See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/305157994

Addressing MAC Layer Inefficiency and Deafness

of IEEE802.11ad Millimeter Wave Networks using
a Multi-Band Approach

Conference Paper September 2016

DOI: 10.13140/RG.2.1.3420.3122

CITATIONS READS

0 91

3 authors, including:

Gek Hong Sim Joerg Widmer

Technische Universitt Darmstadt Madrid Institute for Advanced Studies
10 PUBLICATIONS 7 CITATIONS 166 PUBLICATIONS 8,011 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Anticipatory Networking View project

D2D Millimeterwave Medium Access Control View project

All content following this page was uploaded by Gek Hong Sim on 11 July 2016.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
 Addressing MAC Layer Inefficiency and Deafness
of IEEE802.11ad Millimeter Wave Networks using
a Multi-Band Approach
Gek Hong Sim Thomas Nitsche
Technische Universitat Darmstadt, Germany. Fraunhofer ESK, Munich, Germany. Institute IMDEA Networks, Madrid, Spain.
AbstractAchieving multi-gigabit per second data rates,
millimeter wave communication promises to accommodate future
and current demands for very high speed wireless data trans-
mission. However, the mandatory use of directional antennas
brings significant challenges for the design of efficient MAC
layer mechanisms. In particular, IEEE 802.11ad for the 60 GHz
band lacks omni-directional transmissions and carrier sensing.
This prevents stations from overhearing the actions of other
stations, the so called deafness problem, which substantially
impairs the efficiency and fairness of CSMA/CA medium access.
Most existing solutions to this problem depend on properties of
lower frequency bands and thus do not apply to 60 GHz. In
this paper, we propose a dual-band MAC protocol combining
60 GHz communication with co-existing 5 GHz interfaces. By
broadcasting control messages on 5 GHz frequencies, we solve
the deafness problem and can use the 60 GHz band exclusively
for high rate data transmission. While our approach occupies
air time on the 5 GHz band for control messages, it does
achieve a net throughput gain (over both bands) of up to 65.3%
compared to IEEE 802.11ad. In addition, our simulation results
show an improvement of MAC fairness of up to 42.8% over
IEEE 802.11ad.
I. I NTRODUCTION
Communication in the 60 GHz and other millimeter-wave
(mm-wave) frequency bands (between 30 and 300 GHz) is a
highly promising evolution of WiFi and is considered a key
technology for 5G networks. However, strong signal attenu-
ation in this frequency range requires the use of directional
antennas, which in turn makes the design of efficient MAC-
layer protocols more difficult. As we show in this work,
reduced efficiency and fairness of carrier sense multiple access
with collision avoidance (CSMA/CA) MAC mechanisms are
the major draw-backs in this context.
The use of directional antennas for communication requires
accurately aligning the beams of transmitter and receiver. This
shift to directional communication and the lack of omni-
directional communication raise new challenges for the MAC
layer design of WiFi systems. In particular, the deafness
problem stations that are located outside of the transmitter
antennas boresight not being able to overhear or sense other
transmissions needs to be addressed.
The IEEE 802.11ad amendment for 60 GHz WiFi [1]
addresses these challenges by defining a hybrid medium access
mechanism which includes polling and time division multiple
access (TDMA) in addition to the well known CSMA/CA
[2]. However, polling introduces overhead at all stations, and
TDMA requires additional coordination and communication
Joerg C. Widmer
to determine the schedule. In contrast, CSMA/CA is very
efficient in handling unpredictable bursts of traffic, such as
traffic generated by web browsing. Hence, adaptation of
the CSMA/CA scheme to the constraints imposed by mm-
wave communication is highly relevant. In its basic form,
CSMA/CA suffers from several drawbacks when applied to
directional communication systems, the main challenge being
incomplete carrier sensing. Further, erratic deferral behavior
and increased collisions lead to reduced fairness in terms of
channel access. While long term fairness is still achieved,
a group of active nodes may dominate the channel whereas
long time inactive users have a low probability of successfully
accessing the medium.
Even though several works have addressed deafness in
WLAN and WPAN [3], most solutions are designed for
lower frequency communication where omni-directional com-
munication for coordination purposes is feasible. At mm-
wave frequencies, however, increased attenuation requires a
directional antenna configuration at least at one side of a
communication link. The most suitable adaptations of CS-
MA/CA for mm-wave frequencies are proposed by IEEE
802.11ad [1] and work by Gong et al. [4]. With IEEE 802.11ad
CSMA/CA, a data exchange between two nodes is protected
with a directional RTS/CTS exchange, which prevents stations
with aligned antenna beams from creating interference as
they overhear messages from either transmitter or receiver.
However, as messages are unlikely to be overheard by nodes
with antenna beams in other directions, these nodes do not
defer during ongoing transmissions but unsuccessfully try
to access the channel and then excessively increase their
contention window. A different approach is proposed in [4],
where CTS messages are broadcasted omni-directionally by
the AP. To achieve sufficient link budget to receive the omni-
directional CTS messages, every station by default directs its
receive beam towards the AP. Unfortunately, this approach
still suffers from colliding directional RTS messages, which
lowers the effectiveness of the deferral process. As a result,
both techniques leads to a fairness problem as stations that
successfully access the channel have a substantially higher
chance to subsequently win the contention again.
In this paper, we use a multi-band approach to address
the deafness problem, which severely impacts fairness and
efficiency of uplink transmissions. Use of multiple bands in
high speed mm-wave WiFi is common, due to easily blocked
 2

