All Starlink captures were centered in the downconverted L-band at 1450 MHz using the HIGH side of the LNB, meaning all were centered at the original Ku-band at Fcr = 12050 MHz. Each capture's sampling rate is embedded in the file name. This is necessary because the dual-capture application doesn't produce an XML file. For example starlink_20230427_1_45Ghz_62_5Mhz is sampled at 62.5 MHz. All GPS captures were centered at L1 (1575.42 MHz) and sampled at 5 MHz with the iq16s bitpacking convention. GPS captures have been carefully processed to obtain accurate time stamping of the GPS capture samples (to within a few ns). Since the GPS and Starlink captures were locked to the same clock, this implies that the Starlink sample timing is known to within a few nanoseconds with respect to GPS time. A detailed description of each data set appears below in the following order: starlink_20230427_1_45Ghz_50Mhz_001 starlink_20230427_1_45Ghz_50Mhz_003 starlink_20230427_1_45Ghz_55Mhz starlink_20230427_1_45Ghz_58Mhz_001 starlink_20230427_1_45Ghz_58Mhz starlink_20230427_1_45Ghz_62_5Mhz ---------------------------------------------------------- starlink_20230427_1_45Ghz_50Mhz_001: STARLINK-1550 (v1.0) t0 = 1682617170 t (sec) SNR (dB) beta _______ ________ ___________ 0 -2.6392 -4.8524e-06 5 -6.3178 -2.4262e-06 10 -8.1378 0 15 5.5779 0 20 3.1526 0 25 3.1153 2.4262e-06 30 -1.3349 2.4262e-06 35 -1.0756 4.8524e-06 40 -1.807 7.2787e-06 45 -3.5384 7.2787e-06 50 -2.7396 9.7049e-06 55 -3.9707 7.2787e-06 60 3.4601 9.7049e-06 65 4.9324 1.2131e-05 Amazing data set with 4 beam switches! 0 - 12.43 seconds: Low-SNR composite signals *beam switch* 12.43 - 27.43: ~7 dB SNR. Clock jitter normal. Unusually stable clock: adjustments are less than 50 ns. *beam switch* 27.43 - 42.43: 2.5-5dB SNR. Composite signal. From 32.22 to 35.7 seconds, signals from two different beams (on same channel!) are so close in SNR that they can't be distinguished. Surely the RX is not the target of either beam. Maybe they're targeted to cells on either side of the RX's cell. *beam switch* 42.43 - 57.43: ~4 dB SNR. Strange wall-to-wall pattern. Clock jitter normal. Good stable clock. *beam switch* 57.43 - 67.6: ~6-7 dB SNR. Standard pattern. Nice clean signal. Clock jitter slightly abnormal, with some clock deviations that look strange even though the train of correlation peaks is perfectly clear and regular. I believe this clearly indicates that even when the signal is strong (maybe primary beam), you can get some clocks that aren't so good. ---------------------------------------------------------- starlink_20230427_1_45Ghz_50Mhz_003: STARLINK-5113 (v1.5) t0 = 1682617728 t (sec) SNR (dB) beta _______ ________ ___________ 0 4.3717 -1.2131e-05 5 3.7763 -1.2131e-05 10 1.9825 -9.7049e-06 15 1.0673 -9.7049e-06 20 1.7945 -9.7049e-06 25 -9.5736 1.6984e-05 30 6.7081 -4.8524e-06 35 5.88 -2.4262e-06 40 -0.63865 -2.4262e-06 45 -0.39689 0 50 -2.2186 0 55 -9.5513 -1.6984e-05 0 - 8.67 seconds: ~7 dB SNR. Good steady pattern. Non-composite. Nasty large jitter with a sawtooth pattern. Huge clock rate adjustment at 6.53 seconds. *beam switch* 8.67 - 23.67 seconds: 4-6 dB SNR. Can't tell if it's composite or just an example of power modulation into the same cell. If I take only the strong correlations, which occur near the beginning of bursts of frames, I get a high-jitter (+/- 30 ns) sawtooth pattern. If I take all clean correlations, I get the same level of jitter just more densely filled in. In other words, I don't think it's composite because the clock behavior of the full ensemble is identical to that of the strong signals, which are 1.5x higher in amplitude (3.5 dB). But again I must explain the terrible sawtooth clock pattern. Just a really terrible clock? But why such a regular sawtooth? *beam switch* 23.67 - 38.67 seconds: ~7 dB SNR. Strong non-composite signals but sparse in time. Maybe this is a "heartbeat" segment? Clock jitter is +/- 15 ns at the beginning of the segment, reducing to +/- 10 ns at the end. That's interesting! The clock jitter seems to get better over this segment. The SNR also rises, from about 8 dB at the beginning to 9.3 dB at the end. But 8 dB is plenty to make a clean frame arrival timestamp! And 9.3 dB isn't considerably better than 8 dB. So what happened? Did the clock just get better? *beam switch* 38.67 - 53.67 seconds: 2-4 dB SNR composite signal. Pretty high jitter throughout. *beam switch* 53.67 - 57.2 seconds: Nothing much there. ---------------------------------------------------------- starlink_20230427_1_45Ghz_55Mhz: STARLINK-1707 (v1.0) t0 = 1682616494 t (sec) SNR (dB) beta _______ ________ ___________ 0 3.2279 -4.8524e-06 5 0.19378 -4.8524e-06 10 -0.13631 -4.8524e-06 15 0.29135 -2.4262e-06 20 -3.3428 -2.4262e-06 25 -4.8481 0 30 -0.64477 0 35 -0.74564 2.4262e-06 40 -2.1028 2.4262e-06 0 - 12.54 seconds: 6.5 dB SNR but decreasing to 4 dB. Clearly composite, as evidenced from correlation analysis. Lots of jitter. *beam switch* 12.54 - 27.54 seconds: 5 dB SNR but decreasing to 1.5 dB. Clearly composite. Increasingly hard to distinguish a dominant stream. Did not attempt to track frame timing. *beam switch* 27.54 - 42.54 seconds: 4 dB SNR. Somewhat sparse. Also composite, but a clear dominant stream. One can see at least two minor streams, one thst is closely time-aligned with the dominant stream and one that is off by ~150 us. Could these be coming from a different SV? I don't think so -- we have a fairly narrow beam width and we see fairly constant amplitudes over time. Lots of jitter. *beam switch* 42.54 - 43.3 seconds: 2 dB SNR. Dense. ---------------------------------------------------------- starlink_20230427_1_45Ghz_58Mhz_001: STARLINK-3532 (v1.5) t0 = 1682616397 t (sec) SNR (dB) beta _______ ________ ___________ 0 -7.5261 3.1541e-05 5 -10.214 -3.6393e-05 10 -3.3891 -4.8524e-06 15 -1.6641 -2.4262e-06 20 -1.3658 -2.4262e-06 25 -3.0832 0 30 -3.2676 0 35 -1.2136 2.4262e-06 Good data set. Three clear beam switches. Also interesting because at 24.44 seconds, which is at one of the 1-second boundaries, there is a radical change in the frame Doppler even though the frame timing is not discontinuous beyond what is normal at the 1-second boundaries. This proves that SpaceX is willing to make radical changes to the frame clock Doppler at the 1-second boundaries. 0-5.19 seconds: Low-SNR composite signals *beam switch* 5.19 - 20.19: ~3 dB SNR signals with various patterns. Clock jitter normal: +/- 5 ns. *beam switch* 20.19-35.19: 2.5-3dB SNR. Clock jitter reasonable: +/- 6 ns. *beam switch* 35.19 - 36.7: ~ 2 dB SNR. Terrible clock jitter: +/- 15 ns ---------------------------------------------------------- starlink_20230427_1_45Ghz_58Mhz: STARLINK-1094 (v1.0) t0 = 1682616087 t (sec) SNR (dB) beta _______ ________ ___________ 0 -10.607 1.4557e-05 5 2.2674 -4.8524e-06 10 3.7152 -2.4262e-06 15 -8.802 -2.4262e-06 20 -8.2296 0 25 -10.497 0 30 -9.9526 -2.4262e-06 0 - 14.68 seconds: 6 dB SNR. Sparse frames. High Jitter. *beam switch* 14.68 - 29.68 seconds: Weak signals. ~1 dB SNR. Not worth analyzing. *beam switch* 29.68 - 32.4 seconds: Weak signals. ~1 dB SNR. Not worth analyzing. ---------------------------------------------------------- starlink_20230427_1_45Ghz_62_5Mhz: STARLINK-3610 (v 1.5) t0 = 1682615750 t (sec) SNR (dB) beta _______ ________ ___________ 0 6.5234 -4.8524e-06 5 6.9094 -4.8524e-06 10 4.8992 -2.4262e-06 15 3.3269 0 20 2.3694 2.4262e-06 25 8.591 2.4262e-06 Good data set with decently strong signals and two beam adjustments over a 29-second interval. t0 = 0 beta = -4.245883e-06 SINR = 5.7 dB Moderately strong signals at beginning (~7 dB SINR after filtering to remove edge effects) with dense traffic. After a gap of about 16 frames, a new traffic pattern begins at 7.171 seconds with SINR = ~5 dB. This is the beginning of a new Fixed Assignment Interval (FAI). Sparser pattern persists until 19.71 seconds, at which point a slightly denser pattern emerges, but not a new FAI. SNR has decreased to approximately 2.4 dB by this point. Clear FAI change at 22.173 seconds (15 seconds after most recent start of FAI), with power rising to SINR of 8.6 dB. Q: Signals are coming from the same SV throughout, otherwise our narrow-beam dish would not have tracked them so reliably. We see two "beam switches" here. Yet we seem to be in the primary beam the whole time! But neither one seems to be switching *away* from RX cell to another cell. Could it be that they do both fine and coarse beam adjustments? In fact, I have no evidence for their doing fine adjustments! This data set calls into question my claim that switching to a neighboring cell drops power by 5-6 dB. What if it's just an adjustment of the primary beam into the current cell? Note how quiet it is between beam switches. Maybe because *all* SVs are quiet during this interval? Some clues from the timing analysis: The first segment, from 0 to 7.1 seconds, has what could be called "normal" clock jitter: +/- 5 ns or so, which is close to the sample quantization 4.1 ns = 1/240MHz. The second segment, starting at 7.2 seconds and extending till 22 seconds, has at least double the clock jitter (+/- 12 ns or so). The final segment, from 22.4 to 28 seconds, has extremely intense clock jitter: +/- 35 ns! What's going on? And this jitter is not caused by low SNR -- the SNR at the end is 9.5 dB, which is plenty good. Todd's conjecture: We know there are three downlink phased array antennas on each SV. Maybe each one has its own frame clock! ----------------------------------------------------------