THALES XL modem, channels time lag

 

Thales HF XL SALAMANDRE (1) test transmissions spotted by my friend Geoff (who kindly sent me his recordings) on the upper limit of the 6 MHz band. In these tests the modems are able to use 16 contiguous 3 kHz channels with 2 KHz spacing and occupies a bandwidth of about 80 KHz. As a rule, two stations are used. 

Each transmitted burst consists of an initial XL synchronization preamble (2), followed by 9 frames of alternating data and known symbols. Each data frame consists of a data block consisting of 256 data symbols, followed by a mini-probe consisting of 31 symbols of known data, as per STANAG-4539 (188-110C Ap.C).  All the sixteen channels use PSK8 modulation at the symbol rate of 2400 Bd, user data rate is measured as 3200 bps with Very Short interleaver. Figure 1 shows the analysis of the upper channel and the Thales 124-symbol length extended preamble.

 

Fig. 1

 

A necessary foreword to understand how XL modem works. From what is possible to know from public documents available on the web, the XL multi narrow band (MNB) waveform is implemented by a multithread software running in a Linux PC (figure 2): "[...]The input stream is coded, interleaved and sent to different individual modulators having different data rates, each attached to a single channel. The modems outputs are combined in a frequency division multiplex and then the composite signal modulates the HF radio" (3).

 

Fig. 2 - multi-narrow band “XL” modem approach (Thales - HFIA meeting, San Diego Jan. 31th, 2013)

I realized that, modifying the FFT values, it turns out that the 16 channels do not come out at the same time but rather sequentially; more precisely, by filtering the intermediate channels, it is possible to measure a time lag of 3200 μs between the first and the 16th channel (figure 3). Likely, the parallel to serial buffer at the receive modem shall cancel that delay.

 

Fig. 3

 

Assuming that the XL modem  approach of figure 2 is still used today, I wonder if that time lag - unless it's a something like a required feature - is due to the context switching between the threads, ie between the software-defined  modems. I think that the initial de-multiplexing is irrelevant, since it will start to feed the n modems after the sync preambles have been formed, ie the symbols that are transmitted first are those of the n TLC/AGC sequences.  

Since the process spawns as many threads as the channels to be used,  the time delay is expected to decrease as the number of channels decreases: and indeed that's what happens in case of use of 12 and 9 channels (figure 4).

Fig. 4

I want to say that it's just a my guess and it relies either on the accuracy of SA measurements and the modem approach of figure 2: obviously more samples are needed to support my guess; unfortunately, both for the very nature of these transmissions (trials) and for the difficulty in having broadband recordings, it is a bit difficult to recover/find similar files to be analyzed. It would be great if someone from Thales would came across this post and would shed a light on this behavior.

 

As a final note, since 16 is the maximum number of channels the XL modem can allocate (figure 5), the one shown above is definitely the best performance possible in terms of used bandwidth: that is, 16 contiguous channels in only 80 KHz; but - at the same time - it is also the worst case when considering the total time delay between the limit channels. 

Fig. 5

https://disk.yandex.com/d/WyHqZXAy3yzu1w 
https://disk.yandex.com/d/RsIhraUOmlHaWA

(1)  Système Avancé pour Liaisons HF Adaptatives Multi-bANDes Rapides et Efficientes (adaptable, bandwidth-efficient, multi-band high-speed and high-frequency communication system)

(2) The synchronization preamble consists of three parts:
- a Transmitter Level Control / Automatic Gain Control (TLC/AGC) sequence, including an optional sequence to provide offset between channels;
- the main synchronization preamble, compatible with STANAG 4539 preamble;
- an extended synchronization preamble, specific to HF XL.
This last part, not included when operating according to 4539 or ISB modes, is combined with the main preamble to carry all information necessary to the HF XL waveform, in particular information on modulation choice for each channel. 