directional links and limited range, requiring a fallback mech- ordinator (PNC) in a WPAN to coordinate each transmission.
anism to more resilient lower frequency communication. In In addition, this technique also prevents spatial reuse since
fact, the IEEE 802.11ad standard specifies Fast Session Trans- a busy PNC cannot coordinate new transmissions. The IEEE
fer functionality to transition between multiple bands [1]. 802.11ad amendment itself proposes a directional MAC layer
As a consequence, we expect most devices following IEEE mechanism but does not address the deafness problem that
802.11ad and following mm-wave WiFi standards to be com- leads to a critical fairness issue. We discuss the details of this
patible with our approach. mechanism in Section III. Similar to our approach, [14][16]
Our solution maintains high fairness through a combination exploit the coexistence of a microwave interface. However,
of 60 GHz communication with control messages on legacy there the dual-band approach is used for neighbor discovery, as
WiFi frequencies. When contending for the channel, stations a fall-back mechanism for range extension, instead of solving
exchange omni-directional RTS/CTS messages on the lower the deafness problem.
frequency band and then exchange data frames on the 60 GHz III. FAIRNESS I MPAIRMENTS IN IEEE 802.11 AD
band. The advantage of our approach is twofold. First, control D IRECTIONAL CSMA/CA
message exchange on frequencies of 2.4 or 5 GHz is highly
This section describes the impact of deafness on the
reliable over the typically short IEEE 802.11ad communication
popular CSMA/CA mechanism in directional IEEE 802.11ad
distances. Thus, every station overhears the control messages
networks.The deafness problem entails that in many cases
and can correctly defer, avoiding the IEEE 802.11ad unfairness
other stations will neither overhear frames nor sense that the
problem. Second, by parallelizing control and data transmis-
carrier is busy when two stations are communicating, causing
sion, we free resources on the 60 GHz band for high speed
the following two important performance impairments.
data transmission. RTS/CTS control messages that are used by
Excessive Backoff. Due to limited (or lacking) carrier sensing,
default on the 60 GHz band are no longer necessary with our
the frame collision probability during contention increases.
approach. Thus, this dual band approach enhances throughput
Especially in dense networks, this in turn significantly in-
and MAC efficiency. To the best of our knowledge, this paper
creases the average contention window. Furthermore, as frames
is the first to study and analyze the fairness for multi-band
(including RTS/CTS) are transmitted directionally, stations
WiFi networks operating in the mm-wave band.
outside the transmit beam will not overhear the ongoing com-
Our paper is organized as follows. In Section II we discuss
munications and thus will not defer. Instead, a deaf station may
past proposals to solve the deafness problem. Sections III
try to transmit to an already communicating station, which has
describes impairment on the CSMA/CA mechanisms due to
its receive antenna beam steered into another direction. While
the deafness problem. Our dual-band CSMA/CA solution is
this will not necessarily disrupt the ongoing communication,
described in Section IV. Section V provides the channel and
the deaf station will increase its contention window due to
transmission model we use for the performance evaluation.
the failed transmission. Fig. 1 depicts this excessive backoff
Simulation results are presented in Section VI followed by
problem. Stations two and three initially collide with their
Section VII which concludes the paper.
RTS messages, resulting in increased contention windows
II. R ELATED W ORK compared to station 1. Both have reduced chances to hit
the following contention windows. In the example, their next
As for mm-Wave communication, directional antennas are
RTS messages are lost since the AP directs its receive beam
also used in microwave communication, e.g., the 2.4 GHz and
away from them due to an ongoing data transmission. Stations
5 GHz frequency bands, to improve throughput and reduce in-
two and three further increase their contention window, and
terference. Also here, the directionality of the communication
station one dominates the medium access.
can cause deafness. Solutions to this problem are discussed in
Unbalanced Contention. Stations suffering excessive backoff
[5][12]. [5] and [6] solve the deafness problem for CSMA/CA
have a low probability to win contention, which favors recently
systems by omni-directionally transmitting control messages
active stations with smaller contention windows. While this
(i.e., RTS and CTS). However, by doing so, all overhearing
effect is also present in conventional IEEE 802.11, it is vastly
nodes will defer their transmission, which impacts spatial reuse
exacerbated in IEEE 802.11ad since the contending stations do
in ad-hoc networks. Korakis et al. [9] propose a directional
not know when an ongoing frame ends, i.e., when to resume
system that emulates omni-directional RTS transmission by
contention. This in turn increases the probability that the same
sweeping transmit directions, i.e., transmitting an RTS in each
active station transmits a high number of frames consecutively,
possible direction. While this technique effectively deals with
before another station happens to win the contention for
deaf stations, it has high overhead, especially for systems with
medium access. Note that this effect primarily impacts short
many narrow directional sectors. The works in [10], [11] ad-
term fairness. Over longer time scales, the active station
dress the deafness problem by omni-directionally transmitting
changes sufficiently often to compensate the frame bursts of
control messages but they require additional mechanisms to
active stations and achieve some level of fairness.
prevent interfering transmissions.
Only few works address the deafness problem while taking IV. D UAL -BAND CSMA/CA
this additional challenge on directional transmission into con- In this section, we propose a dual-band CSMA/CA scheme
sideration. Gong et al. [4], [13] use the personal network co- that mitigates the deafness problem for uplink communication.

