Facing the challenges of VoIP WLAN (VoWLAN) design: Part 2Besides the PHY and MAC overheads, there is another problem in VoWLAN that reduces the system capacity even further. An 802.11 BSS typically consists of an AP and stations. ( A typical example is Internet browsing. )
The 802.11 standard specifies that both the AP and the station use the same MAC. Specifically, the AP and the stations use the same media contention/access scheme and back-off periods (contention window values).
This means that if an AP and station both compete to get access to the channel, both are equally likely to get access to the medium. However, in a BSS, there is one AP and multiple stations. This means that if multiple stations and the AP compete for the media, it is more likely that one of the stations gets access to the media.This inherent fairness among the stations and the AP works fine for data communication since most data applications are highly asymmetrical - i.e., they have high download (from the AP to the station) traffic and low uplink (from the station to the AP) traffic. (Since stations rarely compete to access the media in data communication, the AP can easily get access to the media when it needs it.)
However, voice communication is bidirectional and highly symmetrical; i.e., the downlink traffic and the uplink traffic are very similar in terms of bandwidth requirements. This means that stations need to access the media much more often than in data communication.Given that stations and AP are competing on an equal footing and that there are multiple stations involved in voice communication in a BSS, the AP is much less likely to get access to the channel than all other stations combined.
Put another way, since the AP is as likely to get access to the channel as any other station, the probability of a station getting access to the media is higher than the probability of the AP getting access to the media if more than one station is competing for the media.
Combine this with the fact that the AP is handling much more traffic than any station and you have a system where the node handling the most traffic (AP) is not given priority over other nodes. ( Since all traffic must pass through the AP, the AP is almost handling as much load as the combined load of all stations in the BSS. )
This leads to a single point of back-up and congestion in a VoWLAN BSS.
In practice, if there are N wireless IP phones in a BSS making calls to wired networks, the AP is handling N times the load as compared to any other node in the BSS. However, fairness in 802.11 would allow the AP to access the medium only as much as any other node.
The bottom line is that the AP will not be able to transmit the traffic that it is receiving. Bad, as this may sound, things get worse. Since we are dealing with real-time traffic, a packet which gets delayed beyond a limit waiting in the AP queue is rendered useless.
From the 802.11 MAC perspectives, this situation arises because 802.11 requires that every station that finishes a transmission AND has a packet waiting in its queue MUST perform the random back-off. In the build up to a congested network, the AP will almost always have more than one packet in its queue (since all packets must go through it) so it will be backing off.
For an 802.11b network, this backing off will, on average, add a delay of 320 µs6 to its effective packet-transmission time (DIFS = 20 microseconds; CWavg = 32/2 = 16; Delay = DIFS * CW). All phones, however, in most cases will rarely have more than one packet to transmit unless congestion is really heavy or there are PHY layer problems (such as moving out of range of the AP).
The previous sections have argued that the maximum number of voice calls that can exist in a 802.11b BSS is severely limited. Analysis has quantized this observation. as shown in Table 5.1 below.
|Table 5.1: VoIP Capacities for 802.11b|
This analysis has been confirmed by observation too. Put briefly, the number of VoIP connections that can exist in a BSS increases with the use of larger packetization periods. However, higher payload sizes mean larger end-to-end delays in VoIP.
So, this increased system capacity comes at the cost of an increased end-to-end delay in VoWLAN systems. Hence, simply using very large packetization periods is not a viable solution for increasing system capacity in VoWLAN networks.
Earlier in Section 5.3.1, we explained that stations transmitting smaller packets are at a loss with respect to stations transmitting larger packets, since DCF (the default 802.11 MAC protocol) specifies the same back-off times for both regardless of the payload size they intend to use.
Now, consider an 802.11 BSS where some stations are being used for VoWLAN communication whereas others are being used for data applications like Internet browsing. Since VoWLAN stations would be using smaller payload sizes, they are inherently at a loss when competing with data stations for access to the media.
This situation is in fact the exact opposite of what is desired. Voice communication requires low delay and low jitter and it should therefore be given a higher priority. In 802.11 we have, in effect, voice traffic being given lower priority than data traffi c. What is needed then is a way to prioritize voice over data traffic.
The IEEE 802.11 group were aware of the problems with using DCF for real-time traffic. They therefore defined the point coordination function (PCF) to be used with real-time traffic (Figure 5.6 below).
The PCF divides time into superframes. Each superframe is further subdivided into a CFP (contention free period) followed by a CP (contention period).
|Figure 5.6. PCF Timing|
Therefore, in the PCF, CFP and CP alternate over time with the additional requirement that the CP be long enough to ensure the delivery of at least one MSDU. During the CP, channel access is still controlled as in DCF. However, in the CFP, channel access is controlled exclusively by the PC (point coordinator), which is a logical entity typically colocated at the AP.
