Tandem Free Operation is activated and controlled by the Transcoder Units after the completion of the call set-up phase at both ends of an MS-MS, MS-UE, or UE-UE call configuration. The TFO protocol is fully handled and terminated in the Transcoder Units. For this reason, the Transcoder Units cannot be bypassed in Tandem Free Operation. This is the key difference with the feature called Transcoder Free Operation (TrFO) defined in TS 23.153. In return, the Transcoder Units continuously monitor the normal Tandem Free Operation and can terminate TFO as soon as necessary with limited impact on the speech quality. Before TFO is activated, the Transcoder Units exchange conventional 64 kbit/s PCM speech samples coded according to the ITU-T Recommendation G.711 [13] A-Law or μ-Law. The Transcoders can also exchange TFO messages by stealing the least significant bit in every 16th speech sample (see Annex A for the specification of the TFO message transmission rule and clauses 6 to 8 for the description of the TFO procedures and messages content). If compatible Speech Codec Types and Configurations are used at both ends of the MS-MS, MS-UE, or UE-UE call configuration, the Transcoders automatically activate TFO. If incompatible Speech Codec Types and/or Configurations are used at both ends, then a codec mismatch situation exists. TFO cannot be activated until the codec mismatch is resolved. This capability is an optional feature involving other network elements of the Radio Access Network. The rules for finding a common codec type and solve the codec mismatch are defined in clause 11 and clause 12. Once TFO is activated, the Transcoder Units exchange TFO Frames carrying compressed speech and in-band signalling, which structure is derived from the GSM TRAU Frames defined in the TS 48.060 and TS 48.061 (see clause 5). The exchange of TFO messages is still possible while TFO is active. In this case, the stealing process will result in embedding a message in the synchronisation pattern of the TFO Frame. When TFO is activated between two end connections using the GSM_HR speech codec, the TFO Frames are carried over 8 kbit/s channels mapped onto the least significant bit (LSB) of the 64 kbit/s PCM speech samples. When TFO is activated between two end connections using the GSM_FR or GSM_EFR speech codecs, the TFO Frames are carried over 16 kbit/s channels mapped onto the two least significant bits of the 64 kbit/s PCM speech samples. When TFO is activated between two end connections using the AMR speech codec, the TFO Frames are carried over 8 or 16 kbit/s channels mapped onto the least or two least significant bits of the 64 kbit/s PCM speech samples. The format depends on the codec configuration (Optimized Active Codec Set). To facilitate a seamless TFO interruption, the six or seven MSB of the PCM speech samples (not compressed) are transmitted to the far end unchanged. Like GSM TRAU Frames, the TFO Frames have a fixed size (and duration) of:

• 160 bits (20 ms) for the 8 kbit/s format;
• 320 bits (20 ms) for the 16 kbit/s format.

Figure 4.2.1-1 provides a reference model for the functional entities handling Tandem Free Operation. The TFO Protocol is fully described in clauses 9 (State Machine) and 10 (Detailed Protocol).

The same TFO protocol and Frame Format is used irrespective of the PLMN types at both ends of the call configuration. Figure 4.2.1-2 shows a normal TFO configuration involving the same or two different GSM networks.

Figure 4.2.1-3 presents a TFO configuration involving two GSM-evolved 3G Networks. Note that the same protocol and Frame Structure are also used irrespective of the type of Transmission Network connecting the two 3G networks (ATM or STM).

Finally, Figure 4.2.1-4 presents a TFO configuration involving two different network types (GSM and 3G). Similar configurations could be derived with any network supporting a TFO protocol compatible with the present document.

In case of AMR-WB the TRAU/TC performs in uplink direction the wideband decoding and a successive lowpass-filtering, downsampling to 8kHz sampling rate and PCM (G.711) encoding, before its sends the narrowband version of the speech signal towards its destination. This downsampled speech signal in PCM (G.711) representation allows interworking with the narrowband world (PSTN etc.). If a 64kbit/s channel is used, then a transcoded wideband signal (7 kHz speech bandwidth and 16kHz sampling rate) would anyway not fit into it. An efficient way to transport the wideband signal via such a channel is to use TFO (or TrFO) which delivers the compressed (encoded) speech. The encoded speech has a bandwidth significantly lower than 64kbit/s. In TFO_State OPERATION the TRAU/TC sends the AMR-WB TFO Frames within the LSBs of this PCM signal. In the other, downlink direction the TRAU/TC performs G.711 decoding, upsampling to 16 kHz sampling rate, lowpass- filtering and wideband encoding before it sends the AMR-WB parameters down to the A/Iu interface. In TFO_State OPERATION the TRAU sends the AMR-WB parameters as received via the TFO Frames downlink. A listener on the A/Iu interface will always hear the narrowband version of the speech conversation, while both ends send and receive the wideband version. The basic principle for TFO operation for WB speec codec tpyes is the same as for narrow-band speech codec types (see section before). The following items must additionally be considered:

• A new size of 640 bits for the 32 kbit/s TFO Frames format is needed in case the highest AMR-WB modes shall be used (the related TRAU format is defined in TS 48.060).
• The scenario in Figure 4.2.2-1 shows the situation when AMR-WB TFO has not yet been established while the call started with a narrowband codec. This is a likely starting scenario, because it it not desirable to occupy radio ressources unnecessarily with wide-band signals, until TFO is operational.
• Figure 4.2.2-2 describes the situation after AMR-WB TFO establishment
• Because of the higher speech signal bandwidth (up to almost 24 kbit/s for AMR-WB) up to the four LSBs must be stolen by TFO franes.
• In case of TFO interruption, the remaining MSBs of the PCM speech samples (not compressed) might not only be less than for narrowband TFO, the transcoded bits carry a different kind of signal: The downsampled signal has narrowband properties (as depicted in Figure 4.2.2-3). Because of the significant difference of the narrowband speech signal's impression (possibly even distorted by the stealing of four LSBs) to the wideband signal's quality, AMR-WB TFO interruptions should be avoided as best as possible.