CTS1
AP
Backoff SIFS SIFS
RTS1
STA1
RTS2
SIFS CTS Timeout DIFS
STA2
Backoff
RTS3
SIFS CTS Timeout DIFS
STA3
t1
Fig. 1. Excessive backoff behavior of CSMA/CA in IEEE802.11ad
Our approach leverages omni-directional transmissions on
legacy WiFi band to coordinate high throughput transmissions
in the 60 GHz band.
We assume an IEEE 802.11ad compatible transceiver de-
sign and an infrastructure based network architecture with an
AP. Further, we require all stations to be able to communicate
over a 60 GHz and legacy WiFi interfaces. This type of
transceiver architecture is very likely, as IEEE 802.11ad makes
use of a multi-band fast session transfer (refer [1]) for range
extension and seamless failover in case of link breaks. Thus,
we expect typical IEEE 802.11ad devices to comply with the
requirements of our dual-band CSMA/CA scheme. For sim-
plicity we omit details about beam training on the directional
60 GHz interface and assume pre-trained directional links for
all stations to the AP. In general, this assumption is satisfied by
the association beam training process described in [1]. Dual-
band CSMA/CA access can then be enabled as an addition to
the IEEE 802.11ad hybrid MAC architecture.
A. Dual-Band CSMA/CA Protocol
Our dual band CSMA/CA protocol follows the random
backoff and deferral mechanism of IEEE 802.11ad. How-
ever, the contention mechanism together with the RTS/CTS
exchange occur on the omni-directional legacy WiFi band.
The IEEE 802.11ad interface is exclusively used for data
transmission and acknowledgments.
When applying the contention mechanism on legacy WiFi
interfaces, only one message can be exchanged at a time.
Thus, for our approach, we assume to have data frame sizes
that exceed the duration of the RTS/CTS exchange on the
lower frequency. This is a valid assumption since aggregation
of data frames is beneficial (to reduce overhead time) for
transmissions at very high data rate, i.e., 60 GHz transmission.
When the aggregation level is low, it is possible to use in-
band RTS/CTS as in conventional IEEE 802.11ad for small
data frames for which dual-band RTS/CTS creates too much
overhead. The detailed on the switching between in-band and
dual-band RTS/CTS is out of the scope of this paper.
When frames are transmitted according to the dual-band
mechanism, stations need to sense the lower frequency band
to be idle for at least a DIFS time before starting the contention
mechanism. The transmitted RTS frame will reference the end
of the latest known transmission on the 60 GHz band plus
a SIFS interval. Receiving RTS and CTS messages omni-
directionally ensures that the corresponding transmission is
3
Unsuccessful RTS
ACK1 CTS1
SIFS DIFS Backoff SIFS SIFS
DATA1 RTS1 DATA1
Backoff
RTS2
SIFS CTS Timeout DIFS Backoff
Backoff
RTS3
SIFS CTS Timeout DIFS Backoff
Excessive backoff
known to all stations. Note that the RTS/CTS exchange can
already occur during the transmission of some previous data
frame on the 60 GHz band to avoid unnecessary delay.
Fig. 2 shows an example frame flow for our dual-band
approach. Three stations (distinguishable by the subscript in
frame descriptions) intend to transmit a frame to the AP
at the same time. In contrast to deaf IEEE 802.11ad CS-
MA/CA, backoff happens on the legacy frequency band and
the RTS/CTS messages are overheard. As can be seen from
frames Data1 and RT S2 , data frame transmission and backoff
procedure happen in parallel.
B. Improvements to Millimeter Wave CSMA/CA
A fundamental advantage of our dual-band approach is
that all stations can overhear the RTS/CTS exchange on lower
omni-directional frequencies, thus solving the deafness prob-
lem. As a result, dual-band CSMA/CA does not suffer from the
impairments described in Section III. Our approach achieves
a flawless deferral behavior, avoids excessive contention win-
dows, and increases the fairness of medium access.
A second benefit results from the parallelization of RT-
S/CTS exchanges and data transmissions on two separate
frequency bands. This removes idle time and RTS/CTS trans-
missions from the 60 GHz frequency band. As RTS/CTS
exchanges are transmitted with the most robust and thus lowest
rate coding and modulation scheme, inefficient use of the
60 GHz channel is avoided. Instead, all available channel
time with only a SIFS interval between data frames (and
acknowledgements) can be utilized for very high throughput
transmissions on the 60 GHz channel. Sacrificing transmission
time on the legacy WiFi band for control traffic improves
efficiency as the band supports much lower transmit rates
compared to the 60 GHz band.
V. S IMULATION M ODELS
This section discusses the simulation model, implemented
on an event-based MATLAB simulator, that is used to produce
the numerical results in Section VI. We consider an indoor
wireless network with a single access point (AP) and a set
of S non-AP stations that are randomly distributed within the
coverage area of the AP. Let {s, d} be a transmission pair,
where s and d denote the source and destination stations,
respectively.
Channel model. The received power at d is
Pr (s, d)(dBm) = Pt (dBm) + Gs (dBi) + Gd (dBi) PL(ls,d ),

