VoIP Meets Wi-Fi: Internet Phone Calls Go Wireless

It's an almost irresistible match. VoIP (Voice over Internet Protocol) provides inexpensive phone calls over the Internet. Wi-Fi, the wireless LAN technology, cuts the corded connection. Together, in a combination that's increasingly called VoFi, they allow inexpensive, wireless phone calls through any accessible Wi-Fi connection.

And the match of VoIP and Wi-Fi has been a happy one. Large businesses have embraced VoFi to drastically reduce their long-distance phone bills, and it has worked well for them. If voice quality doesn't always match the plain old telephone system (POTS), it at least comes close.

But now VoFi is moving into the consumer world, where operating conditions are a little rockier. VoIP pioneer Vonage begins selling a portable Wi-Fi phone (Figure 1) this summer, and it will have to work on standard 802.11, not special enterprise networks that have been modified and tweaked to handle real-time voice requirments. How VoFi will sound in this environment remains to be seen.

Figure 1:  VoFi service becomes available this summer on the F-1000 Wi-Fi phone made by UTStarcom and marketed by Vonage.

Even VoIP, not to mention VoFi, has to overcome some tough conditions to deliver natural-sounding speech over the Internet. In both cases, the underlying difficulty is simply that packet networks, such as LANs, Wi-Fi, and the Internet, weren't designed for voice traffic. In order for a conversation to sound natural, voice signals must travel without noticeable delay between the conversants. In packet networks, however, there's no guarantee that packets containing voice data will arrive in a timely manner or even at all. Wi-Fi, being wireless and inherently less reliable than wired LANs, makes timely delivery and high voice quality even less likely. Consumer Wi-Fi, often operating under less-than-optimum conditions, makes the situation even worse.

In wired VoIP, fortunately, techniques for dealing with inconsistent packet delivery have solved many voice-quality problems. When problems arise, it's often because the Internet is experiencing unusual problems or a LAN is congested with data traffic and doesn't give VoIP applications priority.

In VoFi, as in wired VoIP, most voice-quality difficulties revolve around latency, jitter, and packet loss. Packets take time to reach their destination (latency), that time is variable (jitter), and some packets never make it (packet loss). Latency, jitter and packet loss are all higher on Wi-Fi than on a wired LAN and therefore create more difficult voice-quality problems.

Some latency, of course, is present in any VoIP or VoFi application. An Internet ping test might reveal, for example, a 67-msec round trip between Boston and Dallas; 133 msec between Boston and Geneva, Switzerland; and 262 msec between Boston and Adelaide, Australia. The International Telecommunication Union (ITU), in standard G.114, recommends a one-way delay of less than 150 msec for good conversation quality. By that measure of quality, all of the above times are acceptable, even when packets travel halfway around the globe.

Problems arise, though, when delays become longer (Figure 2). Internet latency is very unpredictable and can increase unexpectedly. Data congestion on LANs can also cause significant delay, and normal VoIP processing, such as speech coding and decoding, adds to delay. When one-way latency exceeds about 200 msec, the flow of a conversation gets interrupted and the participants tend to interrupt each other and talk at the same time. High latency can also cause echos.

Figure 2:  Network delays reduce sound quality in VoIP and VoFi phones. The Mean Opinion Score is a measure of sound quality as judged by phone users. (Graph courtesy of Global IP Sound)

The most commonly encountered VoIP problem is jitter, the arrival of voice packets at their destination with irregular timing (Figure 3). The result of high jitter (over about 50 msec) is that speech sounds jerky. Severe jitter can actually cause voice-data packets to arrive out of order, causing sounds to be jumbled.

Figure 3:  Jitter, which indicates the arrival of voice-data packets with irregular timing, degrades voice quality without use of a jitter buffer. (Graph courtesy of Intel Corporation)

Jitter is particularly prevalent in wireless LANs, and therefore a significant problem for VoFi. According to Jan Linden, vice president of engineering for Global IP Sound, it's not unusual for jitter in an 802.11b network to have a standard deviation of 50 to 100 msec. In contrast, an enterprise LAN rarely experiences jitter greater than a couple of milliseconds.

The most severe voice-quality problems, however, result not from jitter, but from packet loss. For each packet of voice data that doesn't arrive at its destination, at least 20 or 30 msec of speech-depending on the voice codec used-will simply be absent. In some situations, losing even 1% of packets can result in a significant loss of voice quality.

And Wi-Fi networks tend to have a lot of packet loss, unfortunately, because there's often a poor connection between a Wi-Fi user and a Wi-Fi access point. It a data packet doesn't make it from sender to receiver, the network can resend it, of course, but resending a voice packet merely adds to the problems of delay and jitter. And, if resending a packet causes too great a delay, a VoFi system will discard it anyway.

