For network planners who are deploying a VoIP over WLAN application, one of the first issues to be addressed should be network capacity. To ensure the network is able to deliver the required QoS capabilities for a voice application, designers must anticipate and analyze how the WLAN will be used. Several questions, such as the following, must be answered:
- What types of QoS capabilities will be deployed?
- How much network capacity must be set aside for these QoS capabilities?
- What is the projected growth rate for QoS capabilities on the WLAN?
The purpose of this discussion is to explore the various facets of network capacity planning for the future deployment of WLANs. While the intent here is to analyze VoIP-enabled systems, network designers should also expect a significant amount of multimedia traffic over home/SOHO WLANs as well as video conferencing traffic over enterprise WLANs.
The remainder of this section describes the following:
- Over-subscription of voice networks (voice concentration)
- Throughput requirements for typical voice, video and media applications using IP packet technology
- WLAN network capacity for enterprise applications
- RF frequency planning and reuse for large network deployments
- WLAN network capacity for home applications
- Consideration of wireless repeaters (mesh) to extend home coverage
It is important for designers of VoIP over WLAN applications to understand some of the basic concepts that have been applied for years in the PSTN. A basic understanding of oversubscription, for example, can assist network planners who are evaluating network capacity for enterprise VoIP over WLAN applications.
Telephone systems have been very closely monitored for over 100 years. The public telephone system has always incorporated "statistical over-subscription" of phone lines. In the United States, there are typically between four and eight phones per active (served) phone line in the network. POTS (plain old telephone system) networks are designed to have a specific probability that a call can be blocked from time to time. In the United States, call blocking is typically limited to 1% or 0.5% of total calls.
Phone lines typically are terminated at a Class 5 switch or a Digital Loop Carrier (DLC) connected to a Class 5 switch. The Class 5 switch manages call connections and rejects calls when the system capacity has been reached. (A caller is aware of this when receiving a fast busy tone or the "all circuits busy" message.) In cellular networks, some consideration is given to reserve a fraction of the active phone line capacity for handoff purposes between one cell and the next.
A measure of phone usage capacity is the ERLANG function, which equates to one active call hour (or 3,600 call seconds) of voice line use. The amount of phone concentration (oversubscription) can be determined with the ERLANG-B function, (the ERLANG blocked call function). Because the telephone network must be designed for the worst-case load, phone usage is defined as that level that is achieved during the busiest hour of the day. Accurate average measurements for peak busy hour phone usage in the United States are as follows:
- 0.15 ERLANG (15 me) for a business phone
- 0.1 ERLANG (10 me) for a residential phone
Figure 2
As the pool of attached phone lines increases, efficiency, in terms of fewer blocked calls, and over-subscription also will increase. This should not be surprising because the efficiency of all systems that use statistical multiplexing improves as the number of channel resources increases at the multiplexer. The concentration level moves from around 2:1 at 10 subscriber phone lines to more than 3:1 at 60 user phone lines.
Voice, Video and Media Throughput over IP
The following sections discuss several WLAN network capacity issues as they relate to the transmission of voice, video and multimedia data using IP.
Voice Compression and VoIP
Voice compression algorithms help network designers derive as much capacity from an infrastructure as possible, but compression algorithms involve tradeoffs between efficiency and overhead that planners should consider.
In wireless networks, voice is digitized with the G.711 coding standard and transported at 64 Kbps. While G.711 is the mainstream digital codec for toll-quality voice services, a number of more efficient codes are used for both cellular and voice "pair gain applications." In an IP network, voice codecs are placed into packets with durations of 5, 10 or 20 msec of sampled voice, and these samples are encapsulated in a VoIP packet.
The following figure illustrates the encapsulation for various protocols, including IPv4, UDP and RTP (Real Time Protocol). For IPv4, the packet overhead is 40 Bytes. As the industry transitions to IPv6, this overhead will grow to 60 bytes.
