubesat signal.
2. if the channel is not only used by one satellite, which is often the case, then it's getting more difficult to detect strong sat signals from weak sat signals.

For case 1, we wil try to match the noise in one channel with the noise in the neighbouring channel. If there is noise in both channels at the same time, then it can be clearly detected as noise and not as a signal.

For case 2, we will try to apply the standard devition of the signal levels as the detection threashold two times. With the first iteration, we intend to detect the strong peaks and thus signals. And with the second iteration, these found strong signals will be ruled out and corrected standard devition will be calculated generating a ower standard devition and this threshod. And in that way, lower peaks can be detected leading to find small bandwidth signals. This idea was proposed by stupig,org members who were interested to hear what we do with DGSN. So we discussed it during their meetup at Stuttgart hackerspace shackspace.

Nevertheless, with the current method we were able to match the signal levels with three cubesats, namely GOMX-1, GOMX-3 and PRISM. We did that again with GPREDICT, but to be sure, we will ask them also if they can confirm our findings.

It was funny to find out, that we most likely tracked an airplane tracking cubesat and seeing them transmitting their data. That is GOMX-3's mission and you can all see the results on FlightRadar24. Pretty awesome, isn't it? :)