To minimize the effects of jitter, VoIP systems use a jitter buffer. A jitter buffer collects incoming voice packets, which may arrive in irregular time intervals, and plays them back with regular spacing to give the packetized voice data a natural sound.

A jitter buffer can be a two-edged sword, however. Although a jitter buffer can remove the jitter from arriving packets, it does so by increasing overall delay. A small buffer will introduce less delay than a large buffer, but it can overflow and lose voice packets when jitter is high, with a resulting loss in voice quality. A larger buffer, on the other hand, will lose fewer packets but can introduce unacceptable delay.

A dynamic jitter buffer solves the buffer-size problem. Implemented in software, a dynamic jitter buffer detects changing network conditions and adjusts its length accordingly. Typically, a dynamic jitter buffer is configurable for a maximum percentage of packets that it will be allowed to drop. In effect, it trades off jitter and packet loss in an attempt to get the highest possible voice quality.

Packet loss, depending on how it occurs, can have a widely varying effect on voice quality. According to some sources, losing up to 10% of all voice packets, still yields acceptable voice quality. Other sources, however, say even a 1% loss can result in terrible quality. The discrepancy, apparently, is because packet loss can be either random or bursty. Sometimes a VoIP system will lose 20% to 30% of packets over a period of several seconds. This situation creates a very noticeable voice-quality problem, even though the percentage of packets lost over time might be low.

To deal with packet loss, VoIP and VoFi systems use techniques known as packet loss concealment. These techniques don't recover the lost packets; they just mask the lost packets' undesirable effects on speech. Surprisingly, perhaps, these techniques are very effective, enabling a tolerance of up to 30% packet loss in some cases (Figure 4).

Figure 4:  Packet loss concealment can significantly improve VoIP and VoFi voice quality. The lower (green) line in this graph shows voice quality without packet loss concealment. The two upper lines show improvements with different packet loss concealment techniques. You can hear some effects of packet loss concealment on the Global IP Sound Web site.

Essentially, says Jan Linden of Global IP Sound, packet loss concealment involves adding to each transmitted voice packet some information from the preceding packet. Then, if a packet doesn't make it through to the receiving end, the receiver can wait until it gets the next packet and use the information in that packet to describe the preceding packet.

The most common approach to packet loss concealment is simply to replace a lost packet with the entire preceding packet. "Because speech is correlated over time," Linden says, "you get something that is similar to what it should have been." But, Linden adds, there's a downside-a transition effect that doesn't sound very good.

A better approach to concealing packet loss, Linden says, is to divide a sequence of voice data into two chunks and send them in separate, consecutive packets. With appropriate data sampling, each packet contains a rough representation of the same sound sequence. If both packets arrive, the receiving end can use both to reconstruct the sound sequence perfectly. If one packet gets lost, the system can use the other one to create the same sound with lower, but still reasonable, quality.

802.11e doesn't, however, guarantee that voice data will always arrive as quickly and as regularly as it needs to, although it does increase the likelihood. Also, to be very effective, networks using 802.11e must propagate their QoS requirements all the say from source to destination, and that propagation depends on all network components being 802.11e-compliant.

It's also important for VoIP and VoFi applications to have adequate bandwidth, even though bandwidth isn't usually much of a constraint. Even an 802.11b network is capable of 11 Mbps, after all, and 802.11a and 802.11g provide 54 Mbps. A residential broadband Internet connection, often the slowest link, usually delivers at least 1 Mbps.

Nevertheless, bandwidth can be a factor, because wireless LANs often don't deliver anything like their specified data rates. Less-than-optimum conditions can easily decrease the data rate for 802.11b, for example, to 2 Mbps or even 1 Mbps. Add to that some heavy network traffic and bandwidth can start to be limiting.

Various voice codecs can reduce VoIP and VoFi bandwidth requirements, but they must be used with care. In traditional telephony, codecs that meet the requirements of ITU standard G.711, encodes speech into a 64-kpbs data stream. In VoIP, a G.729 codec requires only 8 kbps and yet provides virtually identical sound quality. With higher compression, however, packet loss can be more noticeable, because each packet contains more voice data.

Roaming can effect voice quality, because the handoff of a VoFi connection from one access point to another doesn't occur instantaneously. It takes a finite amount of time, and if it doesn't happen quickly enough, speech quality will suffer.

Cellular, of course, has never provided great voice quality anyway, but most users have tolerated reduced quality to get the benefits of mobility. The combination of Wi-Fi and cellular will also be mobile, but VoFi by itself will compete not just with cellular, but also with traditional land-line phones, where quality is almost always excellent.

Perhaps only time will tell, then, if users will value cost and convenience over quality or, alternatively, if VoFi will come to match the quality of the traditional phone company. Either way, it looks like a lot of our phone calls will be going over the Internet.