The Headache Between the Worlds of Broadband and Mobile HD

While France Telecom (News - Alert) has over half a million broadband HD voice users and has launched mobile HD voice service in Moldova, the two worlds won't be able to easily communicate with each other.  And it's all about the codecs; more specifically, the design tradeoffs between broadband and mobile.

 

France Telecom freely admits that there won't be HD voice calls traveling between its mobile and broadband worlds, so both sides will end up defaulting to PSTN-quality until FT works out a transcoding solution it (and most likely other European carriers implementing HD voice on broadband and mobile) can live with. How long that will take is anyone's guess, since FT will have to buy, test and implement a transcoding solution across its territories as it continues to implement HD. 

 

In the broadband world, G.722 is the de facto HD voice standard. Encoding and decoding audio at 16 kHz for a 64 kbps "stream," G.722 can handle both voice and music without breaking a sweat and one can clearly hear background noises. 

 

Adaptive Multi Rate – WideBand (News - Alert), or “AMR-WB,” is the current choice for mobile HD implementations and is optimized for speech with a "sweet spot" of around 13 kbps to 24 kbps for clean speech. It samples at 16 kHz, but doesn't capture as much information when it comes to background noise and music. Since mobile carriers currently treat bandwidth as a scarce resource, they want to squeeze every last bit out of a voice session so they can put more calls per cell tower and minimize data network/backhaul requirements as much as they can.

 

The compromise in voice quality comes when translating between G.722 and AMR-WB. A G.722 voice session at 64 kbps will get "stepped on" to make it fit into the 13-24 kbps of an AMR-WB voice connection. Conversely, an AMR-WB voice session won't have the information and "richness" of background sounds that a native G.722 has.

 

In addition, there's also the overhead in the process of transcoding between two relatively information rich codecs. Since the process of transcoding takes some time and since transcoding is "in the middle" of a real-time conversation, moving between codecs adds some number of milliseconds to the phone call. Add too many milliseconds and you get a clearly definable lag between speakers, which can be just annoying even if it is a good quality voice call. 

 

Since HD calls have more information to be processed in the voice streams, transcoding an HD voice call between HD codecs requires more overall throughput in the transcoding "black box" in the middle as well as more compute power for the actual transcoding process. You need to have efficient algorithms to transcode between formats and at least four or more times more compute power than you would with dealing with a "normal HD" stream if you want to avoid adding in excessive latency.

 

For France Telecom, effectively transcoding HD voice between broadband and mobile worlds would appear to translate into software and processor upgrades – making manufacturers such as AudioCodes and Dialogic (News - Alert) happy.