November 1999

BY JIM MACHI

Relax. This isn’t another article about the compression rates, packet overhead, and mean opinion score (MOS) of the various IP telephony coders. While this is an important and undeniably fascinating piece of the overall coder puzzle, it’s not the whole picture. Besides, we’ve all sat through (or given) presentations endlessly analyzing the specs of every coder. If you attended the recent Internet Telephony EXPO in San Diego, you understand: For a coder with the best voice quality, use G.711 since it has the lowest compression rate; to save bandwidth but get decent voice quality, choose G.723.1 or G.729a.

Now it’s time to take a broader perspective on coders. We’ve reached the “application” phase of Internet telephony. Pure public switched telephony network (PSTN)-to-IP telephony gateways are no longer the only application in town.

What else should we consider besides the compression rate? In the public network, we see PSTN-to-IP gateways incorporating enhanced services like voice mail systems. Also, the “gateway” may not be a PSTN-to-IP gateway. Instead, it may be a gateway to transcode from one type of coder to another. There are consequences as the enhanced services traditionally available in the circuit-switched world come to the packet telephony world. In the enterprise, using IP telephony may involve more than hooking an adjunct gateway to the incumbent PBX so a company can take advantage of voice and fax rate arbitrage opportunities between its far-flung branch offices. We are seeing either pure IP PBXs or IP-enabled PBXs. We are also seeing the rise of CT servers handling many different kinds of applications. In a pure IP telephony world, this can include voice mail, conferencing, auto attendant, and speech-activated dialing.

With IP telephony more than just a transport mechanism for rate arbitrage, does the classic coder analysis still mean anything? Yes. But there are two more key aspects to consider when choosing a coder: transcoding costs and storage costs. In a world where enhanced services are important, the economics of these two functions are crucial. If you are an end user, you want the best solution for the lowest price. Your choice of coder will affect this. If you are a developer, you need to know the tradeoffs between choosing one coder over another. Also, you’ll probably want to give your customer some choices.

How do the coders compare? To find out, we conducted a business analysis. It wasn’t a full business analysis considering costs like gatekeepers, PC server, network management software, and application. Instead, we stripped away these costs to compare the coders head-to-head at the lowest possible level.

ECONOMICS OF TRANSCODING
Transcoding costs are related to bandwidth usage, since the gateway needs to transcode the circuit-switched stream to data packets before it goes to the IP network. Also, the number of channels the bandwidth supports directly relates to the DSP transcoding costs. In the US, if we assume the bandwidth cost per year is $3 per kbps per year and our enhanced service system is servicing 2,400 users, we can create Table 1 to analyze various coders.

In creating Table 1, we also factored in enough headroom for the coder to properly function in a given bandwidth. With all the data packet traffic, we need to increase the kbs needs to avoid collision scenarios. Also, as the bit rate gets lower, we need more headroom because the voice payload becomes a smaller proportion of the actual data carried. We also assumed there were 25 users per channel.

From Table 1, it is clear that the G.723.1 coder, probably today’s most widely used coder, is not necessarily the best coder to maximize bandwidth costs. In this case, G.729a performs better. Keep in mind that internationally, it may not be possible to get $3 per kbps per year. (It might be reasonable to double that figure.) The simple analysis still shows G.729a as the leading candidate.

To figure the media cost per channel, consider average costs of DSP resource cards and how many channels per coder can be used for a certain coder. Since there is downward price pressure here, Table 2 assumes a constant cost of $100. According to Table 2, G.729a does not come out as the clear winner. If the media cost per channel is constant, the cost of the bandwidth is the differentiator.

STORAGE COSTS
In an enhanced service system (for example, a voice mail system), the storage costs of the various coders are also very important. If we consider the previous example, an enhanced service public network voice mail system servicing 2,400 users, and assume that each user requires one hour of storage space, we need to analyze the total system storage cost. We assume the basic cost of storage to be $70 per GB based on a composite scan of PC server company Web sites that offer RAID storage systems.

In this analysis, the G.711 coder is the leading candidate. Since the G.711 coder has the worst compression rate, it will require more hard disk storage space. Does our conclusion make sense? Yes, because hard disk storage is less expensive than the media costs of the systems transcoding the circuit switch to data packets. However, if we assume a pure IP world and there is no need for transcoding between PSTN and IP, the storage cost of the coder is the only cost. Thus, G.723.1 would be the most economical way to store data packets. In creating the scenario in Table 3, we assumed that most systems, even in a pure IP world, need to convert from one type of coder to another somewhere along the line. Someday we will no longer have circuit-switched phones, just our own IP address. However, it hasn’t happened yet.

FINDING YOUR IDEAL CODER
As IP telephony moves into exciting public network and enterprise application areas, we need to rethink the economics of some of our basic assumptions, especially considering the types of applications and the type of network on which these applications will be deployed. Although there is no simple answer to the question of which coder is best, by understanding your network and your application, chances are you will find the coder that fits your needs.

This much is certain. If you are using a managed network for which you pay yearly, it will be useful to perform the type of business analysis we did here to decide which coder best meets your needs. If you are using the public Internet, a compression coder such as G.723.1 or G.729a is still your best bet. c

Jim Machi is director of product marketing, Internet Telephony, for Dialogic Corporation (an Intel company). Dialogic is a leading manufacturer of high-performance, standards-based computer telephony components. Dialogic products are used in fax, data, voice recognition, speech synthesis, and call center management CT applications. The company is headquartered in Parsippany, New Jersey, with regional headquarters in Tokyo and Brussels, and sales offices worldwide. For more information, visit the Dialogic Web site.