27 September 2022

unid 150-300Bd/1000 FSK (150-300Bd/500 same system?)

This FSK signal has the same characteristic switch of the modulation speed (150Bd during reversals, 300Bd in traffic mode), but a different shift (~1000Hz instead of ~500Hz), compared to the one seen here (150-300Bd/500Hz FSK). The different speeds are well visible in the 93.33 ms raster in figure 2.  However it must be said that I rounded the value of the shift to the nearest thousand Hz, given that - in some parts and without filtering - the signal can vary from 997 to 1000Hz. By the way, the same rounding was also applied to the measure of the shift  of the similar 150-300Bd/500Hz signal.

Fig. 1
The similarities are not limited only to the FSK parameters but also to the bitstream' formation (figure 2):
* 24-bit length period with the presence of a "phasing" bit (the column of "0s");
* initial sequence generated by the polynomial x^12+x^10+x^9+x^3+1;
* even parity-check on the differential decoded stream. 

Fig. 2

Me and my friend cryptomaster also verified that the bitstream complies the same (8,24) check matrix generated with the polynomial x^7+x^3+x^2+x+1

1 0 1 1 0 0 1 0 1 1 1 1 1 0 0 0   1 0 0 0 0 0 0 0
0 1 0 1 1 0 0 1 0 1 1 1 1 1 0 0   0 1 0 0 0 0 0 0
0 0 1 0 1 1 0 0 1 0 1 1 1 1 1 0   0 0 1 0 0 0 0 0
0 0 0 1 0 1 1 0 0 1 0 1 1 1 1 1   0 0 0 1 0 0 0 0
0 0 1 1 1 0 0 1 1 1 0 1 0 1 1 1   0 0 0 0 1 0 0 0
1 0 1 0 1 1 1 0 0 0 0 1 0 0 1 1   0 0 0 0 0 1 0 0
0 1 1 0 0 1 0 1 1 1 1 1 0 0 0 1   0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0   0 0 0 0 0 0 0 1

that detects and fixes the polarity reversal of the ninth column of the bitstream and performs a H(24,16) coding (figure3).

Fig. 3

That said, we think that these two FSK waveforms belong to the same system although user and purposes are still unidentified. Monitoring is quite difficult since - apparently - a scheduling does not seem used (short transmissions have been heard "by chance" on 4535, 6511, and 8084 KHz).


20 September 2022

R&S GM2100/2200 waveforms' preamble and its similarity to Link-11 SLEW

Me and my friend ANgazu from radiofrecuencias.es have recently studied curious, and apparently unid, burst transmissions occurred for about two consecutive days (7-8 Sept, ceased on 9 morning) on 12431 KHz USB. The idea of this post just comes during the analysis of that signal, noting an interesting similarity of its preamble to the one of Link-11 SLEW. The post then retraces those steps: from the analysis of the bitstreams (and recognition of the waveform) up to the in-depth examination of the preamble sequences and their coding.
The used waveform is the quite common and well-known PSK-8 & 2400 Bd occupying a 3 KHz bandwidth. The bursts do not appear synched or following a certain timing pattern, even if the sequence highlighted in the FFT-spectrum/time of figure 1 seems to be used.

Fig. 1

The good quality of the recording (courtesy of ANgazu) allow its demodulation using the PSKn demodulator of SA. The resulting bitstreams have a 480/96 bit length period, corresponding to 160/32 tribit PSK8 symbols (figure 2). Since these are bursts, the initial part is likely to consist of one or more TLC sections used for transmitter level control and receiver AGC settling (1)

Fig. 2

The most interesting thing is that, after reshaping the bitstream to the 96-bit period, I found a 192-symbol  sequence which is the same as the one used in Link-11 SLEW acquisition preamble (figure 3). Indeed, quoting STANAG-5511 paragraph "[the preamble] consists of a 192 tri-bit known sequence generated from a pseudo random code [...] these symbols are not scrambled and are applied directly to the 8 PSK modulator".

Fig. 3 -  the evident matching between the two 192-symbols sequence (*)

Judging from the preamble, it could be said that the bursts are Interrogation Messages of  Link-11 SLEW... but actually their durations and the structures following the preamble are not the same: indeed, Link-11 bursts have a well defined 45-symbols length header [1] which is not reflected in the analyzed bursts:

Link-11 SLEW waveform structure (Figure B-9, Stanag-5511)

So, a question arises: what other waveform has a 192-symbol length preamble? 