(3) HF XL: ADAPTIVE WIDEBAND HF TRANSMISSIONS - NordicHF 2013

CIS MPSK-16, PSK2 500Bd

Interesting CIS transmission spotted by my friend KarapuZ, whom I thank for sharing. The signal is a wideband MPSK-16 waveform, with 2500 Hz channel spacing, and use a PSK2 modulation at the symbol rate of 500 Bd in each channel. The 16 channels transport the same data.

Fig. 1 - PSK parameters of a single channel

A better view of the decoded bitstream of a single channel is obtained by using a 36-bit length period (figure 2):

Fig. 2 - 36-bit period

The messages consist of 5-FGs groups and a separator character (111110). In my opinion each group is coded into a 5-bit Figures, (after the removal of the common trailing "0") and has a constant ratio "1s"/"0s" = 3/2, except the separator which consists of all ones bit (11111). Nothing changes if the reverse polarity is taken into consideration. By using the association:

11111 = space/separator
00111 = A
01011 = B
01101 = C
01110 = D
10011 = E
10101 = F
10110 = G
11001 = H
11010 = I
11100 = L

we get :

DBLIC DBLIC DBLIC BBBHI
DHCLG AGDHD DDAAC DHCLC
DHDLD EFEIL BHDLG EFDIC
DLCHD EFEIL DILCF GEEDD
DHCLF DIDLE EHGDI ILAFE
BCDGI LAHEF DIDLE EIDLG
DFHIL EHGFL GFEEA CDIGI
EFDIL DIDLE LEADC DFAIF
GFEEE LIHGH IHGGF AAHFI
BBBHI DBLIC DBLIC DBLIC

Notice the particular "format" that is used for the initial and last four groups of the message: some conjectures can be made about its purpose such as for example the signaling of SOM/EOM or the date/number of the message... anyway only hypotesis. The messages are probably off-line encrypted and maybe using the so-called OTP (one-time pad) technique: anyway, one could test the possible associations as for example the following A=0 - L=9 (or the reverse order) and see what it turns out:.

31982 31982 31982 11178
37296 06373 33002 37292
37393 45489 17396 45382
39273 45489 38925 64433
37295 38394 47638 89054
12368 90745 38394 48396
35789 47659 65440 23868
45389 38394 94032 35085
65444 98767 87665 00758
11178 31982 31982 31982

My friend Cryptomaster came to the same conclusions in his reply in radioscanner taking a different approach: ie 6-bit words consisting of 5-bit code + 1 parity bit. Probably the MPSK waveform is an evolution of the CIS PSK-2 serial tone waveform (same period, same bitstream pattern) [1]: given that the 16 channels transport the same data, they provide a better immunity to noise and interferences.

https://disk.yandex.com/d/uBpZwnz0dv47Rw

[1] http://signals.radioscanner.ru/base/signal172/ 

analyzing the HF network traffic on 5120 KHz (OS BiH)

 

First of all I want to thank IZ6BYY Alain from Martinsicuro (Italy) who allowed me to use his KiwiSDR receiver without time limits: I very appreciated. 
I monitored this Bosnian HF network (I logged them first time on 2016) for more than two weeks: transmissions occur almost exclusively in the morning, not on weekends, and start around at 0730 UTC, likely following a certain schedule. The traffic consists of standards-based email exchange:

– 141A 2G ALE for link establishment
– Up to 2400 bps modem (Serial 110A, 39-tone 110A App.B)
– STANAG 5066 & CFTP client used for reliable over-the-air data delivery
– Standard SMTP email protocols into the wired network

All stations are members of the same ALE network and use the 3-way handshake for link management. In a few cases, a link closure similar to that used in STANAG-4538 is adopted, i.e. the link is terminated by the called station and not by the calling one.
The analysis of the 5066 PDUs after the removal of 110A overhead, figure 1, show the use of HBFTP compressed files (Harris Basic File Transfer Protocol) which, as for 5066 Annex F, is used along with CFTP for transfers from one SMTP server to another. 