Figure 3
Clearly VoIP has an overhead issue that is compounded when high levels of voice compression are deployed in conjunction with voice packets of short duration. The tradeoff between overhead and packet duration is shown in the following table. Other issues affecting VoIP network capacity planning, such as delay, jitter and packet duration, are discussed later in this tutorial.
| Voice Packet Frame duration (msec) efficiency | ||||
| CODEC | 5 | 10 | 20 | 40 |
| IPv4 G.711 | 47.6% | 64.5% | 78.4% | 87.9% |
| IPv6 G.711 | 38.5% | 55.6% | 71.4% | 83.3% |
| IPv4 G.726 | 31.3% | 47.6% | 64.5% | 78.4% |
| IPv6 G.726 | 23.8% | 38.5% | 55.6% | 71.4% |
| IPv4 G.729 | 10.2% | 18.5% | 31.3% | 47.6% |
| IPv6 G.729 | 7.2% | 13.5% | 23.8% | 38.5% |
Table 1
The following table lists the one-way throughput requirements for typical voice codecs using VoIP. For the purposes of capacity analysis, a typical throughput of 64 Kbps per direction was used, assuming a combination of G.726 and G.711 codecs.
| Coding algorithm | Bandwidth | Sample | Typical IP bandwidth (one way) | |
| G.711 | PCM | 64kbps | 0.125ms | 80kbps |
| G.723.1 | ACELP | 5.6kbps | 30ms | 16.27kbps |
| G.723.1 | ACELP | 6.4kbps | 30ms | 17.07kbps |
| G.726 | ADPCM | 32kbps | 0.125ms | 48kbps |
| G.728 | LD-CELP | 16kbps | 0.625ms | 32kbps |
| G.729(A) | CS-ACELP | 8kbps | 10ms | 24kbps |
Table 2
| VoIP Complexity Options To deploy VoIP WLANs, two tiers of voice-capable access points (APs) probably will be needed:
|
Video Media over IP
Although voice will be the first application requiring QoS capabilities over WLANs, several other multimedia applications will soon follow, including the distribution of audio (net radio, MP3 music, etc.) and video (streaming video, DVD, HDTV, etc.) over WLANs. Fortunately for network planners, media compression codecs will ease the bandwidth requirements for these multimedia applications. Specifically, improvements in the quality of video codecs like MPEG4 will allow DVD-quality compression at throughput rates of approximately one Mbps. For HDTV, the standard MPEG2 video stream can be reduced from 19.2 Mbps to around eight Mbps.
The following table illustrates the one-way throughput for various consumer video codec devices using maximum IP packet lengths.
| Video Media | Bandwidth (MBPS) | Packet Size | Packets/sec | Delivered Bandwidth (MBPS) | Overhead (%) |
| MPEG-1 Basic Media | 2.5 | 1,500 | 2083 | 2.5663 | 2.58% |
| MPEG-2 SDTV format (DVD) | 8 | 1,500 | 6666 | 8.2125 | 2.59% |
| HDTV MPEG 2 committee 18 | 19.2 | 1,500 | 16000 | 19.7120 | 2.60% |
Table 3
It should be noted that the FCC has mandated that by 2006 all televisions sold in the U.S. must include digital receivers. At that point, the integration of wireless interfaces into television electronics could be widespread.
Video Conferencing and IP Streaming Media
WLAN planners and designers should realize that video conferencing is an application that will have an impact on WLAN network capacity even though video conferencing has not yet become as pervasive in the enterprise or home markets as had been expected. This will change over the next few years as broadband connections become pervasive in households, the number of telecommuting workers increases, and enterprises improve their IT resources to allow greater use of video. As a result, WLAN designers should consider the requirements of video conferencing as they deploy infrastructure.
