Professional Documents
Culture Documents
www.arubanetworks.com
Aruba Networks
Table of Contents
Chapter 1 Preparing for the 802.11n adoption wave
Introduction
Benefits for the Enterprise Benefits for the user These benefits will be elusive for a while 802.11n migration strategies for enterprises Network design with 802.11n Wired backhaul from APs & LAN implications Implications for rogue APs and WIDS Good-neighbour (or bad-neighbour) strategies
3
3
4 5 5 7 8 9 10 10
12
12
13 14 16 21 22 24 24 25 26 27 27 31
31
31 32 33 33 34 35 36 38
www.arubanetworks.com
Designed for Speed - 802.11n Theory and Practice Other mechanisms for coexistence: Dual-CTS protection (CTS-to-self) Other mechanisms for coexistence: 40 MHz-intolerant indication Using 802.11n in the 2.4 GHz band Using 802.11n in the 5 GHz band 39 40 40 42
43
43
45
47
47
48 49 50 51 52 53
53
55 56
58
58 59 59 60 61
Forms of MIMO
Channel estimation
66
67
68
References
2
69
Aruba Networks
Introduction
Wi-Fi technology has carved a path of ever-increasing performance from the earliest pre-802.11 standards through 802.11b to 802.11a/g, with peak data rates rising from 2Mbps to 54Mbps. The latest set of innovations is a package known as 802.11n. As 802.11n is a very new technology, this paper should be considered a snapshot of progress in late 2007, rather than a definitive guide. We refer in this paper to two key documents that shape the industry, and will endeavour to maintain consistency in using these terms: 802.11n-Draft2.0 is a technical standard developed by the IEEE 802.11 working group in early 2007. This is the base technical standard for all 802.11n products. However, 802.11n is not a finished work, and the IEEE continues to develop, refine and (in some cases) correct it.
www.arubanetworks.com
Designed for Speed - 802.11n Theory and Practice Meanwhile Draft-n is an interoperability certification developed by the Wi-Fi Alliance, a trade association of companies interested in promoting 802.11 products (Wi-Fi is a Wi-Fi Alliance brand). Draft-n is a certification awarded by the Wi-Fi Alliance to indicate a product has passed a set of tests that ensure it will inter-operate with other Draft-n products. For this certification, the Wi-Fi Alliance took parts of 802.11n-Draft2.0 and developed a series of tests involving a testbed of early Draft-n equipment: thus the certification tests only a subset of the full 802.11n-Draft2.0 functionality. (It is important to distinguish Draft-n, a certification, from pre-n which is a label used by many equipment vendors before the certification was available. Many of these products do indeed interoperate, but they are not certified by any independent organization.)
Chapter 1 - Preparing for the 802.11n adoption wave reducing performance. One contributor to this issue is multipath propagation, and the best technology to counter this to date has been antenna diversity nearly every Wi-Fi device sports two antennas, and switches between them so when one is in a multipath null, the other should still have a workable signal. The MIMO technology in 802.11n is extremely effective in reducing the effect of multipath nulls. Lower network costs. This is a consequence of improved range and more uniform coverage: APs can be spaced further apart. This reduces costs in a number of ways: fewer APs, lower installation costs, possibly fewer LAN edge switch ports, and fewer outdoor APs to cover campus areas between buildings. Conversely, 802.11n APs will initially be more expensive than existing equipment, although we expect this premium will erode over time. Also, where networks, or areas within networks are designed to capacity limits (the bandwidth available per user) rather than range, it may not be possible to increase AP spacing.
www.arubanetworks.com
Designed for Speed - 802.11n Theory and Practice Most enterprises upgrade PCs (the most common Wi-Fi client) over a cycle of 2- years or longer. As long as older clients exist, they will affect network performance by connecting at much lower rates than 802.11n clients, effectively slowing cells to near-802.11a/g rates. Most WLAN vendors try to escape this limitation by implementing some kind of per-user bandwidth control, but that is only a partial solution. As long as even a few legacy clients exist, the expected capacity improvements are unlikely to be realized. Legacy devices are also unable to take advantage of the improvements in range and uniformity of coverage offered by 802.11n. If 802.11n APs are spaced farther apart to lower costs, legacy clients will likely run into more coverage problems than before. Moving to 802.11n will drive equipment upgrades in other parts of the network. 802.11n requires new access point hardware incorporating new silicon, while Gigabit Ethernet and powerover-Ethernet upgrades will often be required for wiring and edge switches. The WLAN controllers that are used by state-of-the-art centralised architectures should be considered vendor-by-vendor. For a controller to become 802.11n-capable, it must support the new radio and other features of the access point from a technology perspective, which is not likely to present a problem. However, the increased data rates generated by an 802.11n WLAN may challenge some current controllers using general-purpose computing platforms: traffic analysis will allow the network designer to develop an accurate estimate of WLAN controller capacity requirements. There is already a wave of pre-standard implementations. The WiFi industry is technology-driven and consumer-led. Consumers are inherently less sensitive to standards than enterprises, since a typical consumer WLAN might include a handful of devices working together where compatibility is easily addressed by simply using a single vendors equipment. Thus there are already first and second waves of pre-n silicon and consumer products on the markets, and major vendors are now building Draft-n support into mass-market PCs. This is helps move the new technology into the field, but it is pre-empting the standards process, which is not yet complete. There is no guarantee that Draft-n products shipped today will be fully-compatible with the eventual standard. (There is no concern that such products from different vendors will not work together, but they may not prove capable of realizing the full benefits listed above.)
Aruba Networks
Chapter 1 - Preparing for the 802.11n adoption wave Thus, while the eventual benefits are easy to see, it may prove more difficult to get there from here. These obstacles can be overcome by carefully considering the plan to migrate from current WLAN installations to a new 802.11n infrastructure.
Designed for Speed - 802.11n Theory and Practice These techniques for designing high-performance 802.11 enterprise WLANs have been refined by vendors over a number of years. While the overall network efficiency and performance will fall short of a greenfield uniform 802.11n network, such a mid-life upgrade can provide the desired performance improvements for another 12-24 months, by which time the technology risks of moving to 802.11n will have been allayed.
Aruba Networks
Chapter 1 - Preparing for the 802.11n adoption wave the antennas by at least half a wavelength (6.25cm for 2.4GHz or 2.7cm for 5.5GHz) will give better performance.
Other areas of the LAN should be checked for traffic capacity but will probably not require augmentation. This is because the traffic on the upstream connection from the edge switch still represents the same aggregate number of users and applications as when clients were all wired. The power consumption of 802.11n APs is greater than for 802.11a/g APs, exceeding 802.af limits (12.5W maximum can be delivered to a Class device under 802.11af, now properly termed clause of updated 802.-2005 ). This means the edge switches or in-line power injectors must support the unratified 802.at standard. Alternatively, the AP must use a local power brick.
www.arubanetworks.com
10
Aruba Networks
Chapter 1 - Preparing for the 802.11n adoption wave are prefixed by a preamble that uses legacy modulation, allowing such networks to coexist. Another optional mechanism, Phased Coexistence Operation (PCO) allows an AP to alternate between a 40 MHz channel and each of the constituent 20 MHz channels. The AP sets the Network Availability Vector (NAV) on the 20 MHz channels to inhibit transmissions for a time, during which it and whatever 40 MHzcapable clients it supports switch to 40 MHz operation.