The transmission of a beacon frame marks the beginning of a superframe. The beginning of the superframe also marks the beginning of the CFP. During this time, PC has complete and unilateral access to the channel and no other station may try to access it. In the CFP, the PC polls the stations for a chance to transmit data.
A polled station is allowed to transmit a single MSDU when it is polled. This polling continues till the end of the CFP, which is indicated by the transmission of a CF-End packet.
Ideally, the centralized tighter control of channel access by the PC should have allowed time-sensitive applications like voice to satisfy their QoS (delay and jitter requirements). However, there are many loopholes in PCF which make it unusable.
Consider what happens if the PC polls a station when the CFP is just about to end. Since the 802.11 standard allowed stations to start data transmission even if the MSDU delivery could not finish before the upcoming TBTT (target beacon transmission time), this basically meant that the CFP may actually end up the transmission of the next beacon.
This unpredictability in TBTT stemming from the unknown transmission durations of the polled stations during a CFP is a serious limitation of PCF since it has serious repercussions not only for the QoS of other stations but also for power management in stations.
Consider another case where the PC polls a station and the station has an MSDU to transmit. Since the maximum allowed size of the MSDU is 2304 bytes, this station can effectively capture the channel for a long time, which would in turn delay the transmissions of other stations in the BSS.
Note that this loophole stems from PCF specifying the polled station's channel access in terms of the number of MSDUs rather than absolute time. These problems have made PCF practically unusable and it has seen very little (if any) deployment in the industry.
5.5 Admission Control
Most networks require that a user/endpoint request permission to use the network before it actually starts using the network resources. This is known as admission control and can be broadly classified into two categories: authentication-based and resources-based.
Authentication-based admission control refers to a procedure in which the network ensures the authenticity of the user/endpoint before allowing it to access the network. This is usually done for the purposes of security and billing.
The 802.11 standard allows for authentication-based admission control where the network/AP may reject the admission request of a station if the station fails the authentication process. (The fact that this authentication scheme is weak and suffers from a lot of loopholes is another matter,)
Resource-based admission control has its roots in telephony networks like PSTN and is related to the concept of QoS. The underlying philosophy is that, if a network admits a user to the network, it must allocate/reserve resources for this user to ensure service to the user.
In the case of telephony, we do not want the destination phone to ring if we can't ensure that resources for the call will be available (this is known as a call defect, and has a more stringent failure rate requirement in toll-quality phone networks).
Since the network has limited resources, the network should keep track of what resources it is currently using and allow users to use the network only if there are enough resources available to service this user. Hence, admission control in this case is a matter of network capacity and usage.
Since the 802.11 standard has its roots in data communications, where the network does not reserve any resources for any users, it does not provide any resource-based admission-control procedures.
This, combined with the fact that the capacity of WLANs in terms of VoWLAN is extremely limited, means that congestion in VoWLAN networks is not handled gracefully. In most scenarios, a congested VoWLAN network severely degrades performance for all nodes in the system.
Security in 802.11 networks is a complex issue and has been a focus of a lot of attention. Our concern here of course is that while we might not care if someone snoops to see what web pages we are downloading, we do care if someone can snoop on our conversations.
5.7 Power Save
As discussed previously, the 802.11 power save mechanism required that the AP buffer packets destined for a dozing station and inform the station about buffered packets in the beacon. This means that packets destined from the AP to the station may be delayed for as long as the beacon period.
Since beacon periods are typically configured to be of the order of hundreds of milliseconds, using this power-save approach for real-time traffic would lead to delays of hundreds of milliseconds in the voice path.Arguably, smaller beacon periods would make this approach suitable even for low-delay applications like VoWLAN; but smaller beacon periods also means that when a Wi-Fi handset is not in use, it will have to wake up more often, thus leading to more power consumption and lower battery lives. Clearly, this is unacceptable. Hence, we conclude that the 802.11 doze mode is not suitable for real-time applications like VoWLAN.
5.8 Roaming/Handoffs in 802.11
The wireless medium is a harsh medium for signal propagation. Signals undergo a variety of alterations as they traverse the wireless medium. Some of these changes are due to the distance between the transmitter and the receiver, others are due to the physical environment of the propagation path, and yet others are due to the relative movement between the transmitter and the receiver.
Attenuation refers to the drop in signal strength as the signal propagates in any medium. All electromagnetic waves suffer from attenuation. For radio waves, if r is the distance of the receiver from the transmitter, the signal attenuation is typically modeled as 1/r2. It is important to emphasize that this is radio modeling we are talking about.
