Supporting multiple codecs is one of the most challenging aspects of rolling out VoIP services across telephone, cable, cellular, Wi-Fi, and Internet networks. There are currently more than 25 different codecs in use--each of which may be optimum for a specific VoIP application--and there are many new codecs on the horizon. This multi-codec world makes it necessary for VoIP providers to translate (or transcode) traffic at their borders. The problem is that transcoding can be very expensive and complex. In this article, we’ll look at the transcoding issue and how VoIP providers can best implement future-proof transcoding solutions.

One codec is good, so more are better

Initially, most VoIP operators standardized on a codec called G.711. G.711 became the de facto standard because it represented the best tradeoff between voice quality, bandwidth, and required processing power. G.711 offers a usable audible frequency spectrum from 300-3400Hz and has low processing overhead (0.2 MIPS per channel). On the other hand, G.711 requires a relatively high amount of bandwidth (64 Kbps per channel or up to 110 Kbps with the inclusion of IP headers). In many DSL deployments, total upstream bandwidth may be limited to 128 Kbps, so by using G.711, providers are limited to offering one VoIP line per customer.

Over time, engineers have developed many alternative codecs, each of which has specific advantages. Today, there are specific codecs for wireline VoIP, wireless, music and video. Codec proliferation occurred because engineers have developed additional encoding methodologies to more faithfully reproduce voice (higher quality, as defined by mean opinion score or MOS), allowing them to pack more calls into the same amount of bandwidth. These various codec types are optimized for specific target applications.

Voice codecs have traditionally been implemented within media gateways and customer premise equipment to convert analog and TDM voice to VoIP. Each voice channel (or call) requires network bandwidth, while the encoding/decoding process requires processing horsepower. Ideally, a codec would encode speech to perfectly represent the analog waveform of the voice on the call while using virtually no bandwidth and very little processing power. In reality, the decision to use one codec or another involves a tradeoff between bandwidth and processing power.

For example, the G.729a codec saves 88 percent of the bandwidth from a G.711 call (enabling the carrier to offer more lines per customer), but it requires 38 times the processing power and results in a slight reduction in voice quality. New generations of wideband codecs are becoming adopted (iSAC, G.722, Speex WB, and others) and offer the promise of slightly reduced bandwidth with a much higher call quality utilizing the full audible frequency range of speech (50-8000 Hz). Depending on the VoIP provider’s marketing goals (more channel density vs. better quality for example), the provider may well select different codecs for different applications.

As VoIP networks, usage, and applications expand, the need to translate traffic from one codec to another becomes more and more important to ensure global connectivity.

Transcoding implementations

Codec transcoding provides a means to convert traffic so that two VoIP networks using different codecs can exchange traffic without making codec format changes on their individual network devices. Transcoding efficiently and effectively interconnects traffic at the network peering point, thereby allowing each provider’s network to use its codec of choice. The transcoding location also becomes a central point where all codec treatment can be handled, allowing further flexibility with the addition of different codec types in the future.

A typical VoIP network contains several elements that could potentially support transcoding: media gateways, media servers, session border controllers, and media processing platforms (see Figure 1).

Figure 1: VoIP network architecture

Media Gateways: Media gateways provide codec encoding/decoding processing, and would seem a likely place to provide transcoding capabilities. However, media gateways must have enough MIPS to support the higher processing requirements of any new codec or codecs. Since many carriers built networks to handle the G.711 codec only, upgrading to support additional codecs typically means reducing the number of channels per media gateway. This leads to additional expenses for new media gateways just to handle the previous call volume, let alone provide for future growth. In addition, supporting each additional codec requires that the provider upgrade 100 percent of deployed ports, since each trunk group will require diverse codec coverage.

Media Servers: Media servers typically reside in the backbone of the network and provide announcements, conferencing, and other special services to callers. While media servers have the advantage of being centrally located in a network (and are therefore less costly to upgrade), it is inefficient to have to backhaul each call that requires transcoding. In addition, most media servers are somewhat limited in the number of codecs they support, and are limited in the processing power available to support MIPS-intensive codecs.