www.arubanetworks.com
11
Chapter 2 - Technology in 802.11n Legacy 802.11 equipment only drives one antenna at a time, and only receives on one antenna at a time, usually the same antenna. 802.11nDraft2.0 assumes that multiple antennas are used for either or both the transmit and receive functions. There are four distinct algorithms at work: Maximum ratio combining is a receiver function, where signals received on multiple antennas, whether from one or a number of transmit antennas, can be combined to improve the signal-to-noise ratio. MRC is an antenna diversity technique that can increase range for a given data rate. Space-time block coding can be used where there are multiple transmit antennas, regardless of the number of receive antennas. STBC is an antenna-diversity technique that improves the signalto-noise ratio at the receiver, also allowing an increased range for a given data rate. Spatial division multiplexing is the technique most often associated with MIMO. Rather than increasing range, SDM sends different spatial streams of data from each transmit antenna. Since these streams carry different data, the overall data rate of the system is increased. Under good conditions, a MIMO system of two transmit and two receive antennas doubles the achievable data rate over a single-antenna (SISO) system. Transmit beamforming is a technique where signals sent to multiple transmit antennas can be phased such that the RF power at a targeted receive antenna is maximized. TxBF is discussed here even though the technique is not included in 802.11n-Draft2.0.
www.arubanetworks.com
1
12
2 A B 1
9 S N R (dB) 6 3 0 -3 -6 SNR 1
Combined SNR
SNR 2
time
In the diagram above, each receive antenna would receive an identical signal under line-of-sight conditions. However, due to noise and multipath effects, one or both of the antennas will often receive a suboptimal signal. MRC allows the receiver to process both received signals independently, then combine them, weighted by the signal strength of each signal, to extract a more accurate replica of the transmitted datastream than it could from a single antenna. In simple terms, when one antenna is in a null and has a bad signal, the other is likely to have a good signal. The receiver will only suffer when all antennas have bad signals simultaneously. As the number of antennas increases, the probability of bad conditions for all of them simultaneously becomes increasingly smaller. In order for MRC (and MIMO techniques) to be effective, the receive antennas must receive different versions (distorted by noise and interference) of the original transmitted signal. Accomplishing this goal usually means separating the antennas by at least half a wavelength, in the order of cm for a 5GHz signal. MRC is not covered in 802.11n-Draft2.0 because it can be implemented at the receiver only, with no changes at the transmit end. However, most chip designers are expected to implement a form of receive diversity such as MRC. For MRC to work well, the RF channel must be characterized. In simple terms, this involves sending a known sequence of symbols from the available transmit antennas to calibrate the system. Because the receiver knows the symbols that were sent, it can determine the type of distortion introduced by the RF channel. 802.11n-Draft2.0 provides for these known sequences in the long training fields (LTF).
14
Aruba Networks
Chapter 2 - Technology in 802.11n of receive antennas. STBC uses coding to transmit different (but known) copies of the data-stream from different antennas; assuming the receiver knows the code, it will be able to extract the original data with fewer errors than when a single transmit antenna is used. One form of space time coding, the Alamouti code to spread one spatial stream over two space time streams, takes a pair of data-stream bits, and performs operations on them in consecutive time intervals. This is the basic STBC method in 802.11n-Draft2.0.
2 A B
Spatial stream Tx Tx Space time streams Rx STBC decoding MAC etc
1
Spatial stream
MAC etc
STBC coding
802.11n coding example: for two consecutive symbols, s1 and s2: - In time interval t1, transmit s1 from antenna A and s2 from antenna B - In time interval t2, transmit s2* from antenna A and s1* from antenna B
When processing this sequence of two symbols from two space-time streams, the receiver is able to re-constitute the original data-stream even in the presence of channel noise and distortion. STBC uses the time dimension: consecutive symbol intervals contain the same pair of basic symbols, but modified according to the code. A good choice of code such as Alamouti allows minimum complexity for the transmitter and receiver, but maximum improvement in SNR under normal channel impairments. 802.11n-Draft2.0 defines STBCs to split a spatial stream into as many as 4 space-time streams. This technique can be used when the number of transmit antennas exceeds the number of receive antennas: it can also be used in conjunction with MRC. STBC requires both channel characterization and shared knowledge of the STBC code in use.
www.arubanetworks.com
15
MAC etc
Signal processing
Tx Tx Rx
Rx
Signal processing
MAC etc
Each antenna is connected to its own RF chain for transmit and receive. The baseband processing on the transmit side can synthesize different signals to send to each antenna, while at the receiver the signals from different antennas can be decoded individually. (We will simplify this explanation by showing only one direction of transmission, practical systems will transmit in both directions.) Under normal, line of sight conditions, the receiving antennas all hear the same signal from the transmitter. Even if the receiver uses sophisticated techniques to separate the signals heard at antennas 1 and 2, it is left with the same data. If the transmitter attempts to send different signals to antennas A and B, those signals will arrive simultaneously at the receiver, and will effectively interfere with each other. There is no way under these conditions to better the performance of a non-MIMO system: one might as well use only one antenna at each station.
16
Aruba Networks
2 A B 1
(Nearly all 802.11 stations built to date actually use two antennas. However, this is not MIMO the radio switches a single radio chip from one antenna to the other, so only one is used at any time. Using two antennas in this way helps to negate the effects of multipath, as when one antenna is in a multipath null, the other is likely to have a better signal. It is generally reckoned that using antenna diversity in this way improves overall reception by perhaps -6 dB, although the effect is of course statistical. MIMO is different in that both antennas are driven by or receiving signals at all times, and those transmitted signals need not be identical.)
2 A B 1
However, if there is sufficient RF distortion in the path, receiving antennas will see different signals from each transmit antenna. The transmit antenna radiates a signal over a broad arc, and it scatters and reflects off various objects in the surrounding area. Each reflection entails a loss of signal power and a phase shift, and the longer the reflected path, the more delay is introduced relative to a line-of-sight signal. In the past, multipath has been the enemy of radio systems, as the receiver sees a dominant signal (line of sight if it is present), and all the multipath signals tend to interfere with it, effectively acting as noise or interference and reducing the overall throughput of the system. Multipath effects also change over time, as objects in the path move, and movement of reflecting objects results in a Doppler shift of the frequency of the received signal, further complicating the control mechanisms needed to counter multipath.
www.arubanetworks.com
17
Designed for Speed - 802.11n Theory and Practice To understand how MIMO works, first consider the signal each receive antenna sees in a multipath environment.
b a A B
c 2
a b c
Dominant signal at Rx antenna 1 is from path a. Path b, c and other multipath signals cause some degradation to the dominant signal a similar effect to higher background noise.
time
In this example there are multipath signals arriving at antenna 2. The strongest is signal a, and the information carried in this signal will be decoded. Other signals arrive at lower power levels, and they are timeshifted (or phase-shifted) compared to a, so it is likely they will degrade the overall signal-to-noise ratio associated with a.
18
Aruba Networks
A2 A B A1 B1
B2 2
A1 B1 B2 A2
time
Dominant signal at Rx antenna 1 is from Tx antenna A (B1 signal and other multipath signals cause some degradation) Dominant signal at Rx antenna 2 is from Tx antenna B (A2 signal and other multipath signals cause some degradation)
When multiple antennas are considered, however, MIMO offers considerable gains in throughput. The example above shows that each receive antenna receives its dominant signal from a different transmit antenna: receiver 1 uses transmitter A while receiver 2 uses transmitter B. When the system understands this, it can take advantage by transmitting different signals from each antenna, knowing each will be received with little interference from the other. Herein lies the genius of MIMO. (In fact the technique is more sophisticated, using RF channel characterization as explained earlier in this paper: it is not necessarily the case that individual signal paths can be drawn between pairs of transmit-receive antennas, but given that the nature of the cross www.arubanetworks.com 1
Designed for Speed - 802.11n Theory and Practice coupling is known at the receiver, and that mathematical conditions for the channel are favourable, this is the overall effect.)