Fig. 1 - STANAG-5066 PDUs showing the use of CFTP and HBFTP protocols

After HBFTPn.gz files have been extracted and unzipped, the email headers finally emerge and allow a bit of "intelligence" (figure 2):

Fig. 2 - email headers

 

wmtuser@OMEGA.ok, wmtuser@CIKLON.ok
The email addresses reveal that the messaging system software, and most likely the connected radios, are provided by Harris Corporation: indeed "wmtuser" is the email address default name that is prompted by Harris RF-67x0W Wireless Gateway. The ".ok" e-mail domain name stands for Operativna Komanda or Operational Command,
 

received: from osbihbutmir, received from kstbrvspvo
The server name "osbihbutmir" must be split as OS BiH Butmir where OS BiH (Oružane Snage Bosne i Hercegovine) stands for Armed Forces of Bosnia and Herzegovina, and Butmir is a neighborhood in Ilidža municipality site of the AF Operational Command HQ. Similarly, the name "kstbrvspvo" could be formed by the acronyms KSTBR and VSPVO, where KSTBR may stand for Communications Systems and Technologies Brigade (Komunikacioni Systems i Tehnologije BRigada).
I also noted the server name "jovana" which may have been chosen to honour the memory of Jovana Divjak, a Bosnian army general who died on April 8th, 2021: but that's just a my guess.

X-Mailer
The underlaying PCs run a Microsoft OS, likely Windows 2000 Professional or Windows XP Professional; Outlook 11 is used to draft and send the emails by OMEGA while other nodes seem to use Outlook Express.

X-HSMTP
(likely Harris SMTP, Simple Mail Transfer Protocol) The routing rows show that the recipient node is at 1-hop distance (DP and NP values).

By processing the bitstreams is then possible to derive the the 5066 addresses of the nodes and associate them to the related stations names and 141A ALE addresses  (in brackets):

 

000.000.000.001 OMEGA (OMA)
000.000.000.006 ASTRA (ASA)
000.000.000.007 CIKLON (CIN)
000.000.000.011 GRANIT (GRT)
000.000.000.016 LI(?)A (LIA)
000.000.000.017 LI(?)1 (LI1)
000.000.000.029 ORKAN (ORN)

 

It must be noted that:

1) the 5066 address range 0.0.0.0 — 0.255.255.255 does not have a Regional Assignee, rather the actual block allocation for Bosnia-Herzegovina is 6.6.y.z ( STANAG-5066 Annex N);

2) during the monitoring period I have not heard any other station or ALE address other than those listed. 

 

We also might compare the current station names with the old ones in use in year 2016, assuming that the 5066 addresses of the stations have not changed; notice that at that time the 141A ALE addresses were assigned  by using some popular automotive brands (HFMREZA is the Bosnian translation for HF Nerwork):

000.000.000.001 GAMAHFMREZA (GAMA)
000.000.000.003 FIATHFMREZA (FIAT)
000.000.000.005 FORDHFMREZA (FORD)
000.000.000.007 OPELHFMREZA (OPEL)        
000.000.000.009 SKODAHFMREZA (SKODA)   
000.000.000.011 VOLVOHFMREZA (VOLVO)        

Searching in the UDXF logs, this network appears for the first time in 2014: even in that case the ALE addresses were formed by the union of the first two and the last letter of the station names (TAO = TAngO), the latters consisting of the letters of the Greek alphabet: ALA (=ALFA), BRO (=BRAVO), DEA (=DELTA), GOF (=GOLF), EKO (=ECHO), OMA (=OMEGA), OSR (=OSCAR),TAO (=TANGO), ZUU (=ZULU).

The particular 5066 address (.001), the site (the AF Operational Command HQ), the traffic (OMEGA almost always initiates the ALE sessions) and the software too (Outlook 11 rather than Outlook express), led to think of OMEGA (OMA) as the net-control station as it was for the station GAMA. In addition to the change of station names and addresses, the most relevant change compared to 2016 is the paradigm used for emails: PEM - Privacy Enhanced Mail is now used for secure that traffic (figure 3).

