802.

11n Under the Microscope

Vivek Shrivastava, Shravan Rayanchu, Jongwon Yoon, Suman Banerjee

Department of Computer Sciences, University of Wisconsin, Madison, WI 53706, USA
{viveks, shravan, yoonj, suman}@cs.wisc.edu

ABSTRACT range and more reliable coverage than 802.11 a/b/g net-
We present an experimental study of IEEE 802.11n (high works. In order to provide such gains, it introduces a variety
throughput extension to the 802.11 standard) using com- of mechanisms such as physical layer diversity (using Mul-
modity wireless hardware. 802.11n introduces a variety of tiple Input Multiple Output (MIMO) technology), channel
new mechanisms including physical layer diversity techniques, bonding, and frame aggregation. The goal of this paper is
channel bonding and frame aggregation mechanisms. Using twofold: (1) To provide a better understanding of 802.11n
measurements from our testbed, we analyze the fundamen- by experimentally evaluating the potential impact of each
tal characteristics of 802.11n links and quantify the gains mechanism (and their combination) on client throughput
of each mechanism under diverse scenarios. We show that under diverse scenarios (2) To re-evaluate the effectiveness
the throughput of an 802.11n link can be severely degraded of some prior wireless research in view of these new mecha-
(up to ≈ 85%) in presence of an 802.11g link. Our results nisms (e.g., evaluating the benefits of MAC-diversity [5] in
also indicate that increased amount of interference due to presence of physical layer diversity offered by MIMO).
wider channel bandwidths can lead to throughput degrada- We now briefly describe the different mechanisms used
tion. To this end, we characterize the nature of interference by 802.11n. These mechanisms are illustrated in Figure 1.
due to variable channel widths in 802.11n and show that For detailed information on these mechanisms, please refer
careful modeling of interference is imperative in such scenar- to [7].
ios. Further, as a reappraisal of previous work, we evaluate PHY-diversity (MIMO): IEEE 802.11n employs a va-
the effectiveness of MAC level diversity in the presence of riety of physical layer diversity mechanisms for achieving
physical layer diversity mechanisms introduced by 802.11n. higher throughput and improved packet reception capabil-
ities. In 802.11n, receiver diversity is implemented by us-
ing Maximum Ratio Combining (MRC), a technique which
Categories and Subject Descriptors optimally combines signals from multiple antennas taking
C.2.1 [Network Architecture and Design]: Wireless com- into account the signal-to-noise ratio (SNR) of the signals
munication received at different antennas. The transmit diversity tech-
niques used in 802.11n include Space Time Block Coding
(STBC) and Cyclic Shift Diversity (CSD). These techniques
General Terms improve the signal reception by spreading it over multiple
Experimentation, Measurement, Performance transmit antennas using specialized coding (STBC) or phase
shifting techniques (CSD).
Keywords Frame Aggregation: IEEE 802.11n provides an option of
combining multiple data frames ready for transmission into
802.11n, MIMO, Wireless, Frame Aggregation, Channel Bond- an aggregate frame (Figure 1). Frame aggregation helps
ing, MAC Diversity, PHY Diversity, Performance amortize the channel contention and backoff delays by trans-
mitting the aggregated frame (i.e. multiple data frames) in
1. INTRODUCTION a single transmission opportunity on the channel.
This paper presents an experimental study on the per- Channel Bonding: IEEE 802.11n also introduces two dif-
formance of the new IEEE 802.11n (draft) standard using ferent channel bandwidths – 20 MHz and 40 MHz. Theo-
a real testbed. IEEE 802.11n is a next generation wire- retically, using a 40 MHz band should double the amount
less LAN technology that promises higher data rates, longer of throughput achieved using a 20 MHz band. However, as
shown in Figure 1, all the 40 MHz channels are partially
overlapping in the 2.4 GHz band, as opposed to the 20 MHz
channels 1, 6 and 11 which are non-overlapping. Thus us-
Permission to make digital or hard copies of all or part of this work for ing 40 MHz channels can also lead to degradation in the
personal or classroom use is granted without fee provided that copies are throughput due to increased interference with neighboring
not made or distributed for profit or commercial advantage and that copies channels.
bear this notice and the full citation on the first page. To copy otherwise, to In this paper, we systematically evaluate the implications
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
of using the aforementioned mechanisms and their impact on
IMC’08, October 20–22, 2008, Vouliagmeni, Greece. network throughput. Specifically, we evaluate the following:
Copyright 2008 ACM 978-1-60558-334-1/08/10 ...$5.00.
Figure 1: Different mechanisms used in the 802.11n standard. We summarize our findings regarding each
mechanism.

