October 1998

Quality Of Service In IP Telephony

BY JOEL HUGHES

Every network application has a basic set of requirements that the network must deliver in order to meet the end user's expectations for acceptable performance. These demands are generally referred to as quality of service (QoS) requirements, the most crucial being bandwidth and delay characteristics. IP telephony is simply a real-time voice application with very stringent requirements in the area of packet delay. It is generally agreed that one of the major obstacles to widespread adoption of IP telephony is acceptable quality of service.

HOW WE GOT HERE
The circuit-switched PSTN was designed to do one thing - transmit full-duplex voice reliably. It does this very well, largely due to the fact that only a few equipment and service providers have been responsible for its construction. The PSTN has established the user's expectation for quality and reliability of voice services. The phrase "toll quality" is often used to describe the PSTN-equivalent quality of service for IP telephony offerings.

Packet-switched IP networks were designed to do something completely different - route data packets on a "best effort" basis. They do this very well. The Internet, the most ubiquitous and best-known IP network, has established the user's expectation for reliability of data services such as e-mail, ftp, and Web browsing as well as the expectation of relatively low voice quality. IP networks were not designed to support real-time traffic, such as voice.

Growth of Data Traffic
In the past several years, principally driven by the phenomenal expansion of the Internet, the growth of worldwide data traffic has far outpaced that of voice traffic. During this period we have come to realize that the PSTN will be subsumed by IP networks in the not too distant future. Instead of using the PSTN to carry voice and data traffic (through the use of modem technology), we will be transmitting packetized voice traffic as just another data type over IP networks. This puts us in a tough spot: How do we deliver the quality of service for voice traffic that the user has come to know on the PSTN, over an IP network that was never designed to support real-time voice traffic?

QoS Factors in IP Telephony
There exist a number of factors, which contribute to the overall quality of service, which can be delivered by an IP telephony service provider. Significant factors impacting the service provider include provisioning, billing, and network management. This article focuses on the primary factors directly affecting the perceived quality of service experienced by the end user or subscriber - voice quality.

VOICE QUALITY
Many factors impact the quality of real-time voice delivered to the end user. Although some are not specific to IP telephony, such as echo cancellation performance, the most crucial determinants of voice quality are unique to transporting real-time voice traffic over IP. These include packet latency (delay), packet jitter, and packet loss. These issues simply don't exist in the PSTN due to its use of a dedicated, full-duplex 64 Kbps circuit between two users to guarantee low-latency delivery of uncorrupted voice data. But in an IP network, each router hop encountered along the voice path adds delay and possibly jitter and loss to the data stream. Let's examine each of these factors.

Packet Latency
Packet latency is the time it takes for a voice packet to be transmitted from one user to the other. An IP network can introduce packet delay which, if over about 150 ms in one direction, in addition to the approximately 100 ms latency inherent in the two IP telephony endpoints, is likely to degrade the transmission from a natural interactive conversation to something more like a push-to-talk radio session. A general rule of thumb is that each router hop introduces approximately 10 ms of latency. Also, in an IP network, the path taken by packets sent from endpoint A to endpoint B is likely to be different than the path traveled by packets sent from endpoint B to endpoint A. This means that the latency may be different in the receive and transmit directions as seen by each endpoint.

Packet Jitter
Packet jitter refers to the variability in arrival time between consecutive voice packets. If jitter is not accommodated in some fashion, gaps and dropouts in the decoded speech result due to a mismatch in the asynchronous packet delivery and the synchronous voice playback system. This produces an unacceptable user experience. To combat jitter, IP telephony endpoints employ buffering schemes, which smooth the asynchronous-to-synchronous interface between the packet and real-time data flows. Unfortunately, this jitter buffering adds to the end-to-end latency of the system. The more jitter there is that needs to be handled, the more latency is introduced.

Packet Loss
Packet loss is the actual dropping of data in an IP network between the endpoints. It is not necessarily so that every voice packet sent will arrive at the other end. Most of the voice compression algorithms (vocoders) used in IP telephony, such as G.723.1 and GSM, have built-in capabilities to synthesize lost voice packets in the decoder. These algorithms are able to build reasonable models of the missing speech data by interpolating from previously received packets, but this is only effective up to 5 percent or 10 percent packet loss. And although these vocoders' lost packet reconstruction schemes work best on random packet loss, the real packet loss observed on the Internet and other IP networks is highly correlated and bursty.