V11 U11
V21 V12
U21 U12
V22
U22
The diagram above shows a more detailed explanation of MIMO implementation. At the transmit side, signal processing provides real and imaginary outputs S1in and S2in. These are then mixed with different weights V11 etc, before the signals are combined and delivered to the transmit antennas. A similar mixing function processes signals from the receive antennas using weights U11 etc. Provided the RF characteristics are known, the weights U11 can be calculated and set for optimum throughput, given the RF channel conditions. The most favourable case would be where each transmit-receive pair operates with a completely independent RF path: a 2x2 (2 antennas at each station) system will have double the throughput of a singleantenna 1x1 system, and a x configuration could extend to triple the throughput. 802.11n defines MIMO configurations from 2x1 to 4x4 antennas and up to spatial streams. MIMO is the most difficult aspect of 802.11n to understand: multipath (reflected RF between transmitter and receiver) is normally the enemy of performance, but with MIMO it can be used constructively. Line of sight normally gives the best performance, but with MIMO it provides just baseline data rates. (Note, however, that reflected signals are usually much weaker than primary, line-of-sight signals. Even though losing lineof-sight may allow use of more RF paths and hence the additive MIMO effect, the signal-to-noise ratio of each path may be considerably worse than previously. It is difficult to predict the relative weight of these two opposing effects.) As noted above, MIMO works best when antennas are positioned more than half a wavelength apart. For 5.5 GHz, half a wavelength is about 2.7 cm or one inch. The 802.11n-Draft2.0 standard mandates at least two spatial streams (antennas) for access points and one spatial stream for client devices, with a maximum of 4 spatial streams per device. 20 Aruba Networks
Chapter 2 - Technology in 802.11n One key question in MIMO systems is how to tune the transmit signals at different antennas for optimum reception at the receiver. 802.11nDraft2.0 offers different methods for this. With implicit feedback the MIMO transmitter characterizes signals from the receiver, and assumes that the channel is reciprocal reflections and impairments operate equally in both directions. This is a reasonable approximation for most purposes, but better performance is achieved when the receiver sends explicit feedback messages to the transmitter; with these, the transmitter can accurately tune its signals for optimum reception and best signal to interference and noise ratios at the receiver.
A C
Normal isotropic radiation pattern
Transmitted power level
B
Directional pattern after beamforming
B C
tca tcb
time
B C
time
Identical RF signals are transmitted from each antenna, but very slightly offset in time (phase)
Transmit timing offsets are calculated to match the path delays of the different RF beams, so all signals are directed at the target in phase -
This results in constructive interference, and a higher signal-to-noise ratio at the receive antenna
By carefully controlling the time (or phase) of the signal transmitted from multiple antennas, it is possible to shape the overall pattern of the received signal, emulating a higher-gain, or directional antenna in the direction of the target. The same implicit and explicit feedback mechanisms used to characterize the MIMO channel allow beamforming. www.arubanetworks.com 21
Designed for Speed - 802.11n Theory and Practice In practice, beamforming may be used when MIMO with SDM is not effective. This is because beamforming aims to produce a single, coherent RF signal at the receiver, while SDM relies on multiple, independent signals. Beamforming will only work well with 802.11n clients, as it works best with explicit feedback messages from the client. This highlights a difference from some other adaptive antenna or beam steering technologies: in 802.11n, the beam is not based on an indication of the direction of a client, but rather on the actual RF conditions at its antennas; hence the requirement for explicit feedback messages.
22
Aruba Networks
2 A C B
P1b
P1a P1 P1b
SDM combined with antenna diversity. In this example, the path between the A-3 antenna pair has different RF characteristics from the other antenna pairs: it offers RF diversity, and carries one spatial stream. The other inter-antenna paths, B/C to 1/2, are not RFisolated, so they cannot only be used for one further spatial stream. However, transmit beamforming or receive antenna diversity may be used to optimise this spatial stream. In this case a system with 3 transmit and 3 receive antennas nevertheless supports only 2 spatial streams.
P1a P1a P1 P1b
2
P1a P1 P1b
A B
P1b
MIMO with Spatial Diversity Multiplexing (SDM) and Space Time Block Coding (STBC). Independent paths between pairs of antennas allow data transmission in parallel: data packets (P1) are interleaved and mapped to different paths, where they may be encoded at a different data rate for each path, depending on RF conditions. The receiver interleaver re-builds the original packet.
P1 P1
2
P1
A B
P1
Transmit beamforming. The transmitter sends a single stream of data, adjusting the signal from each antenna to ensure the optimal signal forms at the receiving antenna. This is used where there is little RF separation between the different inter antenna paths, so SDM is not useful.
2
P1
A B
P1
P1
Receive antenna spatial diversity. Working on only one transmitted signal, the receiver can use RF combining techniques on signals from different receive antennas to achieve higher signal-to-noise ratios and higher data rates.
While SDM is a multiplexing technique to increase overall data rate, STBC and MRC are diversity techniques that improve the signal to noise and interference ratio, SNR or SINR. These techniques can be combined under some conditions.
www.arubanetworks.com
2
MAC etc
RF channel
MAC etc
The first decision is how many spatial streams to use. A spatial stream in this context carries data that is independent of other spatial streams: spatial streams are combined by SDM techniques. The number of spatial streams can be no greater than the smaller of the number of transmit or receive antenna chains in the system, where the system covers a transmitting and a receiving unit. Most current 802.11nDraft2.0 chipsets (Draft-n) can process two spatial streams. After the number of spatial streams is defined, there may be excess antennas at either the receiver or transmitter. For instance, most current Draft-n access points sport three driven antennas, while PC clients have two. Therefore, in the direction of transmission towards the 24 Aruba Networks
Chapter 2 - Technology in 802.11n client, the access point has an excess transmitter. This transmitter can be used to provide better performance by implementing STBC. The STBC encoder takes the two spatial streams from the SDM block, and expands them to form three signals to drive the transmit antennas. At the receiver, antenna chains are programmed for the STBC used at the transmitter, and may use whatever MRC capabilities the designer implements. In this case one would not expect any gains from MRC, as there are (to approximate to a simple model) two antennas receiving two spatial streams. The combined system gains bandwidth from SDM, and some SNR gains from STBC: in practice, SNR may translate to a higher data-rate for a given range, or a longer range for a given data-rate, its the implementers choice. The diagram below shows the reverse link, where the transmitter has two antenna chains and the receiver three. Now, it is not possible to use an STBC gain, as there are no excess transmit antennas, but the MRC processing the receiver realizes from its excess antenna will provide approximately the same gain.
Data flow in this direction
MAC etc
RF channel
MAC etc
www.arubanetworks.com
25
Designed for Speed - 802.11n Theory and Practice be expected, this offers approximately twice the throughput of a 20 MHz channel. However, while in the 5 GHz band the channels are defined as pairs of existing 20 MHz channels, they do not line up with commonlyused 20 MHz channels in the 2.4 GHz band, as these channels are not adjacent. This means that when a 40 MHz channel is used in 2.4 GHz, it will interfere with at least one other 802.11b/g channel.
a b c
N N N
N+1
802.11n (also 802.11a/g) transmission is by OFDM symbols (examples N, N+1, N+2). Multipath increases delay spread at the receiver; the guard interval prevents inter-symbol interference. In this example, path b is within the guard interval while c causes inter-symbol interference.
time
Previous 802.11 standards used a guard interval of 800nsec. 802.11n adds an option for 400nsec, negotiated between receiver and transmitter, for cases where the worstcase multipath delay is low. (propagation in free-space, delay = distance x 0.3 metres/nsec, so 400nsec is equivalent to 120metres path difference.)
For best (least-error) decoding, the symbol must arrive at the receiver without any interference or noise. Previous sections of this document have shown how 802.11n-Draft2.0 uses MIMO to improve reception of multipath, but this only works symbol-by-symbol. Inter-symbol interference occurs when the delay between different RF paths to the receiver exceeds the guard interval, causing a reflection of the previous symbol to interfere with the strong signal from the current symbol: a form of self-interference.