The following table provides a summary of the throughput requirements for several typical video conferencing and streaming media applications. Similarly to DVD and HDTV, the same types of improvements in lower resolution video and conferencing compression are expected in the years ahead. Network capacity models, especially in home networks, should anticipate increasing use of these applications.
| Video Product | Bandwidth (MBPS) | Packet Size | Packets/sec | Delivered Bandwidth (MBPS) | Overhead (%) |
| Business-Quality Conference | 915 | 781,693 | 107 | 735,466 | 6.3% |
| NetMeeting Video LAN | 779 | 478,312 | 77 | 445,156 | 7.4% |
| NetMeeting Video DSL | 363 | 187,726 | 64 | 159,800 | 17.4% |
| NetMeeting Video 28K | 288 | 10,497 | 5 | 8,529 | 23.1% |
| Read Audio Radio | 681 | 165,118 | 30 | 152,025 | 8.6% |
| Media Player 80K Stream | 687 | 81,171 | 15 | 74,882 | 8.4% |
| Media Player 20K Stream | 476 | 27,600 | 7 | 24,469 | 12.8% |
| Real video 28K Stream | 384 | 25,173 | 8 | 21,633 | 16.4% |
Table 4
Throughput of WLAN Access Points (AP)
To optimize the network capacity of a WLAN with a voice or multimedia application, network planners must give special attention to the throughput of the APs which govern how quickly data of any sort can be placed on the network.
The following two basic functions affect the throughput of an AP:
- Area and modulation density supported by the cell
- Small cells can support high data rate modulations (peak rates)
- Larger cells will use lower rate 802.11 modulations and are an aggregate sum of areas covered and the modulation rate
- The WLAN MAC protocols have the following effects:
- The Ethernet (CSMA/CA) protocols, DCF and EDCF, limit capacity at approximately 37% of the peak data rate
- Scheduled TDMA protocols such as HCF can theoretically reach around 90% capacity of the network, but under full load they will typically carry only approximately 75% of capacity
- DCF/EDCF MAC protocols do not effectively manage network latencies as the capacity limit is approached
- HCF protocols control latencies by providing fair weighted queuing so that all users will receive service even under full load conditions
| Throughput (MBPS) | ||
| Modulation | HCF (75%) | DCF/EDCF (37%) |
| 54 MBPS OFDM | 40.5 | 19.98 |
| 22 MBPS PBCC | 16.5 | 8.14 |
| 11 MBPS CCK | 8.25 | 4.07 |
| 5.5 MBPS CCK | 4.125 | 2.035 |
Table 5
By and large, network designers do not use theoretical peak performance rates when planning a WLAN. As a rule of thumb, most network planners de-rate the theoretical performance figures to approximately 70% to 80% of the peak capacity.
Note: With packet aggregation and proper use of 802.11 protection mechanisms, DCF/EDCF can achieve higher levels of throughput (approximately 50% to 55% higher) with a limited number of users and limited number of connections requiring QoS capabilities. This does not address the concern many enterprise WLAN designers have for the stability of DCF/EDCF under a high user load.
Enterprise Capacity Analysis
Because an enterprise 802.11 WLAN deployment will involve covering a workplace with a series of APs, the network planner must analyze the bandwidth capacity of each cell and the bandwidth demands that users will make on each cell in the network. In an enterprise deployment, the APs will be connected to a router either directly or through an Ethernet switch. In larger enterprises, multiple sub-nets may be connected hierarchically so that a wireless subscriber actually passes through several routers before reaching the IP network.
This type of WLAN essentially represents a micro-cellular architecture using 802.11 Aps interconnected via broadband IP links over Ethernet. APs have a certain coverage range which provides network access to users in a circular area around the location of the AP.
The analysis of enterprise network capacity that follows was based on the following assumptions:
- The average density of enterprise users is one per 200 square feet of floor space.
- The work day is eight hours long.
- 150 Mbytes of data as file downloads, e-mails and web accesses are transferred per user over the WLAN. No streaming media is supported.
- A sustained peak-to-average data throughput rate of three was used, essentially making the average data load three x 150 Mbytes or 450 Mbytes.
- Users require 0.15 ERLANG (15 me) of voice load. (This is based on current Bellcore and SBC business user peak busy hour loads.)