Session Border Controllers: Session Border Controllers (SBCs) reside at carrier borders, most notably at the access or peering border. As such they are ideally located for transcoding. However, an SBCs’ main function is to provide security and connectivity capabilities, and they are not optimized for scalable codec transcoding. These systems are limited in their available MIPS resources for transcoding functions. It is possible to upgrade SBCs so that they can offer transcoding, but SBC architectures would then limit the density of typical SBC services that could be carried on the same system.

Media Processing Platforms: A media processing platform is a relatively new type of system designed specifically to handle transcoding and voice quality functions. It offers very high density and processing performance and can be deployed as a general resource for all existing media gateways, media servers, and SBCs in the network. A media processing platform is most commonly deployed at the access or peering border which provides a natural aggregation point for transcoding, similar in placement to an SBC. Because the media processing platform supports the widest array of codec types with high density, it is far less disruptive to deploy and less costly to install and support than any of the other options for transcoding (see Figure 2). The oversubscription benefits (where only a fraction of total calls may require transcoding) also save operators money from upgrading other VoIP equipment.

Figure 2: Codec Transcoding with a Media Processor

Recommendations for cost-effective growth

The optimal choice in a transcoding solution depends, of course, on the needs of the individual provider. However, providers typically need a transcoding solution that satisfies four key requirements: codec flexibility, high processing power, high call density, and high availability.

Codec flexibility: Carriers need to support a broad range of codecs. With VoIP traffic from different wireline or wireless networks, carriers will need to offer support for over a dozen codec types to ensure proper call completion.

High processing power: The most important requirement for a codec transcoding solution is to have large amounts of digital signal processing (DSP) power. Because the carrier may not know which codecs will be processed at any given time, it must deploy codec transcoding capabilities that can easily scale and adapt to changing codec needs whether the codec types require a little or a lot of processing.

High density: Given the rapid growth in VoIP subscribers, carriers should be prepared to accommodate new users and processing demands along with diverse codec support requirements. Ideally, a transcoding system should accommodate at least 40,000 calls per telco rack.

High availability: A transcoding system becomes a critical piece in the total end-to-end voice call, so it must offer at least telco-grade “five nines” reliability.

Viewing these requirements against the strengths and weaknesses of the transcoding solution types presented earlier, we can come to some conclusions.

• Most media gateways don’t have the native processing power to support transcoding for a wide range of codecs, and their density suffers if they are pressed into transcoding duties. Upgrading all media gateways to support new codecs is a very costly proposition.
• Media servers were designed to provide call features, not to transcode between different codecs. Without costly upgrades, they cannot deliver the processing power needed for heavy transcoding volume. As a centralized resource, a media server’s network location requires all traffic be backhauled for transcoding. This can cause unnecessary delay in the end-to-end VoIP call.
• Session border controllers seem to be a good solution as they are located at the ideal transcoding locations, but these, too, lack the processing power for transcoding the diversity of supported codecs. In addition, most SBCs become quite cost prohibitive when transcoding blades are installed instead of SBC blades.
• Media processing platforms are the only solution expressly designed for transcoding functions. They offer high codec flexibility, drop-in support for future codecs, ample processing power for high density, and high availability with built-in fail-over systems. They can be deployed at the ideal transcoding points in the network (access and peering borders), and benefit both the access network customers as well as carriers by interoperating with backbone equipment (e.g. media gateways, announcement services, and media gateways).

As we can see, media processing platforms offer the best solution for transcoding in terms of performance and availability. Users of this solution also benefit from not having to alter their existing network devices, because the media processing platform can be deployed in existing networks and perform transcoding as traffic is directed to it.

Providers must begin planning to support transcoding to easily support new codecs as they arrive on the scene in the coming years. As the only purpose-built transcoding solution, media processing platforms offer the best combination of performance, density, flexibility, availability, and ease of deployment.

Matt McGinnis is director of product marketing for Ditech Networks.