Solving the WLAN VoIP Challenge: Part 1

John Waclawsky, Cisco Systems; and Jim Gunn, Communication Consultant

July 14, 2004

John Waclawsky, Cisco Systems; and Jim Gunn, Communication Consultant

Delivery of voice and video services is the next frontier for the wireless LAN (WLAN) sector. But, supporting real-time traffic on best-effort WLAN links poses quite a challenge for the design community. Three areas that are particularly challenging are quality-of-service (QoS), throughput, and end-to-end delay.

In Part 1 of this series, we'll look at the the challenges that QoS, throughput, and delay provide in delivering real-time VoIP and video over IP streams over an 802.11 WLAN link. In Part 2, we'll provide a detailed analysis showing how 802.11a, b, and b/g systems can deliver the throughput and delay needed to support voice and video services.

Typical WLAN Environment Characteristics
An overview of IEEE 802.11 WLAN standards is presented in Table 1 below.

WLAN 802.11 supports multiple frequency bands so it can be deployed to avoid adjacent LAN interference. The different versions use either direct sequence (DS) or frequency hopping (FH) spread spectrum transmission (initially in the 900 MHz or 2.4 GHz bands). 802.11a provides a minimum bit rate of 6 Mbit/s to a maximum of 54 Mbit/s in the 5 GHz band and 802.11b provides bit rates from 1 to 11 Mbit/s in the 2.4 GHz band. 802.11g operates in the same band as 802.11b and provides up to 54 Mbit/s bit rates. All WLAN systems have Ethernet compatible interfaces (same headers, CRC and frame size) and use connectionless best effort service at the link layer (no persistence).

Table 1 IEEE 802.11 WLAN Standards

The typical problems with wireless LAN environments are:

• CSMA/CA — Collision detection is difficult to implement in wireless networks because it requires simultaneous transmission and reception on the same band. Access is controlled by collision avoidance (CA) instead. Any access delay is typically small delay due to the higher speeds involved.
• Transmission and propagation delays are low due to small coverage area and high bandwidth.
• Typical frame loss rates are less than 2.5 percent (at maximum frame size).
• Power control is typically not provided (but power usage can be a problem).

WLAN throughput can vary significantly due to the nature of radio transmissions and the behavior of the 802.11 protocol in various LAN situations. For example the number of stations and their distance from the access point will affect performance as the closer and faster devices wait their turn while the more distant and slower stations transmit. An RF signal loses power with distance. So signal strength will diminish as it moves through the air. For example, the line-of-sight propagation range for 802.11b using the maximum transmit power setting and a 2.2-dBi gain diversity dipole antenna provides approximately 1 Mbit/s of capacity at 800 feet. As distance decreases, capacity increases to 11 Mbit/s at 200 feet or less.

Signal loss through metal and concrete can be much more substantial than through the atmosphere. So increasing the distance between the user and/or increasing the number of obstacles between the user (e.g. walls) and the access point reduces signal strength and decreases the speed/throughput in a step fashion as shown in Table 1.

It is best to take measurements to determine the actual speed for any situation. Most WLAN equipment has monitoring software, which will tell what speed the device and access point (AP) are using. Besides distance a number of other factors will impact throughput such as the type of antenna used at the access point and the power output from the AP and the wireless device (they are not all equal and vary by manufacturer).

In addition metal objects in a room, such as a filing cabinet at the office or a refrigerator at home, can both block radio waves and reflect them causing multi-path interference. Also, the actual capacity available is a function of usage because as the traffic increases the number of collisions, and subsequent retransmissions, increase. Finally devices such as microwave ovens and cordless phones that use the same frequencies as the 802.11 equipment can cause significant interference. As a general rule, 802.11b frequencies contain more devices that could cause interference.

Increasing WLAN capacity is not as simple as adding more access points because each access point is a potential source of interference. The signal from one access point can interfere with another access point and reduce capacity. The 802.11 protocols attempt to mitigate interference concerns by allowing different access points to operate on different channels so they don't interfere with each other when in close proximity. The various flavors or 802.11 employ from three to 12 channels (regulations vary around the world, for example 802.11a typically uses eight channels but there are provisions that allow additional channel allocation for higher power, longer distance "outdoor" applications). Using 802.11a, with a larger number of channels for example, is a good choice for high capacity systems. This is true because a large number of channels make it easier to assign channels to access points in a way that minimizes interference for a particular situation (physical layout).