- A VoIP connection places a load on the WLAN of 64 Kbps in each direction (a combination G.726 and G.711).
| Cell Radius (feet) | 50 | 75 | 100 | 125 |
| Users | 39 | 88 | 157 | 245 |
| Active Phone Lines | 12 | 22 | 34 | 49 |
| Concentration X:1 | 3.25 | 4.00 | 4.62 | 5.00 |
| Bandwidth (MBPS) | ||||
| Voice Uplink | 0.77 | 1.41 | 2.18 | 2.18 |
| Voice Downlink | 0.77 | 1.41 | 2.18 | 2.18 |
| Data Downlink | 3.25 | 7.33 | 13.08 | 20.42 |
| Data Uplink | 1.63 | 3.67 | 6.54 | 10.21 |
| Total Throughput | 6.41 | 13.82 | 23.98 | 34.98 |
Table 6
This network capacity analysis shows that even for a small cell with a radius of just 50 feet, a typical 802.11b network would not have the capacity for applications like VoIP or the "completely unwired workplace." However, if an 802.11a/802.11g WLAN with 54 Mbps modulation were combined with an HCF MAC in a cell with a 100-ft. radius, the cell would have nearly 40 percent reserve (excess) bandwidth. Alternately, if the inefficient EDCF MAC were used, a dual-mode 802.11a/g solution would be required to cover the same cell. Two RF channels would be required if the EDCF MAC were used.
| "Wired When Docked" Workplace The analysis presented above is abased on the unrealistic assumption that users of a WLAN would always be completely wireless. In reality, a typical workplace will consist of wired and wirelss users, and most wireless users will be "docked," or connected to a wired network, when they are at their desks. Windows XP supports intelligent docking. Users are automatically switched from the WLAN network to a wired IP backbone when a device is docked. WLAN planners should take into consideration the effects that "wired when docked" will have on wireless networks' capacity requirements. For example, what if fewer than 20 percent of a workforce are un-tethered wireless workers. This has a profound impact on WLAN capacity needs, as shown in the following table.
Based on this analysis, planners can conclude that an enterprise WLAN with a "wired when docked" strategy can be supported by 802.11a/b/g dual-frequency access points using either HCF or EDCF. In other words, a deployment of a "sea of simple Ethernet-powered access points" would be sufficient. |
Table 7
In order to fully utilize the bandwidth of an access point, co-channel and adjacent channel interference must be addressed. The following section will briefly address RF planning.
RF Frequency Planning for Enterprise Deployment
To analyze properly the overall capacity of a WLAN deployment, planners must consider the effects co-channel and adjacent channel interference will have on the throughput and bandwidth of the APs in the infrastructure. As WLAN APs are deployed for wide area coverage, WLAN RF interference issues take on characteristics similar to those that are faced in the planning of micro-cellular RF networks.
RF network planning begins with a consideration of the frequencies that are available. 802.11 a/b/g radios have the following independent frequencies:
- 5.1 to 5.3 GHz with eight frequencies
- 2.4 GHz with three frequencies (There is some discussion in the industry that four frequencies actually could be used.)
Figure 4
For these types of deployments, the cell reuse distance, Ru, can be defined as follows:
- C = 7 (7 frequency): Ru = Rcell*sqrt (3C) = 4.48* Rcell
- C = 3 (3 frequency): Ru = Rcell *sqrt (3C) = 3.00* Rcell
Where:
C is the cluster size, which is the number of frequencies used in the reuse pattern
Ru is the reuse radius of the cell cluster
Rcell is the radius of coverage of a single cell
For distances greater than the AP's cell radius, it is assumed that RF propagation loss will not be free space (R2) but will be R3 to R4. This would result in interference reductions between cells of at least the following:
- C = 7: 19.5 dB to 26.1 dB (allows 36 to 54 Mbps OFDM)
- C = 3: 14.3 dB to 19.1 dB (allows 22 Mbps PBCC to 36 Mbps OFDM)
It is possible to use sectored access points and improve frequency reuse. However, in an enterprise environment this would require very careful placement of the APs and alignment of cell sectors. Where frequencies are at a premium, deployments based on four-frequency sectors per AP can provide optimal reuse. The interference reduction is equivalent to the omnidirectional seven-frequency plan previously discussed. The following diagram illustrates the optimal four-frequency reuse plan.