Checking my blog I found a match to the preamble of Rohde & Schwarz proprietary waveforms implemented in their GM2100/2200 HF modem [2] (although the term GM 2100 refers to the physical modem, from now on I will use it to refer to its characteristic waveform).  The framing consists of a 192-symbol sequence preamble followed by one ore more data blocks each consisting of 64-symbols: 48 unknown symbols (coded data) + 16 known symbols (test sequences). The postamble terminates the data blocks and consists of a 64-symbol End Of Message sequence. Except for the presence of an initial TLC section(s), the total length is then a multiple of 64 symbols.

Fig. 4 - GM2100 "signal format" waveform structure

To thoroughly analyze the bitstream under consideration, it's useful to refer to figure 5 that shows the ACF/period of a GM2100 transmission heard some years ago: since the 2400 Baud, the 133.45ms value corresponds to 320-symbol, or 5 data blocks, period:

Fig. 5 - GM2100 133.4 ms ACF

Back to our bistream, after removing initial TLC section(s) and preamble, the remaining bits, once reshaped to a 64-symbol pattern, are consistent with the GM2100 framing of figure 4. Moreover, it's easy to see in figure 6 that the five 16-symbol test sequences are repeated and are "segments" of the longer 80-symbol sequence (*):
0 2 7 7 3 5 1 0 1 4 0 5 0 0 0 0
4 1 1 2 1 4 1 5 4 2 7 4 5 1 6 4
5 4 3 7 0 7 6 2 6 2 4 6 7 2 4 7
3 0 3 1 3 5 1 2 5 0 1 7 1 4 6 0
0 5 7 7 2 5 2 7 7 4 7 5 5 0 5 6

The repetition of the five test sequences causes the (48+16)×5=320 symbol length period shown in figure 5.
Fig. 6 - GM2100 burst, demodulated bitstream (*)

Thanks to ANgazu, another important confirmation of GM2100 was a 188-141A 2G-ALE handshake (heard in that same frequency) which allowed us to identify the user as the Italian "Guardia di Finanza" (GdF): as it's known, they make a large use of R&S equipments in their onshore/offshore stations. Unfortunately, the handshake was not followed by any data traffic.

After identified the waveform, we agreed that the evident correspondence between the 192-symbol preamble sequences of Link-11 SLEW and GM2100 shall be further investigated anyway!

R&S documentation [2] reports the "autobaud" facility of the GM2100 waveforms: "A great advantage of the transmission method employed by GM2100 is automatic detection of the received signal data rate by means of a code received at the start of reception. This means that the receiving data modem need not be told the data rate of the transmitting modem".
A fairly likely conclusion - in my opinion - is that  R&S - as usual in the practice - use Walsh Orthogonal Modulation sequences in the sync preambles for both syncing and coding data-rate, FEC, interleaver, and modulation used in the following data blocks. 
Walsh Orthogonal Modulation is accomplished by taking each three bits (tribit) symbol and selecting a corresponding 4-fold repeated Walsh sequence, represented as octal characters (0 will be 0, and 1 will be 4); the selected four element Walsh sequence is repeated 8 times to yield a 32-element Walsh sequence used for the sync symbols (Table I)

Table I Channel symbol mapping for sync preamble.

The 32-element sequences of  Table I are then modulo 8 added to the scrambling sequence  7 4 3 0 5 1 5 0 2 2 1 1 5 7 4 3 5 0 2 6 2 1 6 2 0 0 5 0 5 2 6 6 (Table II). The scrambling sequence, from what I could find, is chosen from the long pseudorandom sequence of (2^16 -1) bits generated by the LFSR  x^16+x^15+x^13+x^4+1 [3].

Table II

The receive node, knowing the beginning and duration of each sequence, first "removes" the randomizing sequence from the received signal and then the resulting symbols are determined by the maximum likelihood method. 

So, the reason of the corrispondences shown in figure 3 is that both L-11 SLEW and GM2100 code their preamble by using the same 32-element Walsh randomized sequences (2). Notice that a similar method is also used in the preamble pattern generation of MIL-STD-188-110/FED-STD-1052.
By the way, I processed the recording using two GM2100 decoders but the analysis of  the resulting bitstreams did not reveal the presence of some known protocol, specifically the expected RSX.25 (the R&S implementation of X.25) which constitutes the main payload of these waveforms. The duration of the operations (~2 days), their modality and the lack of "consistent" traffic, lead us think about the execution of some test or training.