100
20 MHz, No Aggregation
20 MHz, Aggregation
80 40 MHz, No Aggregation
40 MHz, Aggregation

Throughput(Mbps)
60

40

20

Figure 2: 802.11n testbed used for the experiments. 0

The nodes are placed at locations L1-L9. 600 1,200
Packet Size (Bytes)

• What are the throughput gains of using each mecha-

nism (MRC, frame aggregation, channel bonding) on Figure 3: Average throughput achieved on a isolated
an isolated 802.11n link? What are the factors affect- 802.11n link under different combinations of channel
ing these gains? width and aggregation mechanisms.

• What is the performance penalty when a 802.11g link

is operating near an 802.11n link? What mechanisms
each component in isolation and in presence of interference.
can be effective in such scenarios?
Further, we also identify the scenarios under which a specific
• What is the impact of using channel bonding on net- mechanism is more useful. First we detail our experimental
work design? Can we use 40 MHz channel efficiently methodology for the paper.
in the 2.4 GHz band? When should we choose a 40 Experimental Methodology: All the experiments reported
MHz channel vs. 20 MHz channel? in this paper are performed on our 802.11n testbed (shown
in Figure 2). The wireless nodes used in our experiments
• What is the nature of packet losses in 802.11n? In comprise of desktop machines (512 MB RAM, 1.2 GHz)
presence of PHY-diversity, is MAC-diversity still bene- equipped with the Edimax (EW-7728In) 802.11n (Draft 2.0)
ficial? Will mechanisms exploiting MAC-diversity like PCI wireless cards. These cards are based on Ralink chipset,
MRD [5], ExOR [1], MORE [2] provide high through- support 3X3 MIMO operation, have three detachable an-
put gains on 802.11n as well? tennas (of 3dBi gain) and operate in 2.4 GHz band. They
support channel bandwidths of 20 MHz and 40 MHz. Un-
We answer these questions by performing targeted exper-
less otherwise stated, we run our experiments at the PHY-
iments on our 802.11n wireless testbed as shown in Fig-
data rate of 300 Mbps, the maximum data rate supported
ure 2. Our main observations from the experimental study
by our card. The cards can also be configured to be used
are summarized in Figure 1. The rest of the paper answers
in 802.11b/g mode. We used RT2860 Wireless LAN Linux
these questions in detail. We first describe the impact of dif-
driver to configure the cards for our experiments. Please
ferent components of 802.11n on client performance and also
note that we intentionally use desktop machines with suit-
evaluate the performance of an 802.11n link in the presence
able processing power as we do not want our experiments
of interference from a slower 802.11g transmitter.
to be impacted by the hardware limitations of the host ma-
chine. All our experiments were conducted at night to min-
2. UNDERSTANDING THE GAINS imize interference from other wireless devices. We operate
In this section, we evaluate the three mechanisms used in a orthogonal channel from that being used by our depart-
by 802.11n – frame aggregation, wider channel bandwidths ment WLAN to prevent potential interference.
and maximum ratio combining. We quantify the gains from Performance of an 802.11n link in isolation: For this
 1 1
80 A2
0.8 A2 0.8
Throughput (Mbps)
70 A3 = MAX

Probability

Probability
60 0.6 0.6
50 20Mhz, Agg 40Mhz, Agg A1
0.4 0.4 A1 A3
40 40Mhz,NoAgg
30 0.2 0.2 MAX
20 20Mhz, NoAgg
0 0
10 18 20 22 24 26 28 30 32 34 36 18 20 22 24 26 28 30 32 34 36
6M 9M 12M 18M 24M 36M 48M 54M
Data Rate SNR (dB) SNR (dB)
(a) (b) (c)

