These voice and video compression standards produce bit streams that are structurally and semantically different. The task of ensuring interoperability among the diverse terminals with different media standards falls to network infrastructure equipment such as media gateways, content servers and videoconferencing equipment. Clearly, this much more complex internetwork of competing protocols needs some mechanism for media transcoding among the various voice and video standards. This mechanism is referred to as "media transcoding."

Basically, media transcoding is the conversion of bit streams from one voice (or video) standard to another, to facilitate the interoperability of devices using different media standards. This conversion may also be referred to as "transcompression."

Media transcoding occurs at various places in the network infrastructure today. Depending on the routing of a voice call or video stream, it can happen once or more often in the network.Transcoding usually takes place in mobile switching centers, media and multimedia gateways, content servers, videoconferencing equipment and multimedia messaging servers, among others.

Transcoding is a computationally intensive process and therefore is not left to the end terminals, which in the case of mobile devices are limited by processing power and memory. Transcoding is required for pure voice as well as for voice-and-video applications, and it's needed for both conversational and streaming applications.

Among the variety of contexts in which voice and video transcoding is necessary are wireless-to-PSTN and wireless-to-packet connectivity; wireless mobile to mobile/fixed voice and videoconferencing; between mobile communications on different networks-for example, GSM and CDMA; video-on-demand systems; voice-over-IP communications; international routing and trunking; video mail; content streaming; MMS/IMS messaging services; and Webcams used in surveillance and monitoring.

Some researchers believe that voice and video specifications will converge into a single standard across wireless and wireline networks, eliminating the need for transcoding and giving rise to wide-scale interoperability. Although a noble goal, this is highly optimistic and even unrealistic. Next-generation networks evolve slowly and take years to deploy completely. Therefore, at any given point in time, a variety of old, current and emerging coding standards will have to coexist, leading to the need for media transcoding.

Wireless networks today, as a subset of all networks, use a range of different voice-coding standards, including GSM-EFR, -FR and -HR; narrowband and wideband AMR; and EVRC, QCELP and SMV. Further improvements and breakthroughs are continually emerging and becoming standardized. Convergence across wireless standards themselves is an unlikely scenario, and therefore, the possibility of convergence among wireless and wireline standards is very remote.

Tandem transcoding is deployed in different ways, depending on whether it is enlisted for voice or video.

In the case of voice, when converting from one voice standard (say, standard A) to another (standard B), the coded voice in standard A is completely decoded to pulse-code modulation (G.711) and then re-encoded to standard B. In the video context, when converting from video standard A to video standard B, the coded bit stream in A is completely decoded to YUV and then re-encoded to standard B.

Although tandem transcoding is the prevalent method used in networks today, it suffers from two major drawbacks. First, there is the issue of the large digital signal-processing resource requirement. While the decoding utilizes a small amount of Mips, the re-encoding process in tandem transcoding requires intensive computation and therefore consumes a large amount of DSP power and memory. The full decode-to PCM in cases of voice or YUV in video-and subsequent re-encode is highly inefficient and reduces the number of channels that can be transcoded using a computational resource such as a DSP.

Another problem is long time delays, stemming from two sources. One is a computational delay caused in the decoding and re-encoding process, and the other is an algorithmic delay due to the requirement of codecs for "lookahead samples." This dictates that the tandem transcoder needs to wait for the next frame before processing and emitting the transcoded bits. This wait can vary from 10 to 30 milliseconds, depending on the compression standard.

As in the case of tandem voice transcoding, tandem video transcoding also incurs both types of delays, and in particular algorithmic delays. This is because the decoder part of the tandem transcoder has to decode the full frame before starting the re-encoding process and the production of transcoded bits. Delay is typically additive and occurs every time the transcoding does. This delay, coming on top of the network delay, can have a significant adverse impact.

See related chart