Ultimately, the practice of coding preambles using 32-element Walsh randomized sequences can lead to inaccurate conclusions or even false positive IDs, especially when analysis is limited to preambles only.


(*) A comment about the bitstream of figures 3,6
SA is an amazing signals analyzer but it's not a "waveform decoder", this means that its PSKn demodulator does not sync on knowns sequences. When used on phase keyed signals, SA produces correct info and images (number of phases, angles, modulation speed, carrier frequency, ...) but for the same signal it returns different demodulated streams due to the inevitable phase-offset errors (figure 7). Hence, each output stream should be analyzed separately.

Fig. 7

(1) Existing HF radios were generally not designed with burst waveforms in mind. For example, MIL-STD- 188-141 military radios are allowed 25 ms to reach full transmit power after keying. While the transmitter radio frequency stages are ramping up, the input audio signal level is adjusted by a transmit level control (TLC) loop so that it fully modulates the transmit power. At the receiver, an automatic gain control (AGC) loop must also adjust to a new receive signal. To accommodate these characteristics of existing radios, the 3G burst waveforms begin with a TLC section of “throwaway” 8-ary PSK symbols that are passed through the system while the transmitter’s and receiver ’s level control loops stabilize. (Johnson, Koski, Furman, Jorgenson, "Third Generation and Wideband HF Radio Communications"). 

(2) According to Table II, the sequence of the link-11 SLEW acquisition preamble consists of the symbols 5,7,6,1,4,0 

7 0 3 4 1 1 1 0 2 6 1 5 1 7 0 3 5 4 2 2 6 1 2 2 0 4 5 4 1 2 2 6   5
7 0 7 0 1 1 5 4 2 6 5 1 1 7 4 7 5 4 6 6 6 1 6 6 0 4 1 0 1 2 6 2   7
7 4 7 4 1 5 5 0 2 2 5 5 1 3 4 3 5 0 6 2 6 5 6 2 0 0 1 4 1 6 6 6   6
7 0 3 4 5 5 5 4 2 6 1 5 5 3 4 7 5 4 2 2 2 5 6 6 0 4 5 4 5 6 6 2   1
7 4 3 0 1 5 1 4 2 2 1 1 1 3 0 7 5 0 2 6 6 5 2 6 0 0 5 0 1 6 2 2   4
7 4 3 0 5 1 5 0 2 2 1 1 5 7 4 3 5 0 2 6 2 1 6 2 0 0 5 0 5 2 6 6   0

[1] https://i56578-swl.blogspot.com/2018/09/link-11-slew-transmission-format.html
[2] https://disk.yandex.com/i/mh2Ev4Bo15Az4Q
[3] https://disk.yandex.com/i/zOmzgHgthPXqMA
[4] https://disk.yandex.com/i/lb5zWmqYic7gBA

16 September 2022

coastal HF Radars' embroideries

Interesting and in some way curious recording sent me by my friend "Sigurd from Netherlands". By varying the db level of the spectrum it is possible to see that actually the "embroidery" is composed of 3 distinct sweepers signals distinguishable by their different strengths, at least as they were received at my friend's site (figure 1).
Fig. 1 - spectral shapes at different db levels

Measurements of the main parameters, carried out separately on each sweep, return the following values:

bandwidth (sweep width): 100 KHz
sweep repetition time (PRF): 2 Hz, slope: 200 KHz/s (500 msec duration)
mode: FMCW - Frequency Modulated Continuous Wave, down-chirped

bandwidth: 50 KHz
sweep repetition time (PRF): 2 Hz,  slope: 100 KHz/s (500 msec duration)
mode: FMCW, down-chirped

bandwidth: 50 KHz
sweep repetition time (PRF): 2 Hz,  slope: 100 KHz/s (500 msec duration)
mode: FMCW, up-chirped

Fig. 2

Note that although the 3 sweeps have same rate, B and C are characterized by lower slopes: that's is due to their scanning width that is lower than the one of sweep A (50/100 KHz). 

Other than the sweep rate, the signals share the same center frequency value: probably they are in some relationship and belong to the same system/organization. The different signal strengths lead to think of different transmitting sites or the use of phased antenna sets, so that the effective radiation pattern of a certain signal is reinforced in the direction of the used receiver (I also add that I have only 15 seconds of recording available so it is not possible for me to know if and how the signals may vary). Since the observed working frequency (around 13 MHz) has been allocated by the International Telecommunication Union to support the use of coastal High-Frequency Radar [1],  those emissions are most likely sourced by some WERA (WavE RAdar) systems located in North Europe: this type of OTH-SW (Over-The-Horizon Surface Wave) radar can pick up back-scattered signals from ranges of up to 200 km [2].

