I did a detailed signal analysis of the GOES-11 weather satellite dataset and posted it to my baudline blog.
The setiQuest GOES-11 dataset is a smorgasbord of signals, mirror aliasing, and the return of an old mysterious friend. Baudline's new IQ display was used to classify the many different modulation types.
Fascinating analysis and set of signals. My favorite, although simple with respect to others, is pasted below this text. By the SETI Institute's criteria, this might be considered a candidate signal: it is Doppler-drifted and narrow-band. The modulation scheme is very interesting. A detector such as SonATA would efficiently integrate along the time axis except for the infrequent "pulse in frequency" excursions. If those pulses in frequency had different intervals they could convey information. Hmmnn ... :-)
SonATA shouldn't have any problem detecting this signal but it might reject it because of its low +0.01 Hz/sec drift rate. I wonder what SonATA's lower threshold is? [SonATA developers please comment]
You're right that this signal could carry information by manipulating the pulse timing. It would be a sort of Pulse Position Modulation (PPM) or Pulse Width Modulation (PWM) or On-Off-Key (OOK) modulation. A disadvantage, in this situation, is that none of those would be an efficient use of bandwidth. But that might not be ET's governing motivator here, such a signal could be more a side effect of the generating system.
Going off on a tangent ... When I first saw this signal it made me think of the Kepler light curves. Multiple specially placed planets could generate a finite repeating sequence. More complex sequences could be generated by deviating from circular orbits and allowing N-body like interactions. Long term stability would be problem but I'm sure advanced alien mathematics could work that out. By playing with orbits ET could create a type of celestial graffiti. I wonder how many years of data the Kepler Team would need to be able to detect such a construction?
Yikes! Is that a histogram of sample values? If so, it should have a normal distribution.
Yes, that image is the histogram of sample values for the I channel. The Q channel looked the same.
The histogram would have a normal distribution if the signal was mostly noise. The GOES-11 dataset contains several strong signals that are modulated sine waves. Here is a histogram image I just made with baudline's Tone Generator of a sine wave + WGN. The measured SNR is +11 dB.
It looks very similar except for the lack of spurs.
It is not clear to me how to convert your dBm (which surely cannot be calibrated) scale to sample values, but if the visible extent of the tails are -128/+127 the attenuation levels were not even close to properly set for this high SNR observation. My first guess would be that range was exceeded in fixed-point math somewhere in the beamformer FPGA code.
The tails in the GOES-11 histogram are almost at the -64/+64 values. This means that only 7 of the 8 bits of quantization are being used. There is one whole bit of dynamic range headroom.
Clipping and fixed point math overflow would cause more severe looking distortions in the data. While the mirror aliasing is significant, the damage it is doing to the signals is much more subtle.
The apparent headroom is likely artificial (part of the problem) given that no values fall in that range. Here is a clearer unnormalized distribution of the real sample values for interested forum members to ponder: