[kqbx]

The article is light on details, but we can make some guesses based on the label "10s integrations, 1Hz channels" in the plot. I assume they have a bank of 1Hz filters, and they split the output of each filter into 10s intervals, and somehow combine ('integrate') each 10s chunk into a single number for each bin.

They need to compensate the Doppler shift so that the signal stays in one bin over the 10s integration time. I imagine they are using non-coherent integration (basically computing total signal energy over 10s in each bin) to take into account that the doppler compensation is not perfect (if it was, you could have 0.1Hz bins with 10s integration time).

If the above is true, then yeah, they can't demodulate any data because the integration time is much longer than symbol duration.

I wonder if with more accurate Doppler prediction, you could get an ever longer integration time and narrower bins, and thus even bigger SNR gain, perhaps allowing signal detection with a smaller dish...

[thomas_tel]

The "live" plot (with the peak) uses indeed 1Hz bins (of 1 second), that we average over the last 2-3 minutes to reduce the noise. We could go even narrower, I might give that a try on the recorded data.

[kqbx]

Thank you for the clarification. Recently I've been reading a lot about tracking of space objects (though much closer ones, on LEO/MEO), so this is some very interesting stuff for me!