CTS1 CTS2
WiFi
AP
60GHz
SIFS DIFS Backoff
RTS1 RTS2
WiFi
STAs
SIFS
60GHz DATA1
Fig. 2. Channel access mechanism of the Dual-band approach
where Gs and Gd are the antenna gains at s and d, respectively.
The path loss PL(ls,d ) of stations that are ls,d apart is
4ls,d (m)
PL(ls,d ) = 20 log10 + ls,d
as presented by Zhu et al. [17]. The oxygen absorption is =
0.02 dB/m and the wavelength at 60 GHz frequency band
is 5 mm. Pt is the transmit power and ranges from 10 dBm
to 27 dBm, the maximum value specified by IEEE 802.11ad.
The maximum equivalent isotropically radiated power (EIRP)
permitted by the FCC for IEEE 802.11ad is 40 dBm.
Interference model. Fig. 3 illustrates the interference sce-
nario in a directional network. A station can only interfere with
a receiving station if it is located within the antenna boresight
of the receiving station. For instance, when the AP is receiving
data from STA1 in Fig. 3, it beamforms to STA1, therefore,
STA2 is a potential interferer but not STA3. In the presence
of interference, the total interference power Pint (s, d) at the a
receiving station d is
X aRTSTime = aCTSTime
Pint (s, d) = Pr (n, d). (1)
n2{S[AP,n6=d,n6=s}
AP
STA1
STA2
Fig. 3. Interference in a directional network
VI. R ESULTS
This section evaluates the performance of our Dual-band
scheme. We investigate the impact of the number of stations
Ns , time interval over which fairness is measured called
fairness interval, and the frame size f . We consider an uplink
scenario, where all stations contend for the channel to transmit
data frames to the AP. Since all stations are attempting to
transmit to the AP, their beams are always directed towards
the AP. However, only one station at a time can win a transmit
opportunity (TXOP) with the AP. Upon winning a TXOP, a
station uses it to transmits a maximally sized data frame that
fits the TXOP before recontending for the channel.
Network topology and configuration. In the simulation,
stations are randomly distributed within the coverage area of
the AP of radius 23 m. This range is the same as the maximum
range we measured on first generation mm-wave devices, a
4
CTS3
ACK1 ACK2
DIFS Backoff
RTS3
SIFS SIFS SIFS SIFS SIFS
DATA2 DATA3
Dell 6430u laptop communicating with a D5000 docking sys-
tem with 13 sector width. The fairness interval over which
fairness is evaluated is chosen to reflect short term ( = 5
(dB) ms) and long term fairness ( = 80 ms), respectively. We
also use different frame sizes f = {1.5, 15, 30, 45, 60, 75}KB
to study the impact of different levels of frame aggregation
on performance. For data frame transmission at 60GHz, we
consider the 12 single carrier MCSs defined in IEEE 802.11ad
[1]. The corresponding transmit rates range from 389 Mbps
to 4620 Mbps. Table I shows the control message transmis-
sion rate for 802.11ad and 802.11ac respectively as well as
simulation settings.
TABLE I
PARAMETERS IN 60GH Z AND 5GH Z FREQUENCY BANDS .
Item
IEEE 802.11ad IEEE 802.11ac
60GHz 5GHz, 80MHz bandwidth
aDIFSTime 13s 34s
aSIFSTime 3s 3s
aSlotTime 5s 9s
MCS0 27.5Mbps 32.5Mbps
8.19s 7.30s
Performance metrics. The main performance metrics are
throughput, fairness, and delay. Throughput is the total amount
of data bits successfully received at the destination station
over the transmission time. To ensure fair comparison, we
STA3
take the throughput loss on the lower frequency band that
our scheme incurs into account. We scale the lower frequency
bands throughput rate by the amount of channel time not used
for RTS/CTS exchanges and add this value to the rate achieved
at mm-wave frequencies to obtain total throughput. For non
dual-band schemes, the full data rate of the lower frequency
bands is added, which is fixed to 433 Mbps, the maximum rate
supported by IEEE 802.11ac without MIMO. Lastly, fairness
is computed based on Jains fairness index [18].
Performance comparison. We compare the performance
of the proposed approach against two other schemes. The
first approach is the IEEE 802.11ad standard [1] where all
messages are transmitted directionally between the source and
the destination stations. The second approach is the centrally
coordinated scheme (denoted Central) of Gong et al. [4],
where an RTS message is transmitted directionally from the
source to the AP, which replies with a broadcast CTS message
to all stations in the system.
We compare throughput, fairness, and the impact of frame
size of the different methods in a homogeneous scenario where
all stations have the same sector width of 30 .
Throughput. Fig. 4 and Fig. 5 illustrate the impact of the
number of stations Ns on frame collision rate and system
 5

