July 1998

BY BROUGH TURNER

The ultimate impact of IP telephony is widely acknowledged. Only the timetable for totally restructuring the PSTN remains in debate. It is clear, however, that the first stage - toll bypass - is well underway. Toll bypass applications use gateways to route voice and fax calls from one telephone switch to another using packet networks instead of the traditional switched telephone network.

Toll bypass is more than a passing fad. New carriers like Qwest and Level 3 have emerged precisely because it is now possible to build an IP-based telecommunications company for one tenth or one twentieth of the investment it took to build AT&T, MCI, or Sprint. So, while the long-term promise of IP telephony is for wonderful new services, the path to change starts with bypass - using many, many gateways. And, with roughly one billion traditional telephones in the world, gateways will be needed for a long time.

WHAT IS AN IP TELEPHONY GATEWAY?
An IP telephony gateway includes the following components:

• Traditional telephony interfaces (to a phone, a PBX, or the PSTN).
• IP network interface(s).
• Call processing (call setup, call teardown, etc.).
• Packet protocol processing.

Real-time media stream signal processing (voice compression and decompression, fax demodulation and re-modulation, and so on), in most cases.

Today's gateways come in innumerable flavors. Because of inherent time-to-market advantages, almost all of the early gateways were built on open telecommunications platforms based on industrial PC technology. Fixed-function gateways are now emerging as traditional equipment vendors (for example, Cisco, Ascend, 3Com, Nortel, and Ericsson) announce gateway functionality for their existing equipment.

So which approach is better? An open telecommunications platform provides the greatest flexibility and quickest time-to-market for new applications, in large part because it has the richest software development environment and the largest pool of programmers. But the embedded solution can optimize cost or performance for a specific environment. So, it is likely both approaches will exist. In stable markets, embedded approaches have the advantage. But in evolving markets, flexibility is key. By the time an embedded solution gets to market, technology has advanced. The open telecommunications systems' time-to-market advantage continues as long as technology and/or applications are evolving.

WHO IS BETTER EQUIPPED TO DELIVER?
Beyond the question of embedded versus open, who is better equipped to provide these gateways - router vendors, traditional telecom equipment vendors, or CT developers? Let's examine this question component by component.

Telephony Network Interfaces
Telephone networks are arcane, country specific, and regulated to the hilt. There are many kinds of telephony trunks, both analog and digital, with dozens of different signaling protocols. Until now, router vendors have only connected to dedicated telephone facilities used for data links. So gateway developers with telecom experience have an advantage here. And, within the telecom sphere, CT developers should continue their lead over more slow-moving traditional telecom vendors.

IP Network Interfaces
On the IP side, router vendors have more experience interfacing to IP networks. But Internet protocols are well-defined and vastly simpler than telephony protocols - one reason the Internet is experiencing its phenomenal growth. So the router vendors' advantage is not as significant. In fact, since CT developers have experience applying computer technology to voice communications, they tend to have a substantial advantage here.

Flexible Call Processing
When implementing call processing functions for an IP telephony gateway, CT vendors have experience the router vendors lack. And, while traditional telecom equipment vendors also have the needed experience, it's difficult to adapt their proprietary systems to new applications. Both traditional telecom equipment vendors and router vendors are addressing this void by forming partnerships with a variety of computer telephony vendors. Bay Networks, for example, made an investment in NetSpeak, a developer of intelligent software modules for IP telephony. Deutsche Telecom purchased an equity stake in VocalTec, a leading supplier of IP telephony gateways. Other carriers are buying complete solutions from Clarent, a leading provider of networked IP telephony solutions. And Cisco is endeavoring to leverage work from Bellcore proposing a simple gateway control protocol (SGCP) that would provide a way to control relatively dumb embedded gateway components with a more intelligent call agent running on a general-purpose computer.

Packet Protocol Processing
There are two issues here: Call setup (on the IP side) and media stream protocol processing. To negotiate IP telephony connections, the VoIP Forum has endorsed the ITU's H.323 standard, a rather heavyweight protocol unlike other Internet protocols. H.323 is a telephony standard developed by extending early circuit-switched (H.320) video conferencing to work over LANs. It provides for the setup and transport of real-time voice, data, and video across a "non-guaranteed quality-of-service" network (a LAN or the Internet).

Because H.323 is such a complex standard, it is at risk of being replaced by a suite of IETF protocols: Session Description Protocol (SDP) (RFC 2327), Session Announcement Protocol (SAP), Session Initiation Protocol (SIP), and the Real Time Streaming Protocol (RTSP) (RFC 2326). While these IETF protocols are still in flux, they have the advantage of being simpler and layered (so they are more flexible). For now, it is likely a gateway will have to support both H.323 and SDP/SAP/SIP/RTSP. Simple devices may only support SIP, relying on proxy software running on a nearby gateway or server to negotiate H.323 sessions when necessary. The good news is that the media streaming protocols for both H.323 and SIP are the same - Real-Time Protocol (RTP) and Real-Time Control Protocol (RTCP).