Figure 4: (a) Throughput of an 802.11n link severely degrades in presence of an 802.11g link operating on
a lower data rate. Frame aggregation improves the throughput of the 802.11n link by providing temporal
fairness.(b),(c) Lesser overlap in the PDFs of RSSI for three antennas indicate increased gains from MRC.

experiment, we fix the 802.11n transmitter at Location L1 used in combination with the 40 MHz channel width, frame
in the testbed and vary the receiver location from L2 to aggregation provides further improvements. This improve-
L9 (as shown in Figure 2). At each receiver location, we ment stems from the fact that frame aggregation helps an
perform a throughput test to the transmitter at location 802.11n link operating at a higher data rate attain a similar
L1. Figure 3 shows the average throughput for two packet temporal share of the channel when compared to that of an
sizes, using both the channel bandwidths, with and with- 802.11g link operating on a lower rate by transmitting mul-
out frame aggregation. For higher packet sizes, throughput tiple data frames during each transmission opportunity on
improvements ranging from 33% upto 2x can be achieved the channel.
using wider channel bandwidth (40 MHz), while frame ag- Benefits of MRC: As mentioned before 802.11n uses MRC,
gregation results in throughput improvements ranging from a physical layer technique which exploits antenna diversity –
10% to 75%. Further we observe that the relative gains from signals from multiple receiver antenna chains are optimally
aggregation are slightly higher in the case of 600 byte pack- combined to improve the packet delivery probability. Al-
ets as compared to 1200 byte packets. This is expected as though we cannot accurately determine the exact amount
aggregation helps amortize the cost of header transmissions of gains derived from this physical layer mechanism, we at-
and is more effective for smaller packet sizes. We now study tempt to characterize the gains (as visible at the MAC layer)
the impact of external interference on gains achieved from by looking at the RSSI of the signals on multiple receiver an-
these two mechanisms. tennas. Analyzing the packet logs from our experiments, we
Performance of an 802.11n link with interference: were able to identify three scenarios which would result in
Previous research [3] in 802.11b multi-rate networks had varying gains from antenna diversity. In order to illustrate
highlighted a performance anomaly – If there is at least these scenarios, probability density (PDF) of SNR values
one host which operates at a lower rate, the throughput at three antennas of the 802.11n receiver (A1, A2, and A3)
of all hosts transmitting at a higher rate is degraded below are plotted along with PDF of the maximum SNR for each
the level of the lower rate. We characterize the impact of packet in the experiment (MAX). Figures 4(b) and 4(c)
this anomaly in 802.11n by evaluating the performance of plot these values for two representative links belonging to
a 802.11n link (transmitter at location L3 and receiver at two different scenarios. In the first scenario (shown in Fig-
location L4) and an 802.11g link (transmitter at location L5 ure 4(b)), the SNR of one of the antennas (A3) dominates
and receiver at location L7) operating on the same channel. the other two antennas all the time. In the second scenario
The packet size is fixed at 1200 bytes. Both the links are (shown in Figure 4(c)), the MAX RSSI is derived from
saturated, that is both the transmitters always have pack- a combination of different antennas at different instants of
ets to send. The data rate of the 802.11g link was varied time, indicating improved gains due to MRC under these
from 6 Mbps to 54 Mbps. We performed the experiments conditions. We found the all the non line of sight 802.11n
using different channel bandwidths and frame aggregation links in our testbed belong to one of the two aforementioned
options for the 802.11n link. The 802.11n link operated on scenarios, showing the usefulness of MRC for such links. Fi-
fixed data rate of 300 Mbps (when using 40 MHz) and 144.5 nally, we found that for the line of sight link (transmitter at
Mbps (when using 20 MHz).1 L1 and receiver at L3), the PDF of all the antennas over-
Figure 4(a) shows that performance anomaly indeed severely lapped, indicating little gains from MRC. The plot for this
degrades the throughput of an 802.11n link to as low as 10 scenario is omitted for brevity.
Mbps (a reduction of ≈ 84%) when an 802.11g link is op-
erating at 6 Mbps on the same channel. We also observe
that this performance anomaly cannot be mitigated by using 3. CHANNEL WIDTHS AND INTERFERENCE
a wider channel bandwidth of 40 MHz. However, using frame Although introduction of wider (40 MHz) channels can
aggregation considerably improves the throughput and when lead to increased throughputs, they also imply an increase in
the observed interference. This is especially true in 2.4 GHz
1 band where the boundaries of 40 MHz channels in 802.11n
These are the maximum possible data rates when using less
than 4 spatial streams. do not line up with the (1, 6, 11) 20 MHz channels of 802.11n
 80 (6) 100