WERA HF coastal radar system (https://www.researchgate.net/figure/WERA-HF-costal-radar-system_fig1_251860940)


[1] https://www.itu.int/en/ITU-R/terrestrial/fmd/Documents/Res%20612.pdf (ITU Resolution 612 of the 2012 World Radio communication Conference)
[2] https://www.radartutorial.eu/19.kartei/10.weather/karte012.en.html

11 September 2022

OFDM-48 and some comments about OFDM analysis with SA

I wanted to further investigate the OFDM-48 signal published a few days ago [1], in particular the "mode" of construction of the signal and consequently the correct PSK modulation used in the channels. Signals Analyzer documentation [2] reports that, according to the method of forming the data channels and pilot tones, it is possible to meet at least three modes of OFDM signals, actually there may be more modes but SA OFDM tool considers the most basic ones:

Mode A:
All channels are formed "as is", including pilot tones. In this case, the pilot tone cannot be chosen arbitrarily, and is assigned from a limited number of suitable candidates. A typical representative of this mode is the WINDRM 51-tone signal.

Mode B:
All channels are configured as potential pilot tones, any channel can be assigned as a pilot. A typical representative of this mode is CIS-12.

Mode C:
Mixed formation type, all channels are formed as they are, according to mode A. But the pilot tone (s) is formed in a special way, according to mode B. In this mode, any channel or channels can be assigned to the pilot . A typical representative is 188-110B-39 tone signal.

In general, it should be noted that mode B is typical for CIS signals and mode C for NATO signals, even if the 16-tone 188-110 App.A signal is formed according to mode B. Of course this is a subdivision that comes from analysis and practice, although it's quite confirmed.  Most likely, in hypothesis, the signals of modes A and C are formed using 2^n dimensional FFT/IFFT algorithms, and signals according to the mode B without these restrictions.

To appreciate their differences, I synthesized two distinct OFDM-48 signals (modes A and B) using the OCG (OFDM Calculator - Generator) tool [3] downloaded from the radioscanner.ru site. The synthesis process requires the calculation of the OFDM parameters ("Calculate") which will then used in the synthesis stage ("Synthese"). The input fields of the "Calculate" tag (see figure 2) must be filled with the appropriate values, among them I chosed the Fmin Fmax values according the bandwidth limits of the original signal after its direct translation (figure 1).

Fig. 1 - direct translation of the original recording

After entered the correct values for FFT_size/2, TonesMin/Max and Fmin/max, the "Calculate" stage returns the relative OFDM parameters (figure 2); it should be noted that the values of SymbolRate and DeltaFreq calculated by the tool correspond exactly to those desired, that is to those obtained at the time from the analysis of the "original" signal (respectively: 50 Baud and 62.5 HZ).

Fig. 2 - computing the OFDM-48 parameters

As said, the values returned by the "Calculate" stage must then be entered in the input fields of the "Sythese" tag to build the desired OFDM signal; values FFT_Size/2 and TonesTotal are the same of the Calculate stage. As shown in figure 3, I synthesized the OFDM-48 signals (MakeOFDM button) according to the A and B modes, both using PSK2 modulation and without pilot tone(s).

Fig. 3 - synthesis of the two OFDM-48 signals

So I went on to analyze the two OFDM signals with the GREAT ADVANTAGE of already knowing their main parameters, especially the "formation" mode and the used modulation (PSK2).
To the facts, if the "mode" used in the analysis match the one used during the formation of the OFDM signal then we will get the actual modulation used in the channels. The following figures 4 and 5 show this evidence: the PSK2 constellation (the modulation actually used in channels) appears only when the "modes" of the analysis and the OFDM formation match; otherwise, the DBPSK constellation appears.

Fig. 4 - analysis of the OFDM-48 PSK2 mode-A signal

Fig. 4 - analysis of the OFDM-48 PSK2 mode-B signal

As proof, I synthesized the same OFDM-48 signal but this time with DBPSK modulation (figures 5 and 6).