26
Aruba Networks
Chapter 2 - Technology in 802.11n The optional 400nsec short guard interval in 802.11n can be used when the path difference between the fastest and slowest RF paths is less than that limit. The diagram includes a quick path loss calculation, but in reality, indoor multipath reflections can reach 400nsec relatively easily. This means the shorter guard interval will be very useful in consumer settings, but cannot be relied upon when planning for enterprise deployments.
26 carriers
26 carriers
28 carriers
28 carriers
-10MHz
fc
+10MHz
-10MHz
fc
+10MHz
52 subcarriers (48 usable) for a 20 MHz non-HT mode (legacy 802.11a/g) channel
57 carriers
57 carriers
-20MHz
-10MHz
fc
+10MHz
+20MHz
The additional subcarriers effectively add bandwidth to the channel, allowing increased data rates for a given modulation type (see the section below on new modulation rates).
Designed for Speed - 802.11n Theory and Practice While 802.11a and g specify 8 rates (6, , 12, 18, 24, 6, 48 and 54 Mbps), 802.11n provides many more: over 00. However, the basic set is of 8 rates: Basic rates of 802.11n (Mbps) 20 MHz channel; single stream; 800 nsec GI; equal modulation 26 52 58.5 1 1.5
6.5
65
This is a set of rates for one spatial stream in a 20 MHz RF channel and with an 800 nsec guard-interval and equal modulation on all spatial streams. This basic set of rates is comparable to the 802.11b/g rates above: each rate is improved by about 8% (e.g. 18 to 1.5 Mbps) by using slightly wider bandwidth, more subcarriers and improved modulation. The 65 Mbps rate has no equivalent in 802.11a/g: it uses 5/6 coding, a higher rate than the previous maximum of /4. Other rates are generally derived as multiples of the basic rates above: Basic rates of 802.11n (Mbps) 20 MHz channel; single stream; 800 nsec GI; equal modulation 26 52 78 104 117 Rates of 802.11n (Mbps) 40 MHz channel; single stream; 800 nsec GI; equal 27 40.5 54 81 108 121.5
1
10
1.5
15
The 40 MHz channel allows slightly more than twice the data rate of two 20MHz channels. Rates of 802.11n (Mbps) 20 MHz channel; single stream; 400 nsec GI; equal modulation 14.4 21.7 28. 4. 57.8 65
7.2
72.2
The shorter 400nsec guard interval allows slightly higher data rates than 800nsec. Rates of 802.11n (Mbps) 20 MHz channel; two streams; 800 nsec GI; unequal modulation 52 65 58.5 78
7.5
28
Aruba Networks
For a given system, the range of choices will be smaller than the tables above or below would indicate, because some of these factors are fixed for a given system: 20 MHz or 40 MHz channel. As discussed elsewhere in this paper, a 40 MHz channel will not often be feasible for an enterprise deployment in the 2.4 GHz band. However 40 MHz channels are likely to be widespread in the 5 GHz band. Spatial streams. As described above, it will be quite usual for the number of supported spatial streams to be smaller than the number of antenna chains either because of silicon processing capabilities or because the client has fewer antenna chains than the AP. For instance, client state of the art x systems in late 2007 only support 2 spatial streams. Also, where there is insufficient RF path isolation between streams, even a 2x2 system may not be able to support 2 diverse streams. Guard interval. The guard interval is the time between OFDM symbols in the air. Normally it will be 800 nsec: the option is for a 400 nsec guard interval, but as noted above, reliable detection with this value may be challenging under indoor conditions. Convolutional coding. When data arrives at the PHY layer for transmission, it is scrambled and coded. This alters its spectral characteristics in order to achieve the best signal-to-noise ratio, and also includes built-in error correction, known as convolutional coding. The 802.11n standard includes BCC (block convolutional coding), as included in previous 802.11 standards, but also adds an option for LDPC (low density parity check) coding, which can improve effective throughput for given RF conditions. Modulation. All spatial streams may use the same (equal) modulation, or they may carry different (unequal) modulation and coding. An example might be where there are three spatial streams with good MIMO characteristics, but one stream has a high noise floor or a low signal level: under these conditions the weak stream would support a lower data rate than the other streams.
www.arubanetworks.com
2
YES YES YES YES YES YES YES YES YES YES YES YES YES YES YES YES NO NO NO NO NO NO NO NO NO NO NO NO
20 20 20 20 20 20 20 20 40 40 40 40 40 40 40 40 20 20 20 20 20 20 40 40 40 40 40 40
800 400 800 400 800 400 800 400 800 400 800 400 800 400 800 400 800 400 800 400 800 400 800 400 800 400 800 400
1 1 2 2 4 4 1 1 2 2 4 4 2 2 4 4 2 2 4 4
6.5 7.2 1 14.4 1.5 21.7 26 28. 1.5 15 27 0 40.5 45 54 60 4. 52 57.8 65 72.2 81 0 108 120 15 150
65 72.2 10 144.4 15 216.7 260 288. 15 150 270 00 405 450 540 600 7.5 108. 156 17. 214.5 28. 202.5 225 24 60 445.5 45
8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 6 6 14 14 24 24 6 6 14 14 24 24
0
Aruba Networks
52 carriers
52 carriers
-20MHz
-10MHz
fc
+10MHz
+20MHz
52 x 2 subcarriers (96 usable) for a 40 MHz duplicate format (802.11n) channel (applicable to greenfield and mixed-modes)
www.arubanetworks.com
1
P1
P2
P3
P1
P2
P3
P1
P2
P3
MAC processing
MAC processing
MAC processing
MAC header
P1
P2
P3
MAC header
P1
MAC header
P2
MAC header
P3
PHY layer
In the A-MSDU format, multiple frames from higher layers are combined and processed by the MAC layer as a single entity. Each original frame becomes a subframe within the aggregated MAC frame, with its own sub-header containing source & destination addresses and length. Thus this method can be used for frames with differing source and destination addresses, but only MSDUs of the same priority (access class, as in 802.11e) can be aggregated. An alternative method, A-MPDU format, allows concatenation of MPDUs into an aggregate MAC frame. Each individual MPDU is encrypted and decrypted separately. Since MPDUs are packed together, this method cannot use the earlier 802.11 per-MPDU acknowledgement mechanism for unicast frames. A-MPDU must be used with the new Block Acknowledgement function of 802.11n-Draft2.0. In order to accommodate aggregated MAC frames, the maximum frame length accepted by the PHY is increased from 405 in previous standards to 6555.
Chapter 2 - Technology in 802.11n often not feasible, given the time constraints of the media, but may be problematic for other real-time applications.
The format of the Block Ack is a bit-map to acknowledge each outstanding frame: it is based on a mechanism originally defined in 802.11e. The bit-map identifies specific frames not received, allowing selective retransmission of only those required.
MAC layer enhancements: Spatial multiplexing power save (SM power save)
The basic 802.11n-Draft2.0 power save mode is based on the earlier 802.11 power save function. In this mode, the client notifies the AP of its power-save status (intention to sleep), then shuts down, only waking for DTIMs (Delivery Traffic Indication Maps) broadcast by the AP, while the AP buffers downlink traffic for sleeping stations between DTIMs. www.arubanetworks.com
Designed for Speed - 802.11n Theory and Practice Power save in 802.11n-Draft2.0 is enhanced for MIMO operation with SM power save mode. Since MIMO requires maintaining several receiver chains powered-up, standby power draw for MIMO devices may be considerably higher than for earlier 802.11 equipment. To mitigate this, a new provision in 802.11n-Draft2.0 allows a MIMO client to power-down all but one RF chain when in power save mode. When a client is in the dynamic SM power-save state, the AP sends a wake-up frame, usually an RTS/CTS exchange, to give it time to activate the other antennas and RF chains. In static mode, the client decides when to activate its full RF chains, regardless of traffic status.