Throughput (Mbps)
(6)

Throughput (Mbps)
70 90
(5) 80 (5)
60 70 (4)
50 (2) (1) (2) (1)
(4) 60 (3)
40 50
30 40
20 (3) (0) 30
20 (0)
10 10
0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Distance (feet) Distance (feet)
(a) (b) (c)

Figure 5: (a) Setup used for channel overlap experiments. The transmitter and the receiver are co-located
while the separation (distance and channel) between the pairs is increased for each run. (b) UDP throughputs
for a channel bandwidth of 20 MHz. (c) UDP throughputs for a channel bandwidth of 40 MHz.

20T−20R 40T−20R
1 1
Normalized I−factor

0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0123456 0123456
1 1
0.8 0.8 Figure 7: Transmit spectrum masks for 40 MHz
0.6 0.6 and 20 MHz channels for the physical layer in IEEE
0.4 0.4
0.2 0.2 802.11n standard
0 0
0123456 0123456
we extend the model developed in [4] to calculate the inter-
40T−40R 20T−40R ference factor (or I-factor) that captures the amount of over-
lap between a transmission on a certain frequency FT and
Channel Separation reception on a certain frequency FR . The amount of overlap
is captured quantitatively by calculating the area of intersec-
tion between a signal’s spectrum and a receiver’s band-pass
Figure 6: Theoretical I-factors for various interferer filter. We incorporate the transmitter and receiver channel
and receiver combinations. bandwidths, bt and br into this model to derive the I-factor:
Z +∞
IF(T,R) (τ ) = ST,bt (F )BR,br (F − τ ) df
and the traditional 22 MHz channels of 802.11 b/g. We il- −∞
lustrate this in Figure 1(b), which shows that: (1) all the In above equation, the parameter τ represents the difference
40 MHz channels are partially-overlapping (2) A 40 MHz in the center frequencies of the channels i.e., τ = FT − FR .
channel might have a significant amount of spectral leakage The parameter ST,bw (f ) denotes the transmitted signal’s
on some of the 20 MHz channels. 2 While it is possible to power distribution across the frequency spectrum when a
use (non-overlapping) 20 MHz channels, it might reduce the channel bandwidth of bt MHz is used. As in [4] we approx-
maximum throughput that can be achieved with wider chan- imate ST,bw (f ) with the corresponding transmit spectrum
nel bandwidths. In this scenario, it becomes imperative to mask. We illustrate the transmit spectrum masks for a 40
understand and characterize the nature of interference intro- MHz channel and a 20 MHz channel of the 802.11n physical
duced due to these variable channel widths. To this end, we layer in Figure 7. Finally, BR,br (f ) denotes the band-pass
extend the model in [4] to characterize the interference on an filter’s frequency response when a channel of br MHz is used.
802.11n link due to partially overlapped channels and vali- Assuming the receive filter for a particular bandwidth to be
date our model using experimental results which we present same as the transmit spectrum mask [4], for 802.11n we get:
next.
Modeling 802.11n Interference: In order to characterize BR,bw (f ) = ST,bw (f ) =
the amount of interference on an 802.11n link due to trans- 8
missions on other 802.11 channels (of 40 MHz or 20 MHz), >
> −45dB if |f − Fc | ≥ (1.5 ∗ bw)MHz
−38dB if (bw)MHz ≤ |f − Fc | < (1.5 ∗ bw)MHz
<
2
Although 40MHz channels could be easily used in 5 GHz > −20dB if (0.55 ∗ bw)MHz ≤ |f − Fc | < (bw)MHz
band but initial 802.11n deployments would also need to
>
: 0dB otherwise
serve 802.11g clients for backward compatibility, forcing
them to operate in a hybrid mode (both 11g and 11n) in where Fc denotes the channel center frequency and bw is the