throughput, respectively. Both simulations have a fixed fairness

Data time MAC overhead time Ilde time Collision time

80
interval of = 80 ms and frame size of f = 15 KB.

Time interval (ms)

In general, Fig. 4 shows that increasing Ns also increases 60

the probability of RTS frame collision. While Central avoids 40

collisions with data frames due to broadcasted CTS messages 20

by the AP, directional RTS frames during the contention phase
0
still can collide since the directional RTS may not be overheard 1 4 16 1 4 16 1
Central(CWmin=15) Central(CW =31) Central(CW =63)
4 16 1 4 16
IEEE 802.11ad
1 4 16
Dualband
by other stations. Further, Central with higher CWmin has a
min min
Number of stations, Ns

lower number of RTS collisions. For IEEE 802.11ad, the curve Fig. 6. Time proportion for data transmission, MAC overhead, idle time and
denoted IEEE 802.11ad only shows collisions of RTS frames, collision time for Ns = {1, 4, 16} in the homogeneous sector scenario.
while IEEE 802.11ad (All collisions) also shows collisions
100
with CTS and data frames. The plain RTS-RTS collision
100

90
rate is comparable to the centralized scheme, however the 95

Fairness index (%)

Fairness index (%)

80
overall collision rate of IEEE 802.11ad is much higher. In 90
70
fact, the majority of the collisions in IEEE 802.11ad occur 85
60
due to the transmission of RTS messages from the deaf
Central(CWmin=15) Central(CWmin=15)
Central(CWmin=31) Central(CWmin=31)
50
stations during an ongoing data transmission. Dual-band in
Central(CWmin=63) 80 Central(CWmin=63)
IEEE 802.11ad IEEE 802.11ad
40
contrast has the lowest RTS-RTS collision rate since RTSs
Dualband Dualband
75
1 2 4 8 16 32 1 2 4 8 16 32
are transmitted omni-directionally and thus stations defer upon Number of stations, Ns Number of stations, Ns

overhearing them. Collisions only occur if two RTS messages Fig. 7. Short term fairness. Fig. 8. Long term fairness.
are transmitted in the exact same slot.
Central(CWmin=15) 3 Central(CWmin=15)
Ns because fewer stations increases the channel contention
1000 Central(CWmin=31) Central(CWmin=31) probability. Further, Fig. 7 and Fig. 8 show that Central and
Number of RTSRTS collisions

Central(CWmin=63) Central(CWmin=63)

IEEE 802.11ad have significantly lower fairness than Dual-

Total throughput (Gbps)

800 IEEE 802.11ad 2.5 IEEE 802.11ad

IEEE 802.11ad (All collisions) Dualband

600
Dualband
band. This is because increased RTS collisions lead to larger
2

400
contention windows. Thus, stations that experience collisions
1.5 have a much lower chance of successfully contending for
the channel.In particular, when many stations contend for the
200

channel, most of the stations excessively increase their backoff

0 1
1 2 4 8
Number of stations, N
Fig. 4. RTS-RTS collisions
From Fig. 5 we see that throughput increases with Ns .
This is due to the reduction of time between two consecutive
transmissions when more stations contend for the channel.