AP
Data
Multicast
Data
AP A
Sleep
Data
Multicast
Data
Trigger/Data
Ack / Sleep
time
Unscheduled PSMP is the simpler mode: it is very similar to U-APSD, supporting both trigger-enabled and delivery-enabled options. Each sleep interval is considered and signaled independently, with the client determining when to wake to receive or transmit data. In the diagram above, the sleep frame informs the AP that the client will stop receiving frames until further notice. When the client wishes to communicate, it sends a regular or trigger frame to the AP, and both parties then transmit whatever data is queued. At the end of this exchange, the client can indicate its return to sleep mode. Scheduled PSMP is very similar to the S-APSD function introduced in 802.11e. The client requests a reservation for a T-Spec (traffic 4 Aruba Networks
Chapter 2 - Technology in 802.11n specification) from the AP, giving details of data rate, frame size, frame interval and access class (QoS priority) of the traffic streams it wishes to send and receive. The AP, once it has admitted this T-Spec, defines and a polling schedule for the client. Since there may be several clients using S-PSMP, the AP defines global PSMP SP (service period) for SPSMP traffic, informing other stations they cannot transmit during these intervals. Once a PSMP SP is declared, the AP first transmits data in the downlink direction to all applicable S-PSMP clients during the DTT (downlink transmission time), then accepts traffic from clients during the UTT (uplink transmission time).
A
802.11n PSMP
C
802.11n Not PSMP 802.11n PSMP
PSMP = power save multi -poll DTT = downlink transmission time UTT = uplink transmission time
AP B
Data A
Data B
Data B
PSMP-DTT
Data A Data B
PSMP-UTT
AP
PSMP-DTT
Data A Data B
PSMP-UTT
Publish schedule
Arequest B C
TSpec request
TSpec
Data
Data
Data
Data
S-PSMP is a very efficient way to transmit streaming or periodic traffic over 802.11n: there is no contention for the medium, as everything depends on a published schedule.
www.arubanetworks.com
5
Designed for Speed - 802.11n Theory and Practice Support for legacy clients. 802.11a/b/g clients can connect to 802.11n APs. They will not be able to use 802.11n features, and their performance will be only marginally improved when connecting to an 802.11n AP. Awareness of neighbouring or overlapping 802.11a/b/g networks. This is particularly important when using the new 40 MHz channel capability, which would impair the performance of such networks.
Unfortunately, as explained elsewhere in this note, working with legacy 802.11 clients and networks degrades the performance of 802.11n considerably, to the point where 802.11a/b/g clients will see very comparable performance whether they are using an 802.11a/b/g or 802.11n access point. In addition, working with legacy clients poisons the 802.11n cell: its capacity will be severely degraded as soon as even one legacy client is present. This does not negate the need for legacy operation, but it does increase the urgency of upgrading the client population to 802.11n. The diagram below shows how introducing an 802.11a client into an 802.11n cell reduces the throughput (based on data rates alone: the overhead introduced by co-existence mechanisms will further reduce throughput).
A
802.11n 52 Mbps 2SS SDM
B
802.11n 2SS SDM 802.11n 104 Mbps
AP
B to AP 1KB @ 104 Mbps AP to A 1KB @ 52 Mbps
A to AP 1KB @ 52 Mbps
AP to A 1KB @ 52 Mbps
2N
2N
2N
time
Data transferred = 6KB, 3KB to/from A and 3KB to/from B ***(for a time interval of 9N*).
A
802.11n 52 Mbps
2SS SDM
B
802.11n 1SS 802.11a 24 Mbps 2SS SDM 802.11n 117 Mbps
AP
C
AP to A 1KB @ 52 Mbps
A to AP 1KB @ 52 Mbps
AP to C 1KB @ 24 Mbps
2N
4N
2N
time
Data transferred = 4KB, 2 KB to/from A, 1KB to/from B, 1KB to/from C ***(for a time interval of 9N) . *** In a time interval of 9N, ignoring contention time
HT-LTF
HT-LTF
HT-LTF
Data
Non-HT format
L-STF L-LTF L-SIG Data
Key Short Training Field Long Training Field Signal Greenfield Legacy (e.g. pre802.11n High Throughput (e.g. 802.11n)
HT mixed format
L-STF L-LTF L-SIG HT-SIG HT-STF HT-LTF
HT-LTF
HT-LTF
HT-LTF
Data
High Throughput (HT). In HT or Greenfield mode, the AP does not expect to connect to any legacy 802.11 clients, and indeed, assumes that there are none operating in the area. Even so, the first part of the preamble is a legacy short training sequence, enabling other devices and APs to sense that there is 802.11 equipment in the area. However, after that no indication is available that will allow older devices to understand the remaining part of the transmission: it is all in HT-format. HT-mode is the only one of the three that is not mandatory in the 802.11n standard. Non-HT format. This is essentially legacy mode. The frames are all in 802.11a/g format (PHY and MAC), so they can be understood and decoded by 802.11a/b/g clients. This mode gives essentially no performance advantage over legacy networks, but offers full compatibility. Non-HT mode cannot be used with 40 MHz channels. HT Mixed Format. As might be expected, this allows operation of 802.11n clients in HT mode, while legacy clients are fully-supported. There is a full legacy preamble, then the option of using HT or legacy format afterwards. The preamble allows legacy clients to detect the transmission, acquire the carrier frequency and timing synchronization, and the L-SIG field allows them to estimate the length of the transmission. This (mixed) mode can be used in a 40 MHz channel, but to make it compatible with legacy clients, all broadcast and non-aggregated control frames are sent on a legacy 20 MHz channel as defined in 802.11a/b/g, so as to be interoperable with those clients. And of course all transmissions to and from legacy clients must be within a legacy 20 MHz channel.
www.arubanetworks.com
7
AP
802.11a (L) 20 MHz
AP A B
RTS (HT)
CTS (HT)
CTS (L)
Sets NAV
Sets NAV
Sets NAV
time
Firstly, for 802.11n clients, the AP advertises a forthcoming switch of operation, allowing these clients to continue communicating in all time slices, whether 20 MHz or 40 MHz. Clearly, throughput is lower during 20 MHz time slices, but nevertheless, two-way transmission between the client and the AP can continue uninterrupted. While the diagram shows all time slices of equal length for simplicity, the AP can choose between a number of defined time intervals for each time slice. For legacy 802.11 clients, only one of the three modes of operation (40 MHz, upper 20 MHz, lower 20 MHz channels) will be possible at any time: these clients will only operate in one of the 20 MHz channels. During time slices when the AP is in one of the other two modes, these clients must be informed that they cannot transmit: the AP will not be able to receive their frames. This is achieved by the AP transmitting a self-addressed CTS (clear to send, see below) frame with a duration value equal to the next time-slice duration. When clients hear this frame, they set their NAV (network allocation vector) to this value: 8 Aruba Networks
Chapter 2 - Technology in 802.11n under the rules of all 802.11 standards, they are not allowed to attempt transmission until this timer has expired. Phased coexistence operation is also a good-neighbour policy because APs and clients in range of the AP will be able to hear the self-addressed CTS messages and set their NAV timers appropriately, avoiding one form of co-channel interference. However, 802.11a/b/g cells operating in range of a PCO cell will experience reduced capacity, as APs and clients will be inhibited from transmitting for a significant percentage of the time previously available.
AP
802.11a (L) 20 MHz
AP A B
RTS (HT)
CTS (HT)
CTS (L)
Sets NAV
Sets NAV
Sets NAV
time
Dual-CTS makes the network a good neighbour to overlapping or adjacent legacy 802.11 networks. It also solves the hidden node problem where different clients in a cell may not be able to hear each www.arubanetworks.com
Designed for Speed - 802.11n Theory and Practice others transmissions, although, by definition they can all hear the AP and its CTS frames. However, the use of RTS/CTS further reduces the data throughput of the cell.