Fig. 3

 

In some cases the contents of the emails are in clear-text, as for example the list of telegrams received/sent by the DK brigade (DK brTP) along with the greetings (*** Greetings from the team DK brTP OS BiH **** ) and the name of the operator of the "Workstation DK 6.pbr"; due to privacy, I have masked his surname:

 

Fig. 4

As said above, in some links the messages are also exchanged using the 39-Tone (110A App.B/FED-1052B) as HF waveform: this evidence proves the use of (at least) two radio networks where all or a subset of the nodes are members of both nets; at this regard, it's to be noted that Harris RF-6750 WG does not allow the use of multiple waveforms/protocols in a same radio-network. Likely, the HF email domain ".OK" coincides with only one radio network.
(yep I know, it's not good and it's definitely not discreet! anyway - to better illustrate my hypothesis - I had recourse to a old copy of the 6750 WG to simulate the software setup that I imagine and which in my opinion comes closest to the configuration which they use)

Fig. 5 - two distinct radio-networks with different waveforms/protocols

Another interesting point is that some bitstreams carried by the 39-tone and 188-110 modems have initial 41-bit length similar patterns (figure 6) that - in my opinion - reveal the use of encryption, therefore in those cases 5066 PDUs are not readable. I tried some analysis of the patterns and maybe they could be "partial" strings of the sequence generated by the polynomial x^42+x^41+x+1. 

Fig. 6 - 41-bit patterns in Serial 110A and 39-tone bitstreams

As far as the encryption device is concerned, my guess is that some links use Datotek encryption which is used in Harris RF-5022 and RF-5800 based radio stations. In that regard, I did some research in the web and found that as early as 2009 they were just using Harris RF-5022 transceivers during their participation as a PfP country in the "NATO Combined Endeavor" 2009 exercise (1). If my guess is correct, the 41-bit sequences could be a kind of "distinctive sign" of the Datotek encryption.

Fig. 7 - Datotek encryption may be used in RF-5022 based radio stations

https://disk.yandex.com/d/B0wHSW1Tfhz2Eg

(1) Bosnia and Herzegovina joined the Partnership for Peace (PfP) programme in 2006.At the beginning of 2021, Bosnia and Herzegovina established the Commission for Cooperation with NATO in order to facilitate the development of their Reform Programme for 2021-2022 and other matters on their path to accession.  
http://mod.gov.ba/Zdruzeni_napor/?id=21449 

 

a strange (if not wrong) use of STANAG-5066

29 Septembere 2021, update

I was WRONG!
Starting from RF-6710W Wireless Messaging Terminal (WMT) v5.0, Harris added new adaptive data rate support for existing HF modem waveforms so that customers can leverage their hardware investments with their new radio purchases. These waveforms include the widely used Parallel tone (39-tone) and STANAG 4285 Coded waveforms. Looking at the 39-tone demodulated stream, it could be that a crypto device is in the path, ie between a pc running STANAG-5066  and the modem

thus the STANAG-5066 PDUs are not in clear text and then no more visible.

24 Septembere 2021
These days I am monitoring some signals on 5120 KHz (from Serbian-Mil) that will be the subject of a next post, and I noticed a strangeness in the use of STANAG-5066 in relation to the "way" the data are sent. The waveforms are shown in the waterfall of figure 1.

Fig. 1 - the waveforms into play

As known from the "operation" of STANAG-5066, before the Data Transfer Sublayer (DTS) forwards the data, the Channel Access Sublayer (CAS) provides the functions necessary to access the physical channel (figure 2), ie the radio spectrum, assuming that the selection of the frequency (physical link setup) is handled by an external process such as ALE, in this case 188-141A. Then, a soft-link session (1) will be started immediately when there is data available for transmission to a remote STANAG-5066 node (2).