RTP processing may be done on the host PC, in firmware on plug-in cards, or embedded in a stand-alone device. In a proprietary solution, this function is typically implemented on an embedded RISC processor. In an open solution of 30 ports capacity or less, RTP processing can be performed on the host PC. For higher port counts, RTP processing can bog down the host and introduce speech quality problems. Scalable solutions perform RTP processing in firmware.

Choice Of Vocoders
A few systems use raw telephony data (G.711 format) wrapped in RTP packets, but most systems take advantage of modern digital signal processors (DSPs) to compress the voice, thus saving bandwidth, and to demodulate fax, again saving bandwidth in the IP network. The standard vocoder for voice over IP (as established by the VoIP Forum) is G.723.1, with fallback to G.711. However, the voice over frame relay standard calls for G.729 voice coding, and many of the data communications vendors support only G.729, even over IP. Meanwhile, other popular vocoders include a Microsoft variant of GSM, Voxware's MetaVoice RT 24 and the Lucent elemedia SX7300P.

This diversity of vocoders is only going to increase. As telephony migrates to IP, new devices will have to interface with legacy equipment. Legacy equipment includes all the vocoders used in the air interfaces of today's wireless phones (at least six to eight different standards here!). Then as new high-quality telephones emerge, wideband coders will become an issue. Three years from now, a competitive gateway may have to support 12-15 different vocoders, dynamically assigned on a call-by-call basis. Flexibility is a major issue and is a big advantage for solutions based on open platforms.

MANAGING THE GATEWAYS
To build a global IP telephony system that allows people to place calls and reach destinations, a lot more is needed than just a couple of gateways at either end. There needs to be a way to manage and monitor the gateways, provide user authentication, determine if the called party is on-line, locate the best path between gateways, and provide billing information - to name just a few. And this management solution must scale. What's emerging are a variety of end-to-end system solutions for phone-to-phone or PC-to-phone connections. CT vendors shipping products in this arena include NetSpeak, VocalTec, and Clarent - which is not coincidentally why carriers and router vendors have formed the partnerships mentioned above.

To date, each vendor of distributed system-level control software has also built their own gateway. The next stage in the development of IP telephony gateways will be standard interfaces for all gateway functions. This will allow distributed network control software to interoperate with gateways from different vendors. H.323, SDP/SAP/SIP, and the recent Bellcore SGCP proposal are all steps in the right direction. A stable standards environment could be in place in less than a year - definitely within three years.

THE EFFECTS OF APPLICATION ON THE GATEWAY DESIGN
The future form of an IP telephony gateway may also depend on where it is located - in a public network, at the center of an enterprise-wide network, or at a branch office.

The Migration Of Carrier-Grade Solutions
In a public carrier or enterprise-level situation, gateway availability is a key issue. Proprietary systems have excelled here, but open platforms using multiple PCs are now able to achieve any desired level of availability. CompactPCI chassis with hot swapable cards can only improve this situation. And indeed, based on RFIs and RFQs of the past six months, it appears many traditional communications equipment vendors are adopting CompactPCI.

The Evolution Of Low-End Solutions
Branch office gateways are another story. Purchasing a separate PC chassis to support a four-port VoIP gateway makes no sense. For now, low-cost, fixed-function, proprietary VoIP gateways will be plug-ins for existing equipment (routers or telephone systems) in this market segment. After all, routers and phone systems exist and have established distribution channels. However, as communications servers evolve, we should see PC-based branch office solutions that provide the IP telephony gateway together with PBX, auto attendant, IVR, and fax server functionality, plus potentially the firewall and WAN access router. Such a communications server is clearly in Microsoft's sights for NT 5.0 and TAPI 3.x. CTI vendors would do well to align with Microsoft on this one.

Further in the future, we will see a new class of open, network-connected, programmable platforms. Since 1950, we've had mainframes, mini-computers, and personal computers. We are due for a new programmable platform (at one tenth the price of a PC and a hundred times the sales volume!). Think set top box, telephone, digital camera...Sun and Microsoft see the potential (thus Java and Windows/CE). As these new devices emerge, they will redefine CTI. They will also provide another approach to the low-end IP telephony gateway.

OPENNESS REMAINS THE BEST BET
IP telephony gateways started on open platforms. For the next few years, the gateway market will be split between open and embedded. And, the rapid restructuring of the PSTN will drive sales of all kinds of IP telephony gateways. But in this rapidly evolving market, flexibility and time-to-market are key. So, those gateways that do go proprietary will migrate back to open telecommunications over time.

Brough Turner is senior vice president of technology at Natural MicroSystems, a leading provider of hardware and software technologies for developers of high-value telecommunications solutions. For more information, call Natural MicroSystems at 508-620-9300 or visit the company's Web site at www.nmss.com. E-mail to the author ([email protected]) is also welcome.