-20MHz -28dBr
fc
fc
11
Channel
10
11
12
13
Centre Frequency 2412 2417 2422 2427 2432 2437 2442 2447 2452 2457 2462 2467 2472
Channels defined for 2.4 GHz band, showing common 20 MHz channel plan and 40 MHz options
40
Aruba Networks
Chapter 2 - Technology in 802.11n This means that in practice, it is unlikely that 40 MHz channels will be used in the 2.4 GHz band. But this does not mean that 802.11n should not be used: there will be performance improvements even when a 20 MHz channel is used, although the presence of legacy clients will reduce the realized benefits. This also illustrates one of the drivers for using 802.11n in handheld clients such as mobile Wi-Fi phones, and even in plug-in PC NIC cards where space is limited and designers are tempted to revert to 802.11g components for power consumption, size and cost reasons. Allowing such devices to work in Draft-n mode will avoid the need for the cell to fall into a PCO or other coexistence mode, and greatly increase overall cell performance, even if a single antenna with a single transmit/receive chain is used. Designers of small clients such as cellphones are already attacking these challenges using techniques such as diverse polarization and attitude, but a PC with antennas embedded in the frame is likely to realize better MIMO performance than a small hand-held 802.11n device for some time to come.
www.arubanetworks.com
41
36
40
44
48
52
56
60
64 5320
100
104
108
112
116
120
124
128 132
136
140
US intermediate band (UNII II extended) 5450-5725 MHz 11x 20 MHz channels 5x 40 MHz channels Requires DFS
Band 149 153 157 161 Edge 5725 5745 5765 5785 5805
US UNII III / ISM band 5725-5850 MHz 4x 20 MHz channels 2x 40 MHz channels
Channels defined for 5 GHz band (US regulations), showing common 20 MHz channel plan and 40 MHz options
We anticipate that Draft-n enterprise networks will make extensive use of the 5 GHz band. The diagram above shows the available 40 MHz channels in the various 5 GHz bands. Note that Dynamic Frequency Selection a technique from 802.11h, is required for operation in some of these bands: Draft-n silicon is generally compliant with the DFS requirement. The plenitude of channels will allow widespread deployment with 40 GHz channels, and will also allow deployment options with parallel 802.11a and Draft-n cells at 5 GHz, allowing the Draft-n cells to load-balance legacy clients to the 802.11a cells.
42
Aruba Networks
APs High Throughput PHY 40 MHz channel in 5 GHz 20 MHz channel in 5 GHz 40 MHz channel in 2.4 GHz 20 MHz channel in 2.4 GHz M O M
O O
O M
Supported
www.arubanetworks.com
4
2x2 / 4x4
Greenfield Mode
Supported.
44
Aruba Networks
PSMP. Power Save Multipoll MTBA Block Ack Quality of Service WMM (including WMM Power Save) Security WPA2 M M
M M
PC
Draft-n
4Q 2007 1Q 2008
AP PC
Draft-n Draft-n
2Q-Q 200
AP, PC
Phones
2010
Handheld devices
46
Aruba Networks
www.arubanetworks.com
47
Designed for Speed - 802.11n Theory and Practice In general, we recommend that when Draft-n access points are used, they should support two radio-sets, one operating with a 40 MHz channel at 5 GHz and the other with a 20 MHz channel at 2.4 GHz. This will provide maximum performance and flexibility for a price premium over a Draft-n + 802.11a/b/g or a single 802.11n access point. Even though phones, location tags and other handheld clients will be slow to adopt 802.11n, it is important that they eventually achieve this goal. Even if they were only to support one antenna, one RF chain and a single spatial stream, the upgrade will prevent access points from dropping into legacy-compatible modes and hence will remove a serious inhibitor of performance as measured by cell capacity. One driver for introducing Draft-n access points in an existing network will be defensive: to provide rogue AP detection that extends to the APs on the shelves of consumer electronics shops. If previous upgrades of Wi-Fi are an indication, the consumer market will lead the enterprise market for several quarters, and while their beacons include legacy transmissions, the only way to reliably classify an 802.11n rogue and its clients will be to use an 802.11n access point. Most WLAN vendors are expected to continue supporting an architecture where a single access point can dynamically switch between coverage and monitoring mode, and in a Wi-Fi service delivery network we expect the usual ratio of 15% to 25% extra access points in order to ensure full monitoring even under heavy load, as well as dynamic network healing in the event of AP failure. If new APs are used for the sole purpose of monitoring, they would normally be installed at wider spacing than indicated above. This may change for 802.11n, since some of the range-enhancing techniques are not symmetrical and depend on the client implementing and enabling particular features.
Greenfield
This strategy is not really a migration: it is building a network with new APs and new clients where there was no WLAN before. This offers an opportunity to start with all Draft-n coverage, and to ensure that clients are Draft-n-capable, or can be rapidly phased-out in favour of capable substitutes. The foremost concern with this strategy is that all clients must be Draft-n-capable, otherwise older clients will have shorter range, similar to 802.11a/g distances, and will experience very low connectivity rates and coverage holes, at the same time reducing the potential performance of 802.11n clients in the cells where they are connected. 48 Aruba Networks
802.11n coverage
As seen from the figures above, it may be possible to reduce the number of APs required to cover an area by a factor of 0% or more, when comparing Draft-n with older 802.11 technologies. This may mean that the capacity of a cell becomes more important than the connection speed of any data client, but this is easily incorporated into the planning stage. The savings indicated above may make it worthwhile to upgrade clients to 802.11n in order to take advantage of the wider AP spacing this enables. Other aspects of a greenfield deployment include rogue detection and intrusion prevention. These functions should work successfully for all types of 802.11a/b/g/n devices. A further consideration will be whether to cover one or both of the 2.4 GHz and 5 GHz bands. Enterprise-class PC Draft-n clients are likely to be 5 GHz-capable, so for these clients, coverage of 5 GHz only will be sufficient. However, the likelihood of requirements for 2.4 GHz-only client support (as discussed above) means it will be important for most enterprise deployments of this type to include dual-radio access points providing coverage of the 2.4 GHz band, even if this is legacy coverage and it becomes intermittent at the edge of the cell (due to Draft-ndriven access point spacing).
www.arubanetworks.com
4
Designed for Speed - 802.11n Theory and Practice RF planning can proceed with 802.11n range assumptions, allowing fewer APs. Use increased AP spacing only if all clients are capable of at least 2 spatial streams. Restrict existing APs to a subset of the available 5 GHz channels, and use the remainder for 802.11n. Consider power to the APs, edge switch ports, etc. Place APs to avoid line-of-sight for best MIMO performance. Disable co-existence mechanisms such as PCO that support mixedmode operation. Enable good-neighbour mechanisms such as dual-CTS if there are other 802.11a APs in the same building, on the same channel.. Ignore 40 MHz-intolerant requests, or reject such clients. Use dual-radio APs where the primary radio(s) is 802.11n at 5 GHz, while the second radio (can be 802.11a/b/g) runs in the 2.4 GHz band. Remember that this scheme does not guarantee complete 2.4 GHz coverage for non-Draft-n clients: there may be areas at the edge of cells where signal strength will not support an 802.11a/b/g signal.
Note that for a lightly-loaded greenfield network, using Draft-n access points with 802.11a spacing, and enabling PCO for a mixed client base presents a perfectly acceptable solution: throughput will initially be low, maybe even lower than an 802.11a network until the number of legacy clients is reduced, but the network can be installed once and grow with the client base, without any required reconfiguration.
AP-overlay
Here, the Draft-n network would be planned as if there were no existing network, for optimum placement of the new APs. The old and new networks will be able to operate in parallel if care is taken with RF channel allocation, and when the new one is complete and all clients are 802.11n-capable, the old APs can be de-installed or abandoned. For this scheme, it is important that the radio planning and management algorithm of the WLAN can manage 802.11a, b, g and Draft-n cells and APs simultaneously.