Fig. 2 - Physical Link Request PDUs related to the S5066 of figure 1

The strangeness is that, in that sample, data are not sent by STANAG-5066 DTS & 188-110 but rather using M-39 (188-110 App.B), ie outside STANAG-5066 (remember that STANAG-5066 is a data-link protocol, NOT a waveform).

 (1) To explain the difference between a physical channel and a session: when a client wishes to send an email to a remote client a physical link is established by the CAS; then a soft-link session is set up over this physical link. The soft-link session is between the local and remote client while the physical link is between the local and remote node. 

 
(2) A correct sequence of operations is shown in figures 3,4 (physical link request/accept, data forward, ACKs, physical link break): all managed by the STANAG-5066 sublayers that is carried by the same underlaying HF wavefrom. Notice the switch of the node address.

 

Fig. 3

 

Fig. 4

Frames captured thanks my STANAG-5066 off-line dissector.

19.5Bd/100 FSK: Rus R-397 “IRTYSH” (“ИРТЫШ”)

About one years ago (may 2020) I came across a short 19.5Bd/100 FSK [1] which, at that time, I chalked up to a some Russian-Mil network. Well, an anonymous reader of the blog, whom I thank, left a link about it [2], without further comments or suggestions: the link refers to an interesting topic posted on the airbase.ru forum about the HF R-397LK receiver - "Lapis" (Р- 397ЛК "Ляпис") used by the Navy Special Forces and - likely - not only by the Navy or by Russian forces. 
The translation from Russian is shown below. 

The R-397 LK radio receiver is intended for tuningless and search-free communication with discrete frequency setting in the short-wave range and provides reception and registration of call commands with visual and sound indication in the memory device, as well as auditory reception of messages by Morse code in the RT and AT modes. The radio receiver provides the following modes:
- Auditory reception of telegraph signals with amplitude modulation (AT)
- Auditory reception of telegraph signals with frequency modulation (FM)
- Continuous reception of commands by signals of RF at a rate of 19.53 baud (CALL CONTINUOUS.)
- Reception of commands by signals of RF at a rate of 19.53 baud with periodic switching on and off of the radio receiver power (CALL CYCLE) 

Fig. 1 - 19.5Bd/100 FSK (R-397 OK)

The “IRTYSH” (“ИРТЫШ”) equipment is designed to provide h24, noise-resistant short-wave special radio communication for the purpose of transmitting short messages with up to 50 digital groups and calls and providing auditory telegraph communication.

RADIO LINE COMPOSITION [3]
R-397 OK - special radio transmitter "OKOLYSH"
R-397 LK - special radio receiver "LYAPIS"
R-397 KC - a set of center equipment "KEDON" as part of the R-160P radio receiver, R-397 OTs demodulator and BPA
R-397 OTs - special demodulator "DEER" included in the set of center equipment "KEDON"
Р-397 LC - the generator of calling commands "LUMEN"
BPA - start-stop telegraph apparatus operating with the MTK-2 code

Conventionally, the radio line "IRTYSH" can be divided into two functionally independent directions:
radio link "OKOLYSH" - "KEDON"
radio link "LYUMEN" - "LYAPIS"
Radio direction "OKOLYSH" - "KEDON" works only for the transmission of information with a special broadband low-energy signal from the correspondent to the Center. The radio direction "LYUMEN" - "LYAPIS" provides an urgent call for communication with a source of intelligence or special intelligence at any time and the transfer of information by auditory telegraph from the Center. Thus, it seems I heard just a short transmission in the Lyumen-Lyapis mode, maybe the R-397 OK2 was the used device.

