8-ary constellation bursts at 12800bps data rate (2)

Some other observations and updates about the S4539 12800bps 8-ary constellation already discussed here: this post was possible thanks to the collaboration of my friends AngazU, Christoph, Martin G8JNJ, and Sergio.

As shown in Figure 1, the polarity of mini-probes matches the 12800Ubps (6,6,2) setting so no doubt about the proper operation of the used decoders, primarily the Harris RF-5710A model.

Fig. 1a) 287 symbols preamble and sync sequence (red);
Fig. 1b) the actual "6,6,2" setting read from the preamble;
Fig. 1c) the theoretic "6,6,2" setting

Now look at the on-air symbols shown in Fig. 2: S4285 symbols (Fig. 2a) are exactly mapped to a PSK-8 constellation but the S4539 symbols being analyzed occupy different points (Figs. 2b,2c). It looks like a subset of the QAM-64 symbols is used for data  while the 4 "circled" points are the QPSK symbols of the mini-probes. Thus, since no interleaving and no coding are used in 6,6,2 mode (12800bps), the source data must be prepared such that after the scrambling the resulting 6-bit numbers will be mapped only to a 8-point subset of the QAM-64 outer ring. This makes sense and clarify the 12800bps speed, though we do not figure out why this is done.

 

Fig. 2

Figure 3 shows the plots of one frame obtained by Christoph: 256 data symbols + 31 mini-probe symbols: the 31 mini-probe symbols were descrambled and are at I=1,Q=0. As you can see the other points fit perfectly the 8 out-of-20 points of the QAM-64 outer ring [7 3 24 56 35 39 60 28].

Fig. 3 - 256 data symbols + 31 mini-probe symbols

These eight symbols have interesting structure: the 3,7,24,28 symbols are the same of  35,39,56,60 unless the left-most bit and they are at the same distance (32)

 3 000011
 7 000111
24 011000
28 011100

35 100011
39 100111
56 111000
60 111100

According to Christoph, the 6 bits are ABBCDD where ABC identify the point and D+B=1 mod 2. The ABC bits stream exhibit a 480-bit leghth period (Fig. 4).

Fig. 4

Back to the transmissions, our monitoring revealed that the entire sequence lasts about 36 seconds and consists of 6 "clusters", or "sets", each consisting of three channels with same spacing and arrangement:

Lately, our friend Martin G8JNJ noticed in the lower cluster A1 A2 A3 one weaker set (TDoA 100% St Eval) every 30 seconds (approx) and one set of stronger ones every three to five  minutes (approx) which he wasn't able to TDoA. "So that I think I'm hearing more than one transmitter site. It's proving to be very difficult to TDoA the second one, as they transmit much less frequently, but there is a big difference in RX signal strength between the transmissions", Martin says.

A friend of AngazU suggested that they could be developing some kind of turbo equalizer or similar. These emissions would be tests of a  training sequence and they would be measuring errors, convergence time and other parameters under different conditions. Just a guess, if they  succeed, we will see the  full constellation.

By the way, subjecting for example the F1 channel to the k500 decoder it prints out only 1536 decoded bits although it correctly recognizes the 12800U setting. As shown in Figure 5, each burst is made up of 13 frames for a total of 256x13=3328 QAM-64 data symbols that make 3328x6=19968 bits of data! (no interleaving neither coding is used in 12800U mode). Thus it seems that only one data frame (256 x 6) is processed by k500 (possibly the first one?): maybe it's a decoder limitation due the short burst duration? Note that it does not happen when I use the RF-5710A modem.

Fig. 5

(to be continued)