Figure 5
Home Capacity Analysis
Unlike the enterprise market where some assumptions can be made about typical usage patterns, network capacity analysis for WLANs in the residential market will be greatly influenced by the rate of market penetration and the implementation of multimedia applications.
The following are some probable multimedia applications for the home:
- 802.11 VoIP cordless phones and home PBX/voice mail integrated into an 802.11 access point
- Streaming audio distribution to 802.11 speaker systems
- Home PC as an MP3 audio service
- Streaming video from a cable television network, DVD system, etc.
- Telemetry applications, such as:
- 802.11-enabled cameras/video for security
- Meter reading for utilities
- Smart appliances
- Wireless print server connections
Considering that the FCC has mandated that all TVs sold in the US must have a digital tuner, there is a very strong possibility of some level of wireless video distribution in the home. Video applications certainly will have the largest effect on the throughput and capacity requirements of home WLANs. The following table lists the bandwidth requirements for a number of current video codecs:
| Video Media | Bandwidth (MBPS) | Packet Size | Packets/sec | Delivered Bandwidth (MBPS) | Overhead (%) |
| MPEG-1 Basic Media | 2.5 | 1,500 | 2083 | 2.5663 | 2.58% |
| MPEG-2 SDTV format (DVD) | 8 | 1,500 | 6666 | 8.2125 | 2.59% |
| HDTV MPEG 2 committee 18 | 19.2 | 1,500 | 16000 | 19.7120 | 2.60% |
Table 8
This data indicates that a single MPEG2 SDTV/DVD quality channel requiring eight Mbps of bandwidth cannot be supported by current 802.11b MAC/PHY components. Fortunately, advances in video compression (MPEG-4) should reduce the bandwidth requirements for video applications to approximately one Mbps for DVD-quality video and about eight Mbps for HDTVquality applications.
Over the next two to three years, many in the industry expect that a typical broadband-enabled household could have WLAN peak capacity needs as indicated in the table below:
| Service | Rate Upstream MBPS | Rate Downstream | Number of Channels | Total Rate Upstream | Total Rate Downstream |
| MPEG DVD-TV | 0.5 | 8 | 2 | 1 | 16 |
| Toll Quality Voice | 0.064 | 0.064 | 2 | 0.128 | 0.128 |
| Streaming Media | 0.01875 | 0.3 | 2 | 0.0375 | 0.6 |
| ABR Web Service | 0.0965 | 0.386 | 1 | 0.0965 | 0.386 |
| TOTAL | 1.262 | 17.114 | |||
Table 9
As shown in the following table, the market acceptance of HDTV and the absence of MPEG-4 compression could increase a home's WLAN throughput needs by a factor of four over the next four to five years.
| Service | Rate Upstream MBPS | Rate Downstream | Number of Channels | Total Rate Upstream | Total Rate Downstream |
| HD-TV | 1.5625 | 25 | 2 | 3.125 | 50 |
| Toll Quality Voice | 0.064 | 0.064 | 4 | 0.256 | 0.256 |
| Streaming Media | 0.01875 | 0.3 | 1 | 0.01875 | 0.3 |
| ABR Web Service | 0.0965 | 0.386 | 2 | 0.193 | 0.772 |
| TOTAL | 3.59275 | 51.328 | |||
Table 10
These numbers indicate that high-throughput 802.11g/a PHY technology will be needed as a minimum in order to support these applications. Further, an efficient MAC (HCF) will be needed to optimize throughput.