Wireless Delays and End-to-End Throughput
To understand and estimate best-case wireless throughput, it is necessary to investigate the source of wireless delays and how most wireless link layer protocols provide reliability. Wireless link reliability is often achieved with processing intensive techniques such as forward error correction (FEC) and interleaving, in conjunction with some kind of (positive) acknowledgement scheme such as automatic repeat request (ARQ). Wireless link layers also significantly contribute to packet delays being experienced due to the forward and reverse direction media access control (MAC) traffic flows required for reliability and control of the wireless medium. Therefore, wireless link layer schemes often delay the transmission of new traffic over the wireless link until the old traffic is received correctly. Often, new traffic can only be transmitted after both the forward and reverse direction MAC layer control and reliability delays are experienced.

Link-layer acknowledgement schemes are very close cousins of the TCP windowing scheme with its acknowledgements. In addition, any TCP or UDP traffic flowing end to end that includes a wireless link will experience wireless link layer delays. Therefore end-to-end TCP and UDP delay and throughput can be limited by the wireless link.

Understanding IP Datagram flow over WLANs
To understand delays in a WLAN network, a designer must first understand the components of delays involved with IP datagrams traversing a wireless link. An IP datagram could be carrying either TCP or UDP traffic. We will begin first by focusing on the more complex scenario of a TCP segment and its acknowledgement over a wireless LAN link. After our analysis of TCP traffic, we then analyze the simpler UDP traffic case with VoIP. This IP traffic situation is shown in Figure 1.

Figure 1: diagram showing end-to-end performance on a WLAN link.

Notice that 802.11 environments are shared-media LANs that operate half duplex over a (potentially) noisy radio channel. Since a radio's receiver is turned off when the transmitter is activated, it is impossible to detect collisions. In addition, the AP must be able to hear all stations associated with it.

It is not, however, a requirement that all stations can hear each other from opposite ends of the coverage area (hidden node problem). So by using carrier sense multiple access with collision avoidance (CSMA/CA; a close cousin of Ethernet's CSMA/CD), the possibility of collisions can be substantially reduced (but not eliminated).

The broadcast nature of wireless WLANs requires that the physical/radio layer must be shared. There are insufficient channels to allow more that three (under the 802.11b or 802.11g standards) or 12 (with 802.11a) unique connections per access point. Communications between an access point and an 802.11 device are limited today to a single connection and cannot handle multiple channels simultaneously. Because of this we will focus our performance analysis throughput discussion on a single channel operating in isolation.

802.11 LANs employ a distributed control function (DCF) as the primary media access to avoid collisions. The fundamental collision avoidance mechanisms are a set of spacing intervals. Even if the medium is idle, stations always wait for an inter-frame spacing interval before transmission. The most common inter-frame spacing intervals are the DCF inter-frame spacing (DIFS) and the short interval inter-frame spacing (SIFS). Also, a rarely implemented point control function inter-frame spacing (PIFS) can be used to support time sensitive QoS traffic.

In addition, each device uses a network allocation vector (NAV) counter for virtual carrier-sense functions. The NAV is used to determine how long the medium will be busy in the future. The NAV counter can be set by a request-to-send (RTS) or clear-to-send (CTS) 802.11 frame, which is used to inform a node's neighbors to set their NAV.

A station assumes the medium is busy whenever the NAV is set and will not try to transmit for that time period. In addition, the waiting intervals can be used as the basis for a quasi priority mechanism since more important transmissions have shorter waiting times. For example, the SIFS interval is used before a station sends an 802.11 ACK. Since it is the shortest interval, this gives 802.11 ACKs the highest priority access to the channel.

Too Limited
Today, 802.11 QoS is considered too limited to support emerging real time traffic needs, such as VoIP. Extending QoS capability is being addressed by the 802.11e-working group. Even if capacity is abundant, traffic growth, changing traffic patterns or connection failures could result in impairments that impact isochronous traffic and is another argument for more effective QoS. The 802.11e standard is expected to be completed in late 2004. It currently specifies two additional mechanisms for QoS:

• An enhanced DCF (EDCF) improves the existing DCF MAC specification. EDCF adds an arbitration inter-frame space (AIFS) where higher priority traffic will have lower AIFS idle times. EDCF also supports up to eight priority traffic classes that map directly to the RSVP protocol priority levels.
• A hybrid coordination function (HCF) builds upon DCFs round-robin polling mechanism by providing more intelligent polling algorithms.

The typical 802.11 transmission process is partially shown in Figure 2. Assume a packet is ready to be transmitted. The station senses for radio energy and checks if its NAV counter is zero, to determine if the radio medium is idle. Notice the 802.11 protocol listens before transmission, so the time to transmit a frame is really the combination of listening time and transmission time. This waiting time (even though it is short) decreases throughput.

Figure 2: Diagram showing WLAN transmission activity and the partial DCF process.