Further, increasing the minimum contention window CWmin
reduces the probability of collision but also causes throughput
loss for lower Ns as stations back off unnecessarily, as can be
seen for the three different CWmin configurations.
Fig. 6 presents the duration for data transmission, MAC
overhead, collision time and idle time. The Central schemes
with higher CWmin incur more idle time than those with a
lower CWmin . Dual-band has increased idle time between
frames, which coincides with the lower throughput for Ns = 1.
This is due to longer DIFS and slot time on the lower fre-
quency band (see Table I). Therefore, the backoff time is 80%
longer than that for IEEE 802.11ad and Central. For larger Ns ,
the benefits of Dual-band outweighs increased inter-frame and
slot times. Further, collision reduction leaves more time for
actual data transmission. Remarks: Dual-band performs well
for Ns > 1 achieving throughput gain of up to 65.3% against
IEEE 802.11ad and 57.9% against Central.
Fairness. Fig. 7 and Fig. 8 show the short-term ( = 5 ms)
and long-term ( = 80 ms) fairness of the schemes. For both,
the frame size is fixed to f = 15 KB and the number of
stations Ns is varied between 1 and 32.
In general, fairness is found to be higher for smaller
16 32 1 2 4 8 16 32
Number of stations, N
s s
interval as the successful contending station repeatedly wins
Fig. 5. Throughput comparison
TXOPs with its minimum contention window.
Somewhat surprisingly, IEEE 802.11ad performs almost as
well as Central in terms of fairness. This is due to the fact that
deaf stations continue reducing their backoff timer during on-
going packet transmissions. They thus have a chance to access
the channel upon the expiry of the backoff time in a random
manner, even when their initial backoff time was very large.
In contrast, with the Central scheme, stations defer due to the
overheard CTS and resulting in a long time to count down
excessively long backoff timers. We also find that the initial
contention window size for Central impacts fairness. Smaller
minimum windows result in reduced fairness, as successfully
contending nodes have shorter contention times.
For the long term fairness analysis with = 80 ms, higher
fairness is found for all schemes, since even a station with
large backoff interval has sufficiently many opportunities to
contend for the channel and eventually will be able to transmit.
Fig. 8 further reveals that IEEE 802.11ad even outperforms
the Central scheme for Ns 32. This due to the deferring
mechanism discussed above.
Further, Fig. 9 and Fig. 10 show the histogram of the aver-
aged maximum per frame delay for Ns = 4 and Ns = 16, over
200 simulation runs. The frame size for these simulations is
fixed to f = 15 KB and the fairness interval is = 80 ms. The
delay duration is the time difference between two successful
transmissions (given that all stations are backlogged). From
 6

