[ Home | Glossary | Acronyms | Links | Contact us ]

Cellular Networking Perspectives

David Crowe’s Wireless Review Magazine Articles

June 1, 2001 Issue

All IP Networks

Physicists dream that one day there will be a single equation to explain all physical forces. While this strongly attracts many, other physicists believe this is the impossible dream. In telecommunications, the parallel is the All-IP network, where video, sound, text files, control information (signaling) and voice all travel over a single massive packet-based backbone. The most challenging aspect of this may be taming the millions of streams of voice, something that may make understanding Superstring theories seem like child’s play.

To people who are used to computer communications, the requirements of voice seem minimal – only 32 kbps to 64 kbps of data per conversation in the fixed network, and as low as 8 kbps on many wireless communications protocols. Meanwhile, even low-end communications is using 10,000 kbps wired ethernet or its wireless equivalent, IEEE 802.11. The most modern computer equipment comes with 100,000 kbps ethernet (100 Base T), and Gigabit ethernet (1,000,000 kbps) is just around the corner from mass market acceptance.

The major challenge with voice over IP (VoIP) is obtaining a high Quality of Service (QoS), particularly in markets like North America where telephony consumers are used to extremely high quality and low cost. While everyone knows quality when they see it, defining it precisely is tricky. Technically, QoS is a list of characteristics that must be met in order for a protocol stack to support a particular application. Email, for example, is quite flexible on some aspects, allowing considerable delays in transmission, and could accept delivery of messages in any order, but it has no tolerance for the delivery of erroneous information to the end user.

Voice, by contrast, is an example of isochronous communications. There is some tolerance in the delivery of this information (400 msec according to an Alcatel study, almost half a second), but this delay must be relatively consistent (i.e. voice has low tolerance for jitter). To deliver one packet with a latency of 400 msec and the next one with only 10 msec (and perhaps out of order) could totally destroy the voice quality. Making life somewhat easier, voice is relatively tolerant to errors or loss of packets, more so than most forms of data communications that prefer no-delivery over delivery of data to the end user with corrupted bits.

Some QoS characteristics for important services are listed below, although they vary significantly, based on consumer expectations. For example, businesses may expect near real-time email delivery, while consumers may be satisfied with delays of many minutes, and in countries where the PSTN is of very poor quality, the QoS expectations of VoIP will be correspondingly low.

Table 1: Permissiveness in QoS Characteristics

Voice in the traditional phone network is transmitted over facilities and routed through switches that are perfectly adapted to its isochronous nature. Voice is digitized to a continuous 32-64 kbps stream, and transmitted over facilities that assign that amount of bandwidth for the entire duration of a call. Switching equipment is very simple, merely copying bits of digitized voice from an incoming timeslot to an outgoing timeslot. Apart from ensuring high quality of the facilities, there is no provision for error handling; systems rely on high quality facilities to make errors extremely rare. Furthermore, overload cannot occur once a call is established, because the stream of voice traffic for each call never varies in size. Overload can only occur with signaling traffic, which only affects the ability to set up calls.

There are a number of drawbacks to this seemingly perfect world. Telecom equipment is, on a bit-per-second basis, very expensive, and it operates in a network that is largely separate from public and private internets. The dream is to put digitized packets of voice on the same network as everything else, meaning that only one network has to be built, managed and upgraded in order for a carrier to provide a full suite of services. Furthermore, there are many convergent services that can integrate a number of different types of information, such as a video conference integrating streaming video with voice, text and graphics. Another advantage of IP telephony is the ability to avoid international tariffs, which further (although artificially) reduces the cost. This is an advantage that many governments would love to remove, if only they could figure out how to do it.

This goal is laudable, but voice is not handled well by an IP network unless considerable adjustments are made. One of the problems is that IP works on a ‘best effort’ basis to deliver packets, assuming that applications like TCP can re-send packets if they disappear. Furthermore, routing of IP packets is much more complex than copying bits across a time-slot interchange switch. The robust routing used by IP allows different packets to take different routes, arriving at different times. Therefore, unless something is done to handle isochronous traffic like voice in a different fashion, the quality of conversations will vary tremendously, declining in quality as the amount of traffic increases.

Maintaining an adequate QoS for voice services is acknowledged as one of the major challenges for VoIP. ETSI has set up its TIPHON group for this purpose, aiming to enhance existing protocols such as SIP and H.323.

One way to improve the QoS of VoIP is to run a managed network. Enough extra bandwidth can be provided that traffic peaks do not cause serious congestion. This is appropriate for enterprise-wide private network solutions, but it is increasingly hard to achieve as the network becomes more public. Another method is to work harder to optimize the path through which the voice traffic passes. Avaya, for example, has created techniques that it calls Shuffling and Hairpinning that can remove servers from the stream of voice traffic, eliminating some of the delays, jitter and sequencing errors. This is similar to the redirection and path minimization techniques used in ANSI-41 for optimizing the path of wireless calls. Other techniques are DIFFSERV (Differentiated Services) and RSVP (Resource Reservation Protocol) that provide information that IP routers can use to give more desirable treatment to voice packets.

While major challenges remain for VoIP in fixed networks, even greater hurdles confront those trying to support wireless access. Although wireless protocols already digitize voice efficiently (8-13 kbps, or even lower during some modes of operation in CDMA systems), the addition of IP overhead will destroy these gains. According to an Ericsson white paper, compression and other techniques will help achieve VoIPoW (VoIP over Wireless) with 3G systems, but will still not achieve the quality or robustness of circuit-switched connections without further work.

Voice over IP systems will likely achieve the most success in areas where either cost or the benefits of convergence with data services are of greater importance than voice quality. Ignoring VoIP now, because of its imperfections, would be to make the same mistake as those who ignored wireless in the early 1980s.