[1] https://yadi.sk/d/z-4ImVGLytR0rQ
[2] http://forums.airbase.ru/2008/10/t62176_10--spetsnaz-vmf.html
[3] https://sdamzavas.net/4-35296.html 
https://studfile.net/preview/7104174/page:25 
http://www.vrazvedka.ru/main/learning/last-confl/afgan-03_07.shtml

unid 1200Bd/850 (G)FSK bursts recorded in Japan

This is an update of the January 9, 2019 post (see below)

Nicely, on the same day two friends of mine linkz (from France) and Eddy (from South Australia) signaled me about the presence of these 1200Bd FSK bursts on 19102.0 KHz (cf). Linkz also DF'ed the signal with good success, identifying the probable Tx site location in the city area of Busan, South Korea "still transmitting the same data over & over":

https://twitter.com/hf_linkz/status/1439165124141592578

So, at present, these (still unid) transmissions have been heard on (KHz): 4584,4626,4756,7531, and 19102 (all cf).

9 January 2019

 

This 1200Bd/850 FSK signal was recorded at different periods using some the KiwiSDRs located in Japan (http://103.2.34.7:8073 http://222.7.151.84:8073 http://kiwisdr-jp7fso.ddns.net:8073), it was observed, at least, in three frequencies: 4765, 4626 and 4584 KHz. During night-time good results are also obtained with the KiwiSDR at Irkutsk (Russia), so the origin of the signal seems to be Japan or surroundings. 

My Spanish friends ANgazu and Rapidbit (from radiofrecuencias group) did a brief analysis measuring the speed (1200Bd) and the shift between tones (825-890 KHz) and suggesting the GFSK mode. On my behalf, I veried their measurements and verified that the bursts are 26 secs spaced and carry the same (encrypted?) text sent in async 8N1 mode (Fig. 21), although there are some difference among old recordings and new ones. The stream obtained after removal the start/stop bits does not offer useful information (encryption? not-standard 8-bit alphabet?), same results after descrambled the stream using the polynomial x^3+x^2+x+1. 

Fig.1

 

https://yadi.sk/d/fUFrXKOTgnEGaw

SDPSK 40Bd 50Hz OFDM-60 (60-out-of-61 + pilot tone)

OFDM signal heard on 5767.0 KHz (cf) and consisting of 61 channels plus a pilot tone located at the lowest frequency (thus 62 channels in all). The modulator actually uses 60 of the 61 available channels (60-out-of-61) since the position #34 is empty (figure 1).

Fig. 1

For what concerns the OFDM formation, the channels have a 50 Hz spacing and are modulated using SDPSK - also called π/2-DPSK (1) - at the rate of 40 symbol/sec (2400Bd as "aggregate" speed).

 

Fig. 2

Fig. 3

The pilot tone at its lowest position, rather than at the usual highest frequency of 3300 Hz, leds to think of a Russian "Serdolik" waveform, indeed it's similar to the Serdolik OFDM-60 [1]: same speed (40Bd) and spacing (50 Hz) but PSK4 modulation. Some friends of mine (Karaputz, linkz,...) confirmed the idea, likely an enhanced waveform.
Reception and recording thanks to the "Tambov" KiwiSDR [2].

 

(1) In SA Phase-Plane using n-Ary = 4 and absolute mode (diff=0) the transitions between states are similar to QPSK but without diagonal paths (no "zero" crossings); in differential mode (diff=1) we see transitions between two states (Fig. 3) thus it's a Differential-PSK or DPSK. DPSK is called Conventional DPSK (or CDPSK) if the phase differences is in the set of [0,π] and Symmetrical DPSK (SDPSK, also called π/2-DPSK) if the phase difference is in the set of [π/2,-π/2].  As you see in Figure 3, the transitions in differential mode (diff=1) are in the set of [π/2,-π/2].
 

https://disk.yandex.com/d/qq37H3KwwaJn9w 

 

[1] https://i56578-swl.blogspot.com/2015/04/serdolik-crowd-36-ofdm-mode-61-tone.html?m=0

[2] http://sdr4.tambov.gq:8901/?f=5767.00iqz7

2000Bd/3250 FSK on 26.9 MHz

Interesting FSK signal spotted on 26.9 MHz by friend killer258 from raduioscanner.ru: modulation speed is 2000 bps and 3250 Hz shift (figure 1).

