Getting Physical With WiFi Telephony

The most unique aspect of WiFi telephony compared with wired VoIP is the affect of using the airwaves as the physical media for carrying packets to and from the wireless telephone devices. A lot of technology goes into making that wireless connection as transparent to the user and the host telephone system as possible, but there are a couple of considerations and tradeoffs to deal with in choosing a specific WiFi network implementation based on the physical radio standard: 802.11a, 802.11b, or 802.11g.

These Physical Layer (PHY) standards describe the radio frequency (RF) spectrum used and how it is partitioned, the modulation type, and power levels. Some of these can be factors in selecting a particular WiFi product, such as the RF spectrum used, while others are only important to product developers to ensure interoperability, such as the modulation scheme used to turn binary data into radio waves. The specific implementation details arent important, except for the ones that affect interoperability, performance and capacity.

The initial 802.11 standard had three PHY options: two using spread-spectrum radio technology, frequency hopping (FH) and direct sequence (DS), and a third using infrared optical technology. Both radio standards used the 2.4 GHz band, which was available for use without licensing restrictions nearly worldwide. The optical approach never took off so the standard now rests peacefully in the standards graveyard.

Even while the initial 802.11 standard was being ratified, work began on standards to increase data rates beyond 2 Mb/s. Two parallel efforts were initiated. The 802.11a Task Group took the approach of moving up in the RF spectrum to where bandwidth was more plentiful and could support data rates above 50 Mb/s. Meanwhile, the 802.11b Task Group stuck with the 2.4 GHz band and looked to build on the DS standard to increase data rates to greater than 10 Mb/s. There were also some efforts to improve on the FH standard outside of the 802.11 committee, but it never caught on primarily due to the widespread industry support of the 802.11a and 802.11b efforts. Following on to these was the 802.11g Task Group, which leveraged the higher speed technology of 802.11a for use in the 2.4 GHz 802.11b spectrum, which offered a lower cost alternative to 802.11a with backward compatibility with legacy 802.11b client devices.

The 2.4 GHz band used by 802.11b/g has a couple of advantages. First, it is now available throughout the world for WiFi networks, although it is also available for other consumer and enterprise wireless products such as cordless phones and RFID tags. The other advantage is that radio signal propagation is pretty good at 2.4 GHz, both in free space but also going through walls. But the biggest limitation of the 2.4 GHz band is that is only has about 80 MHz of radio spectrum available which is partitioned into three non-overlapping channels. Adjacent WiFi access points need to use different channels to prevent interfering with each other, so having only three channels to work with is a network design challenge, particularly in multi-floor, three-dimensional deployments.

On the other hand, above 5 GHz where 802.11a plays, there is enough radio spectrum available to support twelve non-overlapping channels. That makes deploying an 802.11a network much easier in terms of avoiding co-channel interference, but the tradeoff is that radio signal propagation is worse at higher frequencies, so the coverage area for an 802.11a access point is usually less than that of an 802.11b access point. But having more channels allows for denser network deployments and lessens the chance of running up against interference from a neighbors WiFi network. Most enterprise WiFi infrastructure vendors support both 2.4 GHz 802.11b/g and 5 GHz 802.11a PHY standards, and in many cases both are available in a single access point (AP). Besides the advantages or disadvantages associated with the RF spectrum used, what is the impact of the higher data rates supported by 802.11a and 802.11g for WiFi telephony?

Counting Calls
The actual bandwidth used by a telephone call over a WiFi network depends on several factors, starting with the codec used to generate a digital data stream just as with any VoIP implementation. The codec determines the number of bytes that will be carried in each packet for each sample period. For example, full-rate G.711 coding which turns an analog voice stream into a 64 kb/s data stream requires 80 bytes for every 10 ms sample, while G.729 compression at 8 kb/s requires only 10 bytes in a 10 ms sample. Some efficiency can be gained by sending multiple samples in a single packet, reducing the amount of overhead by reducing the number of packets. The tradeoff with sending more information in packets less frequently is that it can affect voice quality by adding delay and echo, and the affect of lost or dropped packets is much more significant.

The overhead required to transmit a voice packet over the wireless LAN and through the wired IP network adds significantly to the packet size. Working our way up the OSI model, headers are required for the 802.11 media access control (MAC) and the logical link control (LLC) portions of the Data Link Layer, followed by the IP header for the Network Layer, and finally the Transport Layer which typically uses something like User Datagram Protocol (UDP) under Real-time Transport Protocol (RTP). Stripping off the 802.11 MAC layer, a wireless VoIP packet looks like any other VoIP packet once it hits the wired network. All together, the various higher layer headers added to a voice packet can more than double the packet size.