50
Aruba Networks
802.11n coverage
802.11a/g coverage
Since Draft-n makes best use of the 5 GHz band (with 40 MHz channels), it makes an attractive overlay on an 802.11b/g network operating at 2.4 GHz: the original network can operate in the 2.4 GHz band (or even 2.4 GHz and several channels of the 5 GHz band) while the Draft-n network can use the remainder of the 5 GHz band. In this scenario, there will be no need for legacy clients to connect to the 802.11n network, so it can operate in HT mode. (It may be useful in the early days of such a network to use it in PCO mode if there are very few 802.11n-capable clients, but this will negate much of the throughput advantages of Draft-n.) Against this, the Draft-n network is additive, so additional cabling, power and edge switch ports will probably be required (but it is likely these upgrades would eventually be needed, anyway).
www.arubanetworks.com
51
Designed for Speed - 802.11n Theory and Practice Switch PCO off, use dual-CTS. Consider power to the APs, edge switch ports, etc. Place APs to avoid line-of-sight for best MIMO performance. Use single-radio APs if there is no future requirement for 2.4 GHz Draft-n coverage. If there is an intent to eventually remove the older infrastructure, it may be better to use dual-radio access points for the overlay so the second radio can be switched on later (becoming similar to the greenfield model over time, as the original infrastructure disappears).
AP substitution
In this model, existing APs are swapped one-for-one with new Draft-n APs. If every AP is replaced with Draft-n, the resulting network will have plenty of capacity we estimate perhaps a three-fold improvement, depending on the mix of 802.11a/g and Draft-n clients but will have more APs that would be strictly necessary. In most networks it will be possible to abandon some AP locations either running an RF planning tool, or using RF self-calibration tools on a network mid-way through migration will give an indication of whether this is feasible.
802.11n Access Points
802.11a/g Access Points
A network of mixed Draft-n and 802.11a/g or even 802.11b APs can operate indefinitely: this is a possible solution where Draft-n is important for classifying rogue APs but not for capacity or range considerations. Note that even though APs can be swapped at a location, the new Draft-n AP may require different power and Ethernet connections.
52
Aruba Networks
www.arubanetworks.com
5
Designed for Speed - 802.11n Theory and Practice Opportunities to mount APs in the workspace - on walls, partitions and on furniture, but not in the ceiling plenum space increase with Draft-n, as the non-line-of-sight characteristics are favourable. This may mitigate some of the installation issues mentioned above: if GE drops already exist in some offices, Draft-n APs can be located to take advantage of these, and local power for the AP via a wall brick may be a reasonable option.
54
Aruba Networks
Chapter 5 - Conclusion
Chapter 5 Conclusion
This note has two goals: to explain the technology and features of the 802.11n standard, and to examine its implications for the design of enterprise Wi-Fi networks, in terms of likely product development cycles. It is clear that 802.11n represents a significant leap forward in technology and performance for enterprise networks. Uniform greenfield 802.11n networks will be able to offer much higher capacity and longer range than current WLANs, and there are potential savings in terms of fewer access points to cover a given area. However, this note has also identified several issues that will delay the adoption of 802.11n in enterprise networks. These include the incomplete state of the standards, infrastructure requirements such as LAN edge switch ports, cabling and power, and the installed base of 802.11a/b/g clients and WLANs that must be considered in any migration strategy. Taken together, these are likely to slow the 802.11n adoption wave, so a number of years will pass before we can realize the full benefits. In the pages above, we have discussed the facts of 802.11n, along with our best estimates of future activity. There is a case in enterprise networks for tuning and extending existing 802.11a/g networks in the short-term, while testing and running pilot trials on Draft-n and eventually full 802.11n-compliant products. When and how to migrate to 802.11n will be an important decision: the information here is intended to assist the enterprise network manager in formulating an optimum upgrade strategy. Despite our best efforts, developments will inevitably prove us wrong in some of the predictions above: our advice to the reader is to check often with the technical press and vendors for current information. 802.11n offers very real and exciting benefits: it will eventually change the way we build and operate enterprise wireless LANs.
www.arubanetworks.com
55
56
Aruba Networks
Appendix capacity. And of course, when the legacy station transmits or receives, it does so at lower data rates, reducing the effective capacity of the cell. The 400 nsec guard interval is not often a realistic option in an enterprise network, where there will be many multipath reflections, some with long delay spreads. Client design for 802.11n is challenging, as it is difficult on many devices to find the space to mount extra antennas, the extra RF chains and processing require more board area and power, and of course 802.11n silicon will command a price premium for several years over 802.11a/g silicon. Not least, NIC cards and clients such as Wi-Fi phones are complex to design: until recently, Wi-Fi phones were limited to 802.11b. All these effects serve to reduce the effective capacity of a cell. Thus, while 802.11n advertises rates to 600Mbps, the expected capacity of an 802.11n cell is between 100 and 200Mbs, and it could certainly be less if clients connect over long distances, transmit short frames, or there are legacy 802.11a/b/g clients present. However, this is still an increase of 5x over 802.11a/g technology. Here are our estimates of reasonable expectations for data rates in an enterprise 802.11n deployment (this assumes greenfield deployment with no legacy 802.11a/b/g clients or APs). Interpret this table as a comparison of the achievable data rate of 802.11n with that of 802.11a or 802.11g at the same distance from the access point. Alternatively, it can be read as the capacity of a cell (with all-802.11n clients) when compared with a cell of the same radius.
Spatial Streams Channel width (MHz) 20 40 40 40 Guard Interval (nsec) 800 800 800 800 Range of PHY rates (MHz) 7.2-72.2 1.5 - 15 27 - 270 40.5 - 405 Performance relative to 802.1 a/g 1x - 1.5x 1.5x - 2x 2x - 4x 4x - 6x
In addition to the PHY effects above, the many MAC enhancements in 802.11n will increase the throughput, and hence the capacity of a cell. Analysis of these effects is challenging, as they are extremely dependent www.arubanetworks.com 57
Designed for Speed - 802.11n Theory and Practice on the data patterns. Some of the enhancements are aimed specifically at streaming media such as video, but there has been speculation that they may actually reduce the throughput of other types of traffic such as voice or file transfers.
Methodology
This explanation is not necessary to an understanding of the results below, but is included so the scientifically-inclined can understand the degree of simplification involved in deriving the results. In deriving the expected data rates and how they vary with distance from the access point, the following steps are employed: Calculate free space loss with distance. Apply a model (Rappaport et al) for typical in-building propagation loss. Calculate the signal to noise ratio (SNR) by distance from the access point. Apply some random statistics to the above to simulate a real-world network. Calculate the Shannon bandwidth of the channel, based on the SNR. Round to the nearest (lower) data rate available
This necessarily includes many approximations. In addition to the methodology above, the following assumptions were used: 58 Aruba Networks
Appendix Propagation loss at 5.5 GHz In-building propagation loss index of 2.7 Assumed transmit power to the antenna of 100 mW, and combined antenna gains of 6 dBi 20 MHz, 40 MHz channel bandwidths for calculating thermal noise and Shannon bandwidth Variation of +/-6 dB in signal strength due to noise and fading effects. This is applied as a random factor uniformly distributed between +/-6 dB of the mean calculated by the in-building propagation formula. No allowance is made for beamforming, as this will not be incorporated in shipping silicon until at least 2008. Only up to spatial streams are analysed, rather than the maximum of 4 in the standard: early products are expected to extend only antennas and 2 spatial streams.
Designed for Speed - 802.11n Theory and Practice any particular case, but much work has been done on characterizing general performance when averaged over many buildings, clients or measurements. Of the many different models proposed, we use one of the simplest: P = x10log10(d) + 47.26 Where is a parameter showing increased loss over distance compared with free space.