Fig. 5
Fig. 6 - analysis of the OFDM-48 DBPSK mode-A signal

This means that we can obtain the correct values of the baud rate, spacing and number of channels but if we do not know a priori the used modulation we could face a margin of uncertainty about it. The best way to fix such impasse is to isolate a single channel and analyze it as if it were a normal PSK-n signal but with the foresight to use the differential mode (Diff = 1) when studying its constellation (figures 7,8). However, this is not always possible because it depends on the quality of the recording.

Fig. 7 - OFDM-48 PSK2, single channel verification

Fig. 8 - OFDM-48 DBPSK, single channel verification

That done, I tried to trace back to the "native" sampling rate (SR) of the original OFDM-48 signal analyzed in the previous post [1].

(the following comments are from "Analysis OFDM with CP in SA versions" [4])
It is well known that in OFDM signals there is a concept called "native sampling frequency" (SR) which has to satisfy some principles:
- SR/Br = x
- SR/Sh = y
where Br is the symbol rate (Baud), Sh is the separation between channels (Hz), and x y are positive integers. The SR frequency has to be a multiple of Br, and its relation to the channel spacing is “native”. On the other hand, we speak of “independence” of the SR frequency and “native” and “non-native” SR frequencies, which must certainly have specific values.

Let's explain this. A given recording has been sampled at a particular rate: we speak of "independence" because that sample frequency does not have to correspond to the "native" frequency. This does not influence or affect the obtaining of the parameters of an OFDM signal, indeed if necessary, the value of SR can be resampled/recalculated as necessary. Since the "native" value of the sample rate is a multiple of both Br and Sh, if we calculate the exact values ​​of Br and Sh, we will have the possibility of estimating a set of SR frequencies SR1, SR2, SR3..., SRn that meet the requirements and in which at least one frequency of them will be the “native” one.

Back to the native SR, if its value is not known but the OFDM signal formation values LU and LG are (1), then the following formula can be used:

SR = (LU + LG) * Br

Well, OCG tool also gives the possibilty to get pairs of LU LG (just "Get LU,LG" tag) according the desidered values of the signal such as channels, shift, and Br. We can choose any LU LG pair but it is better to leave the signal as much as possible as it is, ie without "heavy" resampling and using a pair of values as close as possible to the pair found from the analysis of the original signal [1]. In this case I chose the pair LU = 228  LG = 57 (figure 9), therefore:

SR = (228 + 57) * 50 = 14250 Hz

as you see, the resulting SR frequency is almost the same of the one used when recording the signal.

Fig. 9

The analysis of the signal after its resample at 14250 Hz is shown in figure 10. Since it is assumed to be a CIS signal, and they usually use mode B, it can be said that the modulation used is DBPSK, even if mode A & PSK2 remains equally likely.

Fig. 10 - analysis of the resampled OFDM-48 signal

Without using OCG, a trick to obtain one or more "effective" SR frequencies is to multiply the value of the shift by an positive integer n, ie:
SR = Sh * n  
taking into account the bandwidth occupied by the signal, ie the resulting SR value must be equal to twice the  upper boundary (see Nyquist rate [5]). In this case n would be = 228.
(1) the relation between the duration of LG (guard interval) in samples and the duration of LU (length of useful information) in samples gives the factor K (or "Magic K", visible in the OFDM analysis results), since K = LG/LU. Defining a consistent value of K is one of the primary goal/task of the analysis of signals OFDM, knowing this factor it is possible to receive all much more precisely and faster. 


7 September 2022

Harris wideband operations (a bit "intruding" within the 7 MHz HAM band)

Wideband activity was heard at the end of August around 7 MHz (figure 1) using mainly Romanian and Greek KiwiSDR receivers, my friend KarapuZ sent me his recordings which are of a much better quality than mine and therefore more suitable for analysis. According my friend, this network was set up around March-April 2022 and is well audible in our area since the network is presumably deployed in the south-east of Europe.

Fig. 1 - wideband transfers

Waveforms, durations and signal sequences in my opinion point to Harris devices: they have in fact developed and implemented  a wide band ALE (WBALE) adaptive system that selects the best channel, the available bandwidth and the frequency offset required for optimal wideband communications [1]. As I already mentioned in some blog posts, Harris WBALE relies on 3G-HF STANAG 4538 Fast Link Setup (FLSU) to establish a wideband link:

- the calling station first places a call using STANAG 4538 FLSU to exchange profiles of the two linking radios’ and and negotiate a traffic waveform
– the standard FLSU Request PDU has a traffic type parameter; Harris uses a new value of this parameter (reserved but not defined in  STANAG 4538: see table 4.6.1-2 "second 6-bit argument field") to indicate that a wideband link is to be established
- the radios then use an additional handshake (not defined in STANAG 4538) to negotiate bandwidth and offset (from the assigned frequency, see figure 1) to be used, based on the results of the preceeding "spectrum sensing"  (1).