2.4 GHz band. channel bandwidth (20 MHz or 40 MHz) used.
 Figure 6 shows the I-factor calculated using the above specifically, we answer the following question – In presence
model for four different scenarios – when the interferer and of PHY diversity offered by 802.11n, what gains can we ex-
the receiver are both using 20 MHz channels (20T-20R), pect from MAC-diversity?. The answer to this question has
the interferer is using a 20 MHz channel and the receiver is several important implications. Mechanisms such as MRD
using a 40 MHz channel (20T-40R), the interferer is using a [5], ExOR [1] and MORE [2] depend on MAC-diversity to
40 MHz channel and the receiver is using a 20 MHz channel achieve throughput improvements and hence will be affected
(40T-20R), the interferer and receiver are both using a 40 if the gains due to MAC-diversity decrease substantially
MHz channel (40T-40R) 3 . when PHY-diversity is used.
Results: We used the set up shown in Figure 5(a) where We start with experiments which help us characterize the
four 802.11n nodes are used to form two transmitter-receiver nature of packet losses in 802.11n and understand how they
pairs (Pair-A and Pair-B). The nodes in each pair we kept are different from packet losses in 802.11b/g. We then design
in close proximity of each other and we set up UDP flows experiments to understand the gains due to MAC-diversity
from the transmitter to the receiver. We then measured the in the presence of PHY diversity. Below, we describe the
throughputs for at different physical distances, varying chan- experimental set up used.
nel separation and bandwidths. Figures 5(b), 5(c) show the Experimental Setup: We perform broadcast experiments
average throughput results across varying distances for two with a single transmitter (at L1) and we used two receivers
of the four configurations4 . The numbers in the parenthesis which are co-located with each other. The position of this
correspond to the channel separation between the two pairs. receiver pair was varied from location L2 - L9 during dif-
We observe that as the physical separation between the pairs ferent runs of this experiment. For each run, a total of
increases, the throughput improves due to decreased inter- 100, 000 packets were transmitted in broadcast mode and
ference from the partially overlapped channels. Also, at a the receivers captured the packets in the monitor mode. The
fixed distance, increase in channel separation from 0 to 6 packet size was set to 1024 bytes. We performed a total of
leads to increase in throughput. 10 runs at each location. In order to compare the packet
Further, we note that the observed throughputs correlate losses in 802.11n and 802.11g, we first performed the exper-
with the theoretical I-factor. It is interesting to note that iments with the transmitter set to 802.11n greenfield mode
linear increase in throughputs (due to increased channel sep- (using 40 MHz channel, at a PHY-data rate of 300 Mbps)
aration) correspond to a linear decrease in the I-factor. We and then repeated the experiments with the transmitter set
also note that at a distance of around 120 feet, the through- to 802.11g mode (at a PHY-data rate of 54 Mbps). The cap-
puts of the links reach the maximum for all the partially tured packet logs at the receivers are then used to analyze
overlapped channels, indicating that the channels can be the nature of packet losses in 802.11n.
reused at this distance. For the (20T-20R) case shown in Nature of Losses: In the experiments above, we observed