VoIP applications do not require a significant amount of bandwidth in any of these capacity scenarios. Given the small number of phones in a typical home, the system must be designed for 1:1 concentration (that is, there would be no over-subscription of phone lines in the home). The more important benefits of WLAN-enabled cordless phones are twofold:
- Removing cordless phones as a source of RF interference in the 2.4 GHz and 5.2-5.8 GHz frequencies could accelerate the acceptance of video applications over WLANs.
- A new market for 802.11 cordless phones would be created with a sales potential of approximately 100 million units a year.
| Residential 802.11 Link Asymmetry Usage models of residential applications show that the typical data transfer load is very asymmetric. That is, the downlink from the AP to the subscriber usually requires 10 times more throughput than the up-link from the subscriber to the AP. Radio archetectures for 802.11 APs can be differentiated to improve coverage and throughput by simply "bolting on" a booster LNA and PA capability (much like an "afterburner"). This capability is most appropriate in North America where spectrum rules allow 10 dB greater EIRP (power) than in Europe. The range of an AP can be nearly doubled at the highest modulation rate with simple link budget improvemetns in the AP. |
Application of Repeaters/Small-mesh Access Points for Residential/SOHO Coverage
For developers of WLAN access points for the residential/SOHO marketplace, cell coverage and throughput are the most crucial issues facing WLAN implementations in this market. Wireless repeaters, which can be used to implement small mesh residential networks, are a low-cost method of improving coverage and throughput.
One possible technique for extending coverage and improving residential service is the use of multiple APs in a mesh/repeater architecture. A simple example featuring two access points is illustrated in the diagram below:
Figure 6
Access point B is a repeater (mesh element) to access point A, which connects to the Internet. Access point A functions as a router to access point B. Access point A must maintain a routing list for all clients in the home network while access point B only must maintain a routing list of attached clients. For example, B may be a simple bridge or a more intelligent router. Clearly, the mobility/roaming between the cells in this sort of arrangement will generate overhead messaging to update and maintain the routing information.
A real-world example of mesh WLAN architecture was the Aironet system, which was one of the first large-scale deployment platforms for WLANs. In this system, a client would probe for APs that could provide coverage, and the APs would reply with information on signal quality and on how much of their resources were currently in use. The subscriber would then associate and authenticate with the AP with the best signal quality and lowest usage. Once this was completed, re-routing updates would be completed.
Mesh networks can be nested deeper than a single connection. This is known as multi-hop. However, this creates even greater delays because of the cumulative time needed to route and retransmit from one AP to another. For voice, video phone and video conferencing, the round trip delay would be excessive for any architecture with more than one hop.
There are two possibilities for operating a residential mesh network. They are the following:
- Single-Frequency Mode: Access points are not dual-mode and can only support a single frequency of operation from an AP to another AP and from an AP to a client/subscriber.
- Dual-Frequency Mode: Access points are dual frequency, supporting two separate links on two separate frequencies simultaneously.
The dual-frequency mode requires that all access points support two frequencies simultaneously. Typically, 802.11a (5.x GHz) would be used for AP-to-AP backbone communications while AP-to-subscriber communication would be provided by 802.11b/g (2.4 GHz). Using the 2.4 GHz frequency for subscriber coverage ensures support for low-cost and legacy 802.11b clients/subscribers. Because three independent 802.11b/g frequencies are available in the 2.4 GHz band, WLANs designed with a primary AP and one or two repeater Aps actually improve the coverage of the home. Stated another way, as long as three APs are implemented, the coverage area is greater and throughput will be consistently high without RF interference between the APs.
The dual frequency configuration is shown in the following diagram:
Figure 7
The IEEE community is debating whether to use MIMO and/or beam steering techniques for 802.11 standards as a way to improve throughput and coverage.
A simople mesh extension for the 802.11g standard combined with improved video compression could be available to consumers immediately, and this would provide a "virtual" performance improvement. Final approval of IEEE 802.11 HTSG is at least three years away.
The mesh repeater architechture has another benefit in that it improves signal-to-interference (S/I) performanace because the architecture ensures subscribers are consistently closer to access points. This, in turn, ensures better link margins.