Figure 2 shows the timing diagram of all the signalling required for the Harris WBALE protocol: the timing diagram follows the one described in 188-141D App.G, even if the used waveforms are different!

Fig. 2 - wideband session timing and real-world wideband transfer

Traffic is exchanged using Harris proprietary WHARQ waveforms family, quite well recognizable by their "superframe" consisting of a STANAG-4538 BW6 preamble followed by 8 frames each characterized by a different miniprobe pattern.  An ACK PDU is transmitted by the receive station using a BW6 burst waveform. Figure 3 shows the main parameters of the WHARQ 2400 Bd 3-KHz bandwidth waveform.

Fig. 3 - WHARQ 2400 Bd 3-KHz bandwidth waveform

As I titled, the problem lies in the fact that one of the WB channels occupies about 12 kHz of the low part of the 7 HAM MHz band. It must be said that the 7 MHz band is primarily assigned to radio amateurs, however also shortwave broadcasters and land mobile users have primary allocations in some countries so amateur stations must share bandwidth with these users. 

The choice of the 7 MHz for such milcomms is probably related to the "primary and secondary users" concept [3] which divides the users into primary users (licensed) and secondary users (unlicensed): the first “own” the bandwidth allocation while secondary users are only allowed to use this spectrum in a non-interfering basis:

a) for WBALE primary user mode, stations that link for the purpose of transferring data will use a bandwidth and offset in each direction that is chosen to maximize the signal-to-noise ratio (SNR) with which transmissions in each direction are received. Stations will avoid interference with other stations within the same network, but will make no effort to prevent interference with other stations outside the network, except as a byproduct of optimizing communications within the network.  This can have at least two significant implications:
1. the bandwidth and offset used in each direction of the link may be different;
2. the stations may cause harmful interference to communications in other networks while themselves not experiencing harmful interference. 

b) in secondary user mode, WBALE stations will not (as far as is practical) cause interference to other stations outside the network that are operating within the same channel allocations used by the network. In particular, whenever a link is established for a wideband data transfer, the bandwidth and offset used for the link will be chosen so as to avoid interference with any transmission detected by either side. This is likely to require that the same bandwidth and offset be used in both link directions.  

As you may see in figure 1, the bandwidth and offset used in each direction of each logical link are the same, threfore, in my opinion, it seems that they use this portion of band (7 MHz) in secondary user mode.

Fig. 4 - a Rockwell Collins modem performing the spectrum sensing (2)

(1) To effectively utilize the allocated bandwidth, WBALE will need to listen to an entire wideband channel of up to 24 kHz, detect the presence of interfering signals on the channel that could render all or part of the channel unusable, and identify any portion of the channel that may still be usable even if the channel is partly blocked. This function is referred to as "spectrum sensing".

2) Initial Wideband ALE developing and testing was condected togheter by Harris and Rockwel Collins.
[3] William N. Furman, John W. Nieto, Eric N. Koski: The 10th Nordic Conference on HF (2013)

1 September 2022

CIS-48 calls & OFDM transfers


Interesting and good quality recording sent me by my friend Ary from UDXF group, whom I thank. The initial so-called CIS-48 probes/markers, that presumably(!) act as "calls", are sent at regular intervals and consist of four 50Bd PSK2 tones: notice the typical violation of the bit reversal structure, a "1" is inserted instead of a "0", when the system transits to the traffic condition (figure 1). 
Fig. 1 - the four PSK2 probes/markers

The following OFDM blocks clearly consist of two different sets of channels: 36 and 48, both using PSK2 modulation. As shown in figure 2, the parameters are exactly the same (50 Bd speed, 62.5 Hz spacing, LG, LU) and this leads to think of the same OFDM modulator but with 12 channels suppressed (off). Maybe - at least in this cases - the correct definition should be OFDM-36/48...
Fig. 2 - OFDM blocks 36 (likely 48-12) & 48

Although the classic 3300 Hz pilot tone is missing, the system is attributed to some CIS mil/gov organization.