Figure 5(b), we observe that even at very small distances packet losses ranging from 9% to as high as 80%. Further, we
between the pairs (0-20 feet), the degradation in through- observed that the difference in delivery probabilities between
put is minimal when the channel separation is greater than the two receivers was much higher in 802.11g showing clear
4 as the I-factor is close to zero for these cases. However, benefits of using MAC-diversity. On the other hand, packet
as observed in Figure 5(c), to achieve similar affect when 40 delivery ratios for both the 802.11n receivers were almost
MHz channels are used, a separation of around at least 50 the same for almost all the locations. The delivery ratios
feet is needed even for the maximum channel separation of 6. only differed at locations with very high loss rates (≈ 70%).
This is due to the increased interference as a result of using In order to investigate whether packet losses in 802.11n are
wider channel bandwidths. Hence, there is a clear tradeoff independent at each 802.11n receiver, we take the following
in choosing a 20 MHz vs. a 40 MHz channel. This high- approach: For each set of 10, 000 packets in the above exper-
lights the importance of careful channel width assignment iments, we measure Pl (R1 ) and Pl (R2 ) which are the packet
in 802.11 networks by accurately modeling the amount of loss rates observed at receivers R1 and R2 . We also measure
interference introduced due to variable channel widths. Pl (R1 ∩ R2 ) to represent the number of broadcast transmis-
sions which were simultaneously lost at both the receivers.
In Figure 8(a) we show the scatter plot of Pl (R1 ∩ R2 ) and
4. PACKET LOSSES AND MAC-DIVERSITY the quantity Pl (R1 )∗Pl (R2 ). If the losses at two receiver are
As previously mentioned in Section 1, IEEE 802.11n em- independent, Pl (R1 ∩ R2 ) ≈ Pl (R1 ) ∗ Pl (R2 ), which implies
ploys a variety of PHY-diversity mechanisms (MIMO) for that all the points in the scatter plot should lie on diago-
improving network throughput. Maximum Ratio Combining nal. As shown in the figure, the points are indeed scattered
(MRC) is one such technique which optimally combines sig- very close to the line y = x, which indicates that the packet
nals from multiple antennas taking into account the signal- losses in 802.11n are largely independent. This suggests that
to-noise ratio (SNR) of the received signals. Mechanisms MAC-diversity can indeed be useful in improving the packet
employing spatial diversity can also be implemented at the delivery ratio (even in the presence of PHY-diversity). This
MAC layer [5] by combining frames from multiple receivers. is further confirmed by our experiments that quantify the
We term this technique as MAC-diversity. In this section gains from MAC-diversity in 802.11n. Next we describe our
we first inspect the statistical dependence of packet losses in results from the MAC-diversity experiments in detail.
802.11n MIMO receivers and then try to understand the im- Gains from MAC diversity: The idea of exploiting MAC
pact of PHY-diversity mechanisms on MAC-diversity. More level diversity was explored in [5], where authors showed
3
that packet losses are independent at co-located 802.11g re-
Please note that the I-factor model presented here provides ceivers. They also show that the losses on a single receiver
a rough approximation of the amount of interference due to are bursty in nature, which can be attributed to the short
presence of guard bands, pilots etc. in OFDM.
4 term channel fluctuations. We re-evaluate their findings in
We omit the other two cases due to space constraints
 1 0.8 1.5