Fig. 9 and Fig. 10 IEEE 802.11ad has a higher distribution of band due to broadcasting of control messages, our approach
longer delay for Ns = 4 than that for Ns = 16, as compared to still achieves significant overall throughput gain up to 65%
Central. This is also due to the deferring behaviour of Central and 61.8% over IEEE 802.11ad and the centralized scheme,
when the backoff time is long. respectively.
ACKNOWLEDGEMENTS
80 100
Central(CWmin=15) Central(CWmin=15)
IEEE 802.11ad 80 IEEE 802.11ad
60 Dualband Dualband
This research was funded by the European Research Coun-
Frequency

Frequency
60
40 cil grant ERC CoG 617721, the Ramon y Cajal grant from
40

20
0
0 0.01 0.02 0.03 0.04 0.05
Maximum per frame delay (ms)
Fig. 9. Maximum per frame delay for
Ns = 4.
Impact of Frame Size. Fig. 11 and Fig. 12 depict the impact
of frame size for short term fairness ( = 5 ms) and long term
fairness ( = 80 ms), respectively. The number of stations
for both simulation runs is fixed to Ns = 16. For all of the
schemes, fairness decreases as the frame size increases. For
IEEE 802.11ad, this results from the fact that larger frame
size entails longer transmission time and multiple RTSs of the
same station may remain unanswered during one data frame,
which again leads to excessive backoff intervals. Similarly, for
Central with increasing frame length, stations defer for longer
durations and thus some stations may not be able to transmit
within the fairness interval. For long-term fairness, shown in
Fig. 12, the chances of a station to transmit increases, thus
resulting in a higher fairness for all schemes.
100
Central(CWmin=15)
90 Central(CWmin=31)
Central(CWmin=63)
Fairness index (%)
the Spanish Ministry of Economy and Competitiveness RYC-
20
2012-10788, the Madrid Regional Government through the
0.06
0
0 0.02 0.04 0.06 TIGRE5-CM program (S2013/ICE-2919), the LOEWE initia-
Maximum per frame delay (ms)
tive (Hessen, Germany) within the NICER project.
Fig. 10. Maximum per frame delay
Ns = 16.
R EFERENCES
[1] IEEE, Wireless LAN medium access control (MAC) and physical
layer (PHY) specifications amendment 3: Enhancements for very high
throughput in the 60 GHz band, IEEE Std 802.11ad-2012, Dec 2012.
[2] T. Nitsche, C. Cordeiro, A. Flores, E. Knightly, E. Perahia, and J. Wid-
mer, IEEE 802.11ad: directional 60 GHz communication for multi-
gigabit-per-second wi-fi, IEEE Communications Magazine, vol. 52,
no. 12, pp. 132141, Dec 2014.
[3] S.-K. Yong, P. Xia, and A. Valdes-Garcia, 60GHz Technology for Gbps
WLAN and WPAN: from Theory to Practice. John Wiley & Sons, 2011.
[4] M. Gong, R. Stacey, D. Akhmetov, and S. Mao, A directional CS-
MA/CA protocol for mmWave wireless PANs, in IEEE WCNC, Apr.
2010.
[5] Y.-B. Ko, V. Shankarkumar, and N. Vaidya, Medium access control
protocols using directional antennas in ad hoc networks, in IEEE
INFOCOM, Mar 2000.
[6] A. Nasipuri, S. Ye, J. You, and R. Hiromoto, A MAC protocol for
mobile ad hoc networks using directional antennas, in IEEE WCNC,
Sep 2000.
[7] M. Takai, J. Martin, R. Bagrodia, and A. Ren, Directional virtual carrier
100 sensing for directional antennas in mobile ad hoc networks, in ACM
95
MobiHoc, Jun 2002.
[8] J. Qiao, L. Cai, X. Shen, and J. W. Mark, Enabling multi-hop

