…You say: “So it is like a MS-MS call”. Then I have a question: “Is ter*minating party located on PLMN or PSTN?” Originating party is located in PLMN as Ivann said.
When one MS sends DTMF-tones to remote side in conversational state it sends Start DTMF on FACCH and then (after period of time approximately 40-70 ms) it sends Stop DTMF on FACCH. These nessages is transfered on the signalling channels (not traffic) from MS to MSC. Then MSC generates DTMF tones (according to number of frequency in Start DTMF commands) to remote side. These tones is transfered in voice band on traffic channel (no matter PLMN or PSTN are remote party).
If remote party is in PSTN then collissions (if any) may occur only in MS to PSTN direction. There is no trau in PSTN, hence the TFO signalling in backward direction is not present.
The TFO_Req may be sent up to 35 times if no TFO_Ack from remote side. One TFO_Req – 50-70 bits and is sent during about 100-140 ms (see 08.62).
But as you may see from my questions I have no clear visualization of picture (Is it really DTMF-modems?, modems location? etc.).