CIS-128 and its ACF discrepancy

Looking at a CIS-128 signal heard this morning, I came across the discrepancy about its ACF value and the the right way to calculate it.
The signal exhibits the well known values of this waveform (64+64 channels, 20.9Bd speed and 23.5Hz step) but the modulation seems a BPSK with the add of 4 service tones: unfortunatelly, it's quite difficult to separate a single tone in order to confirm this mode. As it is common for CIS-128, a special char is sent each 5 symbols (Pic. 1) so that the expected value for the ACF should be ~239ms ...but looking for that value we get strong ~477ms spikes in SA module (Pic. 2): it's just the double, or else 10 symbol periods!
In my opinion, the reason for this double ACF is in the way the special chars are sent. 
By looking at the whole 477ms length frame in Pic. 3, it's visible - straining a bit the eyes - the right ACF period that occurs each 5 symbols (~239ms) just as it was expected: but the spec chars sent each 5 symbols are in some way alternately reversed as illustrated by the white-circled bits (Pic. 3). This way, these special chars seem to appear to the ACF-tool eyes as a sort of sequence like 

:: :: :: :: XY :: :: :: :: YX :: :: :: :: XY :: :: :: :: YX :: :: :: ::

and then the repetition period, according to its point of view, is logically 10 rather 5 symbols. Something like an optical illusion. 

Pic. 1

Pic. 2

Pic. 3

The same situation can be found analyzing the 6000Hz 'broadband version' of this waveform (Pic. 4).

In this case the ACF is ~238 ms BUT here the speed is double, so the special chars, which are sent each 5 symbols, should make ~119ms spikes: using the CCF measurement these chars are visible inside the red-marked circles (Pic. 5) and white-marked in the structure of the frame (Pic. 6).

Pic. 4

Pic. 5

Pic. 6

CIS-128 broadband (6 KHz) waveform

Broadband waveform (6 KHz) of the CIS-128 OFDM modem. The values of the manipulation speed and carriers separation are the double of the ones seen in the 3 KHz waveform, i.e. 42 Baud  and 47 Hz, whilst the OFDM formation values remain unchanged (Pic. 1). The waveform allows the two 'data' and 'sync' modes, with the same 1-of-five character feature clearly visible in the data mode. Modulation is always QAM-16 (Pic. 2) and the ACF value is 10 symbols lenght (Pic. 3), like the narrow band waveform.

Pic. 1 - narrow band (3 KHz) and broadband (6 KHz) OFDM parameters

Pic. 2 - QAM-16 modulation in the channels

Pic. 3 -  10 symbols ACF

Below the screenshot of a CIS-128 BB reception (added as update on October, 10) on 14711.5 KHz on USB

CIS-128, OFDM 128-tone 21Bd

CIS-128 is an OFDM 128-tone +  1 'off' tone in the middle of the grid (then 64+64 carriers), it spreads a 3000Hz bandwidth with a step between carriers of 23.5 Hz, manipulation speed of ~ 21 Baud and modulation QAM-16 (Pic. 1). The signal was heard at 0948z on 13399.0 KHz on USB. This waveform  is discussed in radioscanner.ru here: http://signals.radioscanner.ru/base/signal250/
Analyzing the signal, is clearly visible that each 5 symbols a special symbol is sent (Pic. 2) and then the expected ACF value should be 5 symbols; but there is a discrepancy because ACF is effectively doubled, i.e. 10 symbols for a ~477 ms length (Pic. 2).

Pic 1

Pic. 2 - a "special symbol" is sent every 5 symbols

Pic. 3 - 10 Symbols ACF

It seems that during the special symbol, i.e. every 5 clocks, AM modulation is lost since the "shape" of these symbols are pretty the same.

In sync-mode are sent only the special symbols and each 5 symbols all the tones do not change their phase (Pic. 4). The ACF is always 10 symbols lenght but in this case the symbols markers are more evident (Pic. 5). 

 

Pic. 4

Pic. 5 - ACF of sync mode

Dduring a same transmission is possible to see both the data and sync mode, alternatively.