Fig. 1

ACF results show a kind of "interleaved" values of about 161.7 ms that make a 322-bit length period; indeed, looking carefully at the pattern, the stream is actually formed of a 161-bit sequence wich is sent alternatively in positive and negative polarity (figures 2,3).

Fig. 2

 

Fig. 3 - 322/161 bit period

 

Also interesting are the oscillations during modulation in the first part of the transmission which are visible in figure 4. Tones do not preserve their phase (figure 5).

Fig. 4

Fig. 5

Thanks to killer258 who kindly allowed me to use his recordings.

https://disk.yandex.com/d/br7vJ2uGOTFr9w

75(50)Bd 4481F: yet another channel

Yet another 75Bd/50Bd 4481F channel spotted on 9338.0 KHz (cf), most likely from NPM Lualualei, HI (thanks my friend Mike "mco"). After filtering out the column of the replicated bits from the demodulated stream, it turns out the actual 50Bd speed; then resizing the new stream into a 7-bit pattern, the KW-46 sync sequence emerges.

https://disk.yandex.com/d/eGbxz_R34wLOdw

async 5N1 STANAG-4481F, likely tests or training transmissions

5 September 2021 Update

Transmissions are now continuous and after the removal of the start/stop bits the text appears KG-84/KIV-7 secured. It's worth noting that the  128-bit Initialization Vector is splitted in two 64-bit groups and each group is repeated twice rather than four times (as instead it's used to do). That feature has already been osserved in other STANG-4481F transmissions from UK MoD [1] (spotted on 6245.20, 8056.7, 8127.0, and 10272.0 KHz all Cf).

Fig. 1

TDoA results now indicate definitively Crimond (figure 2).

Fig. 2

https://disk.yandex.com/d/Huw79u7T8CKAGQ

[1] https://i56578-swl.blogspot.com/2021/03/async-stanag-4481f-with-kg-84kiv-7.html

3 September 2021

Async (and episodic?) 5N1 STANAG-4481F "segments" spotted on 4539.7 KHz (cf): as a distinguishing feature, it's to notice the presence of a pilot tone at CF-700Hz (4539.0 Hz) preceding each segment and the slight diversity of the durations (both of the segments and of the pilot tone).

Fig. 1

 

I went casually on the signal at about 2150 (Sept. 1st) when it was already active and it lasted almost all night until about 0800 the following morning (Sept. 2nd) when it ceased, all times are UTC: after that I haven't heard it again, unless a sporadic carrier just on 4539.0 Khz. The effective duration of the transmission can be verified by looking at the 24-hour waterfalls as received by the WebSDR receiver in Twente (figure 2).

 

Fig. 2 - Twente WebSDR wtarefalls of 1 and 2 September

 

The 5-bit text after the removal of the start-stop bits seems to be encrypted, definitely not KG-84/KIV-7 or other encryption that I'm aware, or consisting of pseudo-random chars. I tend to think of training or test transmissions - that makes more sense - on a not "usual" frequency: indeed I did not find mentions of S4481/4539.7 KHz in the large collection of logs of UDXF group, only few S4285 logs from Italian-Ny.

Fig. 3

All the direction finding tests (figure 4) point to an area located in north UK, therefore the Tx site could be likely Crimond (Aberdeenshire, Scotland), belonging to UK MoD DHFCS [1]:  it must be noted - however - that the results could suffer from the non-continuity of the signal. By the way,  similar async STANAG-4481F transmissions have already been observed on last March and just from Crimond [2].

Fig. 4 - some TDoA results

https://disk.yandex.com/d/CmkxVf5NwOBJ6A
[1] https://www.scottish-places.info/features/featurefirst94146.html
[2] https://i56578-swl.blogspot.com/2021/03/async-stanag-4481f-with-kg-84kiv-7.html