Conditional Probability
P(R1i+k | R1i) 802.11n
0.7 P(R2i+k | R1i)

Throughput gain
0.8 1 802.11g
P(R1i+k | R2i)
P(R1)*P(R2)
0.6
0.6 P(R2i+k | R2i) 0.5
0.5
0.4 0
0.4
0.3 0.5
0.2
0.2 1 (low) (high)
0 0.1 1.5
0 0.2 0.4 0.6 0.8 1 1 10 100 1000 10000
Frame Loss Rate
P(R1 and R2) Number of Lag Packets (Logscale)
(a) (b) (c)

Figure 8: (a) Pl (R1 ∩ R2 ) ≈ Pl (R1 ) ∗ Pl (R2 ) indicates that the losses are largely independent across receivers
R1 and R2 . (b) Auto-conditional and cross-conditional packet error probabilities for broadcast experiments
performed using two 802.11n receivers. (c) Throughput gains when using MAC-diversity with two receivers
for 802.11n and 802.11g.

the context of 802.11n that already exploits diversity at the In this paper, we extend the I-factor model to incorporate
physical layer. In Figure 8(b) we plot the auto-conditional variable channel widths for characterizing the interference
and cross-conditional packet error probabilities. The auto- in 802.11n networks.
conditional probability denotes the probability that packet
i + k was lost at receiver R1 given that packet i was lost at 6. SUMMARY
receiver R1 . The cross-conditional probability on the other
Our work shows that the packet losses in 802.11n are in-
hand denotes the probability that packet i + k was lost at
deed independent at each receiver and that MAC-diversity
receiver R1 given that packet i was lost at receiver R2 . We
can still be beneficial in presence of the PHY-diversity mech-
draw two observations from the plot: (1) the conditional
anisms introduced by 802.11n. This has an important im-
probability of error decreases with increase in the lag pack-
plication that mechanisms exploiting spatial diversity (e.g.,
ets (k) and becomes constant (approaches the overall loss
MRD [5], ExoR [1]) can still provide high throughput gains
probability) for higher values of k. This clearly reflects the
for 802.11n links. We also observed that the throughput of
bursty nature of the losses as the conditional probability
an 802.11n link can severely degrade in presence of a lower
is higher than average for small values of k. Further, we
rate 802.11g link and that frame aggregation can help miti-
note that after 100 lag packets (k > 100), the losses become
gate this impact by providing temporal fairness among links.
completely independent of the previous packet loss. (2) the
Further, we identified that gains from MRC are likely to be
cross-conditional probability is much lower than the auto-
much higher in presence of interference and NLOS links.
conditional probability for smaller values of k. This shows
Finally, we extended the I-factor model to include variable
that using MAC-diversity can still be beneficial.
channel widths and show that our model can be used to
To quantify the performance gains that can be achieved
understand and characterize the nature of interference in-
with MAC-diversity, we implemented a naive algorithm which
troduced due to the partially overlapping 40 MHz channels
would combine packet receptions from two receivers to im-
in 2.4 GHz band.
prove the overall packet delivery ratio. Figure 8(c) shows Acknowledgments : All authors were supported in part
the throughput gains achieved at different locations for both by the US National Science Foundation through awards CNS-
802.11n and 802.11g. We observe that the gains for 802.11n 0639434, CNS-0627589, CNS-0627102, CNS-0520152, and
vary from 12% to as high as 103%, while the corresponding CNS-0747177. We would also like to acknowledge our shep-
gains for 802.11g reach upto 140%. It is important to note herd, Srihari Nelakuditi, whose comments helped bring this
that although similar loss rates were observed across both the paper into its final form.
802.11n receivers, the losses were actually independent lead-
ing to improvements in throughput due to MAC-diversity. 7. REFERENCES
[1] S. Biswas and R. Morris. Exor: opportunistic multi-hop
routing for wireless networks. In ACM Sigcomm, 2005.
5. RELATED WORK [2] Chachulski S. et al. Trading structure for randomness in
White papers from some of the companies have presented wireless opportunistic routing. In SIGCOMM, 2007.
experimental results on their 802.11n products [7, 6]. How- [3] M. Heusse, F. Rousseau, G. Berger-Sabbatel, and A. Duda.
ever, to the best of our knowledge, ours is the first exper- Performance anomaly of 802.11b. In IEEE INFOCOM 2003.
imental study quantifying the gains of 802.11n under di- [4] Mishra A. et al. Partially overlapped channels not
verse conditions, modeling the interference due to variable considered harmful. SIGMETRICS, 2006.
[5] Miu A. et al. Improving Loss Resilience with Multi-Radio
width partially overlapped channels and evaluating benefits
Diversity in Wireless Networks. In MOBICOM, 2005.
of MAC-diversity in presence of PHY-diversity offered by [6] F. Mlinarsky. Testing 802.11n. In Octoscope, 2007.
802.11n. Further, MAC-diversity has also been exploited by [7] P. Thornycroft. Designed for speed: Network infrastructure
prior research [5, 1] to achieve substantial throughput gains in an 802.11n world. In Aruba Networks white paper, 2007.
(upto 2.3x) over single radio systems. The I-factor model for
partially-overlapped channels was initially proposed in [4].