WiFi APs and client devices are designed to drop back to lower data rates as the signal quality degrades. This is advantageous for maintaining WiFi coverage into every nook and cranny, but it also means that more of the APs bandwidth is used up when the data rate drops, since it takes longer to transmit the same amount of information. Most APs allow the minimum data rate to be selected to prevent low-rate wireless connections from hogging bandwidth, but this requires the deployment to be done such that devices can always operate at the minimum rate or above, no matter where the user is.

Differing from wired Ethernet protocols, which detect and deal with packet collisions, WiFi uses a carrier sense multiple access/collision avoidance (CSMA/CA) protocol to minimize the chance of two devices transmitting over the air at the same time. This protocol requires devices to wait a certain amount of time and then listen to determine if the radio channel is free for them to transmit (referred to as listen before talk). These waiting periods are based on a fixed time followed by a randomly selected time within a certain range, and WiFi quality of service (QoS) mechanisms such as WiFi Multimedia (WMM) assign shorter waiting periods to higher priority applications such as voice and video. So what does this have to do with bandwidth utilization? Well, the waiting time has the affect of wasting available bandwidth. For example, even if just one WiFi device is associated with an AP so there is no chance of another device transmitting at the same time, that device still has to wait the time prescribed by the CSMA/CA protocol. A typical waiting time of 50 microseconds is equivalent to transmitting 550 bits at 11 Mb/s, or about 70 bytes. These timing gaps reduce the effective bandwidth of the AP and need to be taken into account when calculating the actual call capacity for WiFi telephony.

That is why it is more accurate to quantify voice traffic in terms of percentage of bandwidth utilized instead of packet size or data rate. This gives a more realistic measure of the maximum number of simultaneous calls supported by a single AP. For example, taking into account the total packet size and timing gaps gives us around five percent bandwidth utilization for G.711 at 11 Mb/s with 802.11b. So the theoretical maximum number of calls per AP for this example is 20. But that doesnt account for real-world conditions where collisions happen, users roam around and need to handoff to other APs, and other devices need to use the WiFi network. A more realistic maximum takes into account the headroom needed for these things, so capping the real-world maximum capacity to 12 simultaneous calls at 11 Mb/s is more reasonable. But if the WiFi coverage isnt good enough to maintain the 11 Mb/s data rate, the bandwidth utilization goes up significantly. At the minimum 802.11b data rate of 1 Mb/s, a G.711 call u ses more than 20 percent of the AP bandwidth in an 802.11b network, dropping the traffic capacity to only three or four simultaneous calls.

Translating Calls To Users
Telecom managers know that you typically dont need a call resource for every telephone. An enterprise PBX has fewer trunk lines for outside calls than the number of telephone sets installed. The number of trunks is calculated using a probabilistic model of telephone usage that is based on the acceptable chance of a caller wanting to make an outside call and a trunk not being available. The theory behind telephone traffic planning has been around since the early days of the Bell System, and provides a means to determine the actual number of users that can be served by a WiFi access point based on the maximum number of simultaneous calls it can handle. Traffic analysis for WiFi telephony is a little more complex though for a couple of reasons. First, different geographic area may have different traffic requirements. Users may congregate in certain areas such as cafeterias and smoking areas, so the traffic engineering may vary by location and anticipated usage patterns. Second, WiFi telephone users will pr obably have higher than average usage profiles, at least for the initial users since their communication needs were sufficient to justify the investment in their wireless handsets in the first place.

There are a few different traffic models available that vary based on the number of users and how blocked calls are handled. A good source for more information on telephone traffic models, including online calculators, can be found online at www.erlang.com. Of course, the results from the calculations are only as good as the data you put in, so you need to have a good sense of what kind of traffic load to expect from WiFi telephone users.

Choosing which flavor of 802.11 radio technology requires looking at several factors, not the least of which is the bandwidth requirements of both wireless voice and data applications. Most enterprise WiFi telephony applications can be supported by the most widely deployed 802.11b standard, although its maximum data rate of 11 Mb/s may be insufficient for multimedia and other high-bandwidth data applications. Deploying an 802.11a WiFi network may require replacing existing wireless client devices, since most WiFi enabled laptops and PDAs only recently started offering 802.11a radios. A safe bet chosen in many enterprise WiFi deployments today is to opt for dual-radio APs, offering the high-capacity advantages of 802.11a along with broad device support using 802.11b/g. And leveraging that investment with WiFi telephony continues to deliver improved productivity, responsiveness, and mobility of employees in all kinds of enterprise applications worldwide. IT

Ben Guderian is vice president of market strategies and industry relations at Spectralink. For more information, please visit the company online at www.spectralink.com.