SOLUTIONS FOR IMPROVING VOICE QUALITY
Let's examine several approaches to overcoming these challenges and providing a high quality of voice service to the IP telephony user.

Managed Networks
Many large enterprises and service providers have very tightly managed IP networks, which can be configured to support real-time voice traffic. For years, enterprises have managed private leased-line and frame relay networks to transport not only delay-sensitive data traffic but voice as well. Today, most of the IP telephony service providers offering real-time voice services have deployed their own managed IP network backbone and keep all of their traffic "on net." Latency, jitter, and packet loss can be minimized by the carrier because the carrier configures and controls all of the hardware and interconnection inside its network.

"The" Protocol for QoS?
Fortunately, it appears that TCP/IP, "the" protocol for networking, has a shot at becoming "the" protocol for voice transport as well. Unfortunately, there is no particular protocol, or set of protocols, which has emerged as "the" protocol for QoS. ATM is certainly being deployed in the core of the network, but has never fulfilled the promise of delivering end-to-end QoS directly to the desktop. In the past 24 months, there have been many different IETF and proprietary vendor protocols put forth as potential solutions to the QoS - and now Class of Service (CoS) - challenges faced by both real-time and non-real-time IP traffic.

It seems likely that a number of different protocols will be used and that convergence on a ubiquitous, universal QoS protocol won't happen. This means that enterprises and service providers will be picking from a limited set of vendors, each offering an "end-to-end" solution based on a particular, likely proprietary, protocol set. These protocols then become just tools that help the service provider manage its network more easily. "The" protocol does not yet exist, which will allow the Internet, or any unmanaged IP network, to truly deliver QoS for voice traffic.

Fatter Pipes
Many network managers believe that the explosion of bandwidth availability will solve QoS problems. With the rapid decrease in the cost of transmitting a byte of data over SONET, Gigabit Ethernet, and other emerging high-speed networking technologies, it may be best to over-provision the network to minimize packet latency, jitter, and loss. Simply replacing switches, routers, and transmission infrastructure with the latest technology may allow the network architect to stay ahead of congestion and network contention.

Many newly formed service providers, such as Qwest and Level3, have taken exactly this approach. With so much excess fiber capacity, they are not even interested in doing voice compression on the telephony traffic. This means that voice data running over their networks will be using more than 64 Kbps due to the additional IP packet overhead. This is a wonderful, clean-sheet-of-paper approach to building a high-capacity IP network, which is able to carry voice as just another data type. With a big enough pipe, no problem!

I'm certainly drawn to the elegance of the "fatter pipe" solution, but doubt that we'll ever see a successful network, which can be over-provisioned indefinitely. The growth of communications demand seems to expand to fill available communications bandwidth, just as my 2 GB hard drive is filled with a fat OS, huge applications, and megabytes of e-mail attachments.

THE KEYS TO SUCCESS
There are many challenges in delivering high-quality, real-time voice over IP networks. New protocols are being developed to address the underlying technical issues of packet latency, jitter, and loss, which affect the transmission of real-time flows through IP networks. These protocols, in conjunction with the availability of much higher capacity networking equipment, will allow enterprises and service providers to build and manage IP networks capable of delivering high-quality voice services over IP. Even with new protocols and fatter pipes, the keys to success in delivery QoS for IP telephony will be network architecture and management. Only managed networks will provide acceptable quality of service for IP Telephony applications.

Joel Hughes is director of IP Telephony at Natural MicroSystems Corporation. Hughes oversees the company's efforts in the Internet telephony market, including product development and new business development. Natural MicroSystems is a leader in open telecommunications, providing hardware and software technologies for developers of high-value telecommunications solutions. The company's state-of-the-art technology enables a growing international network of OEMs, VARs, systems integrators, and service providers to reduce time to market, leverage development resources, and offer truly global communications products. For more information, visit the company's Web site at www.nmss.com.