It is possible for multiple senders to sense an idle network and simultaneously transmit their messages. When this occurs, both messages collide and are corrupted. This requires retransmissions, which wastes bandwidth and decreases capacity. When 802.11 is lightly loaded an occasional collision is not a problem. But as utilization increases so do collisions which can significantly limit performance and capacity.

If a station has data to send and the channel is idle, then the station waits one DIFS interval and transmits. However, if the radio medium becomes busy, while it waiting during its DIFS interval to transmit, then it will defer its transmission until one DIFS interval after again sensing an idle radio channel.

If a channel is busy otherwise, the station will set its back-off counter with a random number of fixed duration time slots. When the back-off counter reaches zero, the station again waits one DIFS interval and then begins its transmission. If another station begins transmitting while the back-off counter is running, then the counter is stopped while the other transmission is in progress. It resumes when the radio medium becomes idle again.

Lack of ACK
If an ACK is not received after a transmission, a station assumes that a collision has occurred. The station will then increase (double) its back-off counter and repeat the process. Each successive unsuccessful transmission increases (doubles) the back-off counter again until it reaches the maximum value.

In today's 802.11 systems, all stations have equal access to the shared radio channel. There is no mechanism to ensure timing consistency. So as traffic increases, the time to access the channel will increase because stations will try harder to avoid collisions. In a way, when utilization increases, capacity is decreased.

In the WLAN examples to follow in Part 2, we will primarily consider the transmission of a 1500-byte IP packet containing a single 1460-byte TCP segment that generates a TCP acknowledgement. Note that the 802.11 standards allow much larger payload sizes than 1500 bytes. Most implementations, however, connect to existing wireless LAN networks with Ethernet. This limits the wireless frame payload to a 1500-byte maximum (This is the largest Ethernet payload size) plus 802.11 frame overhead. (In fact, meeting this size restriction is needed to obtain certification).

The 802.11 MAC layer specification calls for a positive acknowledgement for every frame transmission to deal with radio medium unreliability. Therefore, each TCP segment transmission and each returning TCP ACK transmission requires both the forward and reverse direction transmission of 802.11 MAC frames. One of the MAC frames carries the TCP segment and another has the MAC layer positive acknowledgement. A MAC frame also carries the TCP ACK and again a MAC layer positive acknowledgement is sent in response.

Consider a typical end-to-end session from a server terminating at a wireless client shown in Figure 1 above. When the TCP segment arrives at the client, a TCP ACK is usually generated.

For simplicity, we'll initially assume that when the TCP segment arrives at a wireless client the TCP ACK is transmitted back immediately. Therefore, a complete client-to-server TCP round trip includes wireless link activity, which at a minimum consists of transmitting four 802.11 frames. One wireless 802.11 frame is carrying the 1500-byte IP packet encapsulating the TCP segment. The other wireless 802.11 frame is carrying the TCP ACK (often just a payload of 40 bytes for the TCP and IP headers). Two other 802.11 frames are carried over the wireless connection to provide the 802.11 MAC acknowledgements. Please note that is some circumstances RTS and CTS control frames may need to be transmitted and will increase overall delay. This is all shown as MAC activity in Figure 1.

802.11b divides data into 8-bit symbols when running at its maximum speed (since each symbol is the same as an 8-bit byte, we will wait to talk more about symbols in 802.11a and 802.11g where symbols provide encoding for a lot more bits). In addition to the payload, frames carrying data have 36 additional bytes of 802.11 MAC frame header and trailer.

The 802.11 MAC header consists of 28 bytes of information for various control and management functions, error detection, and addressing. A SNAP header identifies the network layer protocol for the payload and adds eight additional bytes. The total size of the 802.11 MAC frame for the 1500-byte IP packet becomes 1,536 bytes (or 12,288 bits). The total size for the 802.11 MAC frame carrying only a TCP ACK is 76 bytes (or 608 bits).

On To Part 2
That wraps up our look at the real delay and throughput issues designers will encounter when trying to deliver real-time services over WLAN links. In Part 2, we'll further the discussion by showing how 802.11a, b, and b/g systems can deliver the throughput and delay necessary to support multiple PCM voice channels.

About the Authors
John Waclawsky is a technical leader in Cisco Systems' Mobile Wireless Group. In the past, he has also served as technical committee chair of the Mobile Wireless Internet Forum (MWIF). John holds a master's degree in Computer and Information Sciences from the University of Pennsylvania as well as master' and Ph.D. degrees in Computer Science from the University of Maryland. He can be reached at jgw@cisco.com.

Jim Gunn is a communication consultant, market researcher, and associate at market research firm Forward Concepts. Jim has a BSEE and MSEE from Oklahoma State and Ph.D. in electrical engineering from Southern Methodist University. He can be reached at jimgunn@ieee.org.