Such models are used for simulation and analysis. In real life, radio propagation is much harsher and the signal strength and quality at any given point depend on a lot of other factors too. Attenuation of signal strength predicts the average signal strength at a given distance from the transmitter::
Attenuation explains why all wireless transmissions have a limited geographical range. The signal strength at the receiver decays as the distance between the transmitter and receiver increases. This decay in signal strength means receivers far away from the transmitter are more prone to suffer from transmission errors. To operate reliably in the harsh wireless medium, the 802.11 MAC protocol requires each packet to be individually ACKed by the receiver.
If a packet is not ACKed, the transmitter must retransmit the packet. However, this protocol will work only up to a certain threshold distance, say R (range). If the distance between the transmitter and the receiver is increased beyond R, the received signal strength will be too low to achieve any communication. The value of R, therefore, depends upon three things:
1) Transmitted Power Level: Increasing transmission power will increase the range. However, 802.11 operates in the unlicensed frequency spectrum and most governments restrict the maximum transmission power level in this spectrum to less than 1 watt.
2) Wavelength/Frequency: Higher frequencies are more prone to attenuation than lower frequencies. Therefore 802.11b/g networks, which operate at 2.4 GHz, have a greater range than 802.11a networks, which operate at 5 GHz.
3) Antenna Gains: Since transmitted power level and operating frequency are restricted by laws and standards, respectively, many 802.11 equipment manufacturers concentrate on antenna gains to increase their range.
Most 802.11 equipment uses omnidirectional antennas since they are designed for general-purpose networking. However, it is possible to custom-design WLAN products per specifi c requirements to increase the range of 802.11 networks. Use of intelligent antennae and MIMO (multiple in multiple output) is also getting considerable attention in the industry for this purpose.
The bottom line, however, is that an 802.11 BSA has limited geographical range, typically a few hundred feet. Given that mobility is an inherent expectation in wireless networking, the question is how to provide seamless connectivity to a mobile user in 802.11 networks.
This is where roaming comes in. When a station moves out of the range of its AP and enters into the range of another AP, a handoff is said to have occurred. (The terms handoff and roaming are used interchangeably in 802.11 networks. )
It is important to realize that an 802.11 handoff is a Layer-2 process. An analogy might help make things clearer. Consider a laptop which is plugged into a corporate LAN socket using an Ethernet cable. This laptop has a valid IP address, which applications use to access the intranet and the Internet.
Now, if the Ethernet cable is plugged out from its current LAN socket and plugged into another socket belonging to the same corporate LAN, the applications on the laptop can continue to use its IP address.
In fact, the applications are not even aware of what has happened. The 802.11 handoff is a similar process where the plugging in and out of the Ethernet cable from sockets is analogous to disconnecting from an AP and connecting to another AP.
Therefore, if a higher-layer application uses a reliable transport-layer protocol like TCP, it does not even need to be aware of the 802.11 handoff—the transport layer will take care of retransmitting packets that were lost during the handoff.
A web browser running HTTP (HyperText Transfer Protocol) is such an example. Consider a mobile user running a web browser on his laptop which connects to the Internet via an 802.11 network. When the user clicks on an HTTP link while a handoff is in progress, the underlying transport protocol, TCP, conceals the delay/packet loss due to the handoff by using retransmissions. In other words, higher layers (HTTP, the web browser and the user) are unaware of the handoff and can continue without disruption.
On the other hand, if a higher-layer application uses an unreliable transport layer protocol like UDP, it would "see" the handoff as temporarily increased delay or packet loss. As long as the application can tolerate and recover from this delay or loss, an 802.11 handoff would be "transparent" to the user too.
The challenge for VoWLAN is that voice is extremely sensitive to delay. The end-to-end delay budget for voice is 250 ms; this means that the accumulative delay between the two endpoints involved in a voice call must not exceed 250 ms. This 250 ms must include the total transmission delay, propagation delay, processing delays in the network and and codec delays at both endpoints.
In WLANs, the 802.11 MAC introduces an extra transmission delay in the WLAN due to the contentious nature of the MAC protocol.( The situation is worse if both endpoints in the voice call use VoWLAN.)
This wireless transmission delay increases as networks become more congested or suffer from interference. The bottom line is that the budget for each component of the accumulative delay will typically be specifi ed by the service provider, but we can't assume that we would have all 250 ms available for 802.11 delays. A good VoWLAN implementation would aim to keep the 802.11 delays limited to 40"50 ms.
Given that VoWLAN operates on a very restricted delay budget, the
802.11 handoff times on the order of a few hundred milliseconds are
unacceptable for voice. (Typically
between 200 and 500 ms. )Thus, handoff times are another
important area of product differentiation and vendors are competing to
minimize handoff times in their products.
To read Part 1, go to
The many faces of VoWLAN
Praphul Chandra is currenty senior research scientist at HP Labs, India, which focuses on "technological innovation for emergining countries." David Lide is currently a senior member of the Technical Staff at Texas Instruments, Inc., and has worked on various aspects of Voice Over IP for the past eight years.