60
Aruba Networks
Appendix
Free space and in-building propogation loss model
Maximum data rate (Mbps) 80
70 60 50 40 30 20 10 0
11 12 14 15 16 18 19 21 22 23 25 26 28 29 30 32 33
Distance (meters)
The graph above shows the results of the free-space and in-building propagation models. The free-space line assumes uninterrupted lineof-sight behaviour, and is obviously the upper-bound of what could be expected in practice: it will not be possible to improve on these figures in practical networks. The in-building plot shows a much steeper decline of data rate with distance, as one might expect: generally it assumes non-line-of-sight, and many multipath reflections. The general form of this curve is well-accepted: the slope varies depending on the type of environment. Our assumption is for a propagation factor of 2.7 (estimates vary from 2.0 for free-space to .5 in extremely challenging environments), and we judge this reasonable for a building with few thick internal walls, but with desks, cubicle partitions and thin office walls and doors. A modern office building might be typical of this, whereas buildings constructed with internal walls of brick or concrete, or containing large metal objects would warrant a higher factor.
35
Designed for Speed - 802.11n Theory and Practice The methodology, explained above, is to apply a random variation in the range +/- 6dB to the mean values for in-building propagation above. This gives the figure for a single spatial stream in the plots below. To model multiple spatial streams (SS), a number of independent random events are calculated, one for each SS in the system. These are then added using a variety of weightings, as it is not reasonable to expect perfect RF isolation of the spatial paths: adding a second SS will cause some interference and degradation of the first. Subtracting from this effect, however, the transmitter can use feedback from the receiver to characterize the combination of channels and adjust its antenna weightings. Overall we expect these effects to balance: the calculations above expect that the second and third spatial streams are directly additive to the overall data rates.
Data rates varying with distance
Maximum data rate (Mbps) 160 140 120 100 80 60 40 20 0
10
12
14
15
17
19
21
23
24
26
28
30
32
Distance (meters)
Free space
The diagram above also includes a simplification, in that it assumes non-line-of-sight, but with significant multipath: good conditions for MIMO with SDM. Our practical experience in our own office building suggests that actual data rates can be lower at short distances from the access point, because there is more likely to be line-of-sight, especially with ceiling-mounted antennas. This means that at short range, lineof-sight negates the MIMO effect, and data rates are similar to single 62 Aruba Networks
33
Appendix spatial stream figures. As distance increases, the dominant line-of-sight/ near-line-of-sight signal is attenuated, promoting multiple, more isolated RF paths, and overall data rates increase. Now the results must be matched to actual 802.11 data rates, depending on different coding combinations. The plot below shows this for 802.11a, which can be used as a benchmark for comparison with the 802.11n results.
Data rates varying with distance - 802.11a
Maximum data rate (Mbps)
80
70 60
50 40
30 20
10 0
11
12
14
15
17
19
20
22
23
25
27
28
30
31
33
Distance (meters)
Using the same methodology we model 802.11n data rates for different combinations of spatial streams and channel bandwidth.
www.arubanetworks.com
35
6
250 200
150 100
50 0
11
12
14
16
18
20
22
24
26
28
30
31
Distance (meters)
1 SS, 20MHz
802.11a (20MHz)
The results demonstrate that even with one spatial stream, 802.11n gains range over 802.11a/g. This can be ascribed to STBC, MRC and a series of minor PHY-layer improvements (beamforming is not included in this model). The extent of this improvement of 802.11n range over 802.11a is perhaps 5% for single antennas, increasing to 50% where extra antennas are available. Also, the effective range of 802.11n (the distance at which the lowest data rate becomes unreliable) increases slightly with the number of spatial streams: maybe 20% with the second spatial stream. While this may seem to be a small increase, it will allow an increase in the spacing of access points, and a reduction in their number of perhaps 0-50%. Note that as this model deals only with range and data rates at the PHY layer, the achievable throughput will be less than indicated above sometimes considerably less, especially when older 802.11a/b/g clients are present in the cell. The table below shows the estimated contribution of different 802.11n features in extending range (range here is defined as the distance at which the error rate at the lowest defined data rate becomes excessive: the range is compared to 802.11a). 64 Aruba Networks
33
Appendix
Feature Increased channel bandwidth, 20 MHz channel, modulation & coding improvements Increased channel bandwidth, 40MHz channel, modulation & coding improvements Diversity effects from multiple antennas, 2x1/x1 Diversity effects from multiple antennas, 2x2, x2 Transmit beamforming 2x1 Estimated Improvement 5% Notes More subcarriers are used, but the improvement is directed to a higher data rate rather than increased range. As above.
5%
0%
STBC or MRC. This improvement will be realized in 2007. STBC and MRC. This improvement will be realized in 2007. This will not be implemented in silicon (at least with intervendor interoperability) in 2007. (Not additive with diversity). (This is not additive with diversity effects or beamforming.) (This is not additive with diversity effects or beamforming.)
50%
50%
20%
0%
The practical solutions available in 2007 should benefit from diversity effects and the two-spatial-stream opportunity, giving the range estimates (based on two spatial streams) used through this paper for a 2007 greenfield 802.11n network. Note that MAC enhancements do not affect the maximum range of an 802.11n signal, although they do serve to increase the overall throughput at any particular AP to client distance.
www.arubanetworks.com
65
Forms of MIMO
There are already more than enough terms and acronyms in this document. We have deliberately avoided introducing more than were necessary to explain 802.11n. But the reader may have encountered SIMO and other terms. The figure below shows that these are just degenerate forms of MIMO with a single antenna chain at one or the other end of the connection:
SISO
(Single input, single output)
MISO
(Multiple input, single output)
Tx Tx Rx Tx Rx
SIMO
(Single input, multiple output)
Rx Tx Rx
MIMO
(Multiple input, multiple output)
Tx Tx Rx Rx
66
Aruba Networks
Appendix
Channel estimation
As explained above, all the techniques used for MIMO require the receiver to estimate the channel characteristics. It does this by receiving a sequence of known signals. The diagram below explains how this is done, both in general matrix algebra and with an example. This is for a 2x2 system.
TA hA2 = 0.7
2
hB2 = 0.5 hA1 = 0.3
R2
A
TB
hB1 = 0.9
R1
To determine the channel characteristics, known symbols are sent from antennas A and B. If enough known symbols are sent, it is possible to construct these equations: R1 = TA x hA1 R2 = TA x hA2 + TB x hB1 + TB x hB2 Algebraic form Numerical example R1 = 0.3TA R2 = 0.7TA + 0.9TB + 0.5TB
TA = R1 x qA1 TB = R1 x qB1
+ R2 x qA2 + R2 x qB2
Now, for every received symbol (RA and RB in this case), the original transmitted symbol can be determined.
www.arubanetworks.com
67
Glossary and References RF RIFS RTS SISO SM SNR STBC STBC/SM TRQ TxBF UTT WLAN Radio frequency Reduced inter-frame spacing Request to send Single input, single output Spatial multiplexing Signal to noise ratio Space time block code Space time block code / spatial multiplexing Training request Transmit beamforming Uplink transmission time (Power save multi-poll) Wireless Local Area Network
References
The IEEE is the fount of 802.11 standards. While it does not publish standards until final ratification, much useful information may be found here: http://grouper.ieee.org/groups/802/11/. In particular, there is a useful summary of each forthcoming amendment (currently under WG Info / 802.11 Quick Guide) and a current estimate of schedules (use the search box for timeline). The information in this paper was taken from and checked against the IEEE document P802.11n draft 2.0. The Wi-Fi Alliance certification is Draft-n. The WFA web site is here: http://wi-fi.org/
www.arubanetworks.com
6
2007 Aruba Networks, Inc. All rights reserved. Aruba Networks, BlueScanner and RFprotect are trademarks of Aruba Networks, Inc. All other trademarks or registered trademarks are the property of their respective holders. All rights reserved. Specifications are subject to change without notice.
70
Aruba Networks
www.arubanetworks.com
WP_80211n_US_071018