Fairness index (%)

90
80 IEEE 802.11ad
Dualband
concurrent transmissions in 60 ghz wireless personal area networks,
70 85 IEEE Transactions on Wireless Communications., vol. 10, no. 11, pp.
80 Central(CWmin=15)
38243833, Nov 2011.
60
Central(CWmin=31) [9] T. Korakis, G. Jakllari, and L. Tassiulas, A MAC protocol for full
75
50 Central(CWmin=63) exploitation of directional antennas in ad-hoc wireless networks, in
40
70 IEEE 802.11ad
Dualband
ACM MobiHoc.
65 [10] A. Arora, M. Krunz, and A. Muqattash, Directional medium access
1.5 15 30 45 60 75 1.5 15 30 45 60 75
Frame sizes, f (KB) Frame size, f (KB) protocol (DMAP) with power control for wireless ad hoc networks, in
IEEE GLOBECOM, Nov 2004.
Fig. 11. Impact of frame size on Fig. 12. Impact of frame size on long
[11] D. Han, J. Jwa, and H. Kim, A dual-channel MAC protocol using direc-
short term fairness ( = 5ms) term fairness ( = 80ms)
tional antennas in location aware ad hoc networks, in Computational
Science and Its Applications ICCSA, ser. Lecture Notes in Computer
VII. C ONCLUSION Science, Jan 2006, no. 3983, pp. 594602.
In this paper, we address the deafness problem that affects [12] Y. Wang, H. Garg, and M. Motani, Directional medium access control
for ad hoc networks: A cooperation-based approach, in IEEE ICON,
throughput and fairness for directional 60 GHz transmissions Dec 2013.
with CSMA/CA channel access. To the best of our knowledge, [13] M. Gong, D. Akhmetov, R. Want, and S. Mao, Multi-user operation in
we are the first to design and analyze a dual frequency system mmwave wireless networks, in IEEE ICC, Jun 2011.
[14] J. Qiao, X. Shen, J. Mark, Z. Shi, and N. Mohammadizadeh, MAC-layer
to mitigate the deafness problem in mm-wave systems. Our integration of multiple radio bands in indoor millimeter wave networks,
approach takes advantage of a coexisting 5 GHz interface to in IEEE WCNC, Apr. 2013.
coordinate the data transmission at 60 GHz. This is beneficial [15] H. Park, Y. Kim, T. Song, and S. Pack, Multi-band directional neighbor
discovery in self-organized mmwave ad-hoc networks, IEEE Transac-
in two ways. First, due to omni-directional transmission on tions on Vehicular Technology, vol. PP, no. 99, Jun. 2014.
5 GHz the deafness problem is solved. Second, moving robust [16] T. Nitsche, A. B. Flores, E. W. Knightly, and J. Widmer, Steering with
low rate control messages to 5 GHz allows to use the 60 GHz eyes closed: mm-wave beam steering without in-band measurement, in
IEEE INFOCOM, Apr 2015.
band exclusively for high throughput data transmission. We an- [17] Y. Zhu, Z. Zhang, Z. Marzi, C. Nelson, U. Madhow, B. Y. Zhao, and
alyze throughput and fairness of our proposed scheme through H. Zheng, Demystifying 60GHz outdoor picocells, in ACM Mobicom,
extensive simulations to compare it against IEEE 802.11ad and Sep 2014.
[18] R. Jain, A. Durresi, and G. Babic, Throughput fairness index: An
a centralized scheme called central. In terms of fairness, we explanation, Tech. rep., Department of CIS, The Ohio State University,
improve by up to 42.8% over IEEE 802.11ad and 34.5% over Tech. Rep., Feb 1999.
the centralized scheme. Despite using air time on the 5 GHz