Directional Antenna Diversity For Mobile Devices: Characterizations and Solutions
Directional Antenna Diversity For Mobile Devices: Characterizations and Solutions
Directional Antenna Diversity For Mobile Devices: Characterizations and Solutions
General Terms
mobile health monitoring, and video chat, are increasingly employing the uplink. Therefore, the mentioned interference has
become a key bottleneck in the capacity of infrastructure networks with a large number of mobile clients. Similarly, the power
waste also makes wireless transmission one of the most powerhungry activities on mobile devices, contributing a critical barrier
to the usability and wider adoption of mobile Internet. Furthermore, devices that are far away from access points or base stations, also suffer from low uplink data rates since most of their
limited transmission power is radiated to undesirable directions.
Directional transmission can alleviate these problems by focusing
the radiation toward the intended receiver. Beamforming employs
multiple RF chains to achieve directionality. However, its cost
and power overhead has prevented it from being immediately
used on smartphone-like mobile devices. In this work, we study a
much more efficient way of realizing directionality: miniature
passive directional antennas. Many authors use directional antenna to refer to beamforming based on smart antennas. In this
work, we use antenna to refer to the passive antenna without
the RF chain. By placing multiple directional antennas on the
surface of a mobile device, the device can opportunistically select
one for directional transmission without adding RF circuitry, in
contrast to the simultaneous use of multiple RF chains by beamforming. However, because only one antenna can be used at a
time, the device will not be able to find out the best antenna easily, especially when it moves and rotates.
The goal of this work is to realize the benefits of directional antennas for mobile devices. Toward this goal, we experimentally
answer three questions.
First, how do smartphone-like mobile device rotate during wireless access? By using a directional antenna, the device orientation
becomes critical because a mobile device can rotate and the rotation changes device direction much faster than mobility does. We
collect accelerometer and compass readings along with network
usage information from 11 smartphone users, each for one week
in the field. From such field-collected traces, we are able to estimate the orientation and rotation of the smartphones during wireless usage. We show that smartphones rotate relatively slowly;
compared to how fast packets are exchanged during wireless
communication. Moreover, the orientation is quite predictable in
short intervals. We report the characterization in Section 3.
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
MobiCom10, September 2024, 2010, Chicago, Illinois, USA.
Copyright 2010 ACM 978-1-4503-0181-7/10/09...$10.00.
Keywords
Passive directional antennas, mobile devices, orientation estimation, smartphone rotation
1. INTRODUCTION
trolled orientations, but also rotation according to the fieldcollected traces. We show that directional antennas outperform
omni ones for a considerable range of orientations even in NLOS
indoor environments. More importantly, the directional channels
are highly reciprocal for 802.11-like frequency bands, and their
performance is quite predictable in short intervals even under
realistic rotations. We report the characterization in Section 4.
Finally, how can a device dynamically select the best antenna?
Leveraging the discoveries from the two characterizations, we
design a multi-antenna system (MiDAS) that consists of an omni
directional antenna, one or more directional antennas, and an
antenna switch. MiDAS works with existing mobile devices that
usually have a single RF chain and can use only one antenna at a
time. We further provide two antenna selection methods for MiDAS. The packet-based method uses one packet to assess an antenna without any changes to the network infrastructure. The
symbol-based method uses the PHY training symbols so that all
antennas are assessed with a single packet. It is much more efficient than the packet-based selection but requires changes to the
PHY layer. We report the design of MiDAS in Section 5 and the
two antenna selection methods in Sections 6 and 7, respectively.
We implement MiDAS and its antenna selection methods on the
WARP platform from Rice University and evaluate them with the
computerized motor platform that replays the field-collected rotation traces from smartphone users. Our evaluation shows that
with three 5dBi directional antennas placed on the surfaces of a
device, MiDAS can achieve a median gain close to 3dB, without
any changes to the network infrastructure.
To fully realize the benefits of the link gain of MiDAS, we incorporate rate adaptation and power control mechanisms into our
system. Using real-time experiments, we demonstrate that MiDAS can improve the link goodput by 85% when the SNR is low.
Also, when the SNR is high, it can save 51% of the transmit power, while increasing the goodput by 7%. We report the prototype
and evaluation in Section 8.
To the best of our knowledge, this work represents the first publicly available characterizations of the rotation of smartphone-like
devices and its impact on the performance of directional antennas.
MiDAS and its antenna selection methods are also the first reported directionality implementation for a device that cannot only
move but can also rotate. We will discuss related work in Section
9.
Making smartphone-like devices directional is a radical departure
from existing and emerging wireless technologies. It provides an
inexpensive and immediately deployable solution to improve
network capacity and device efficiency. While we demonstrate
the effectiveness of MiDAS in improving link goodput and device power efficiency, more network support is needed to fully
realize its potential in improving network capacity. From this
perspective, our work complements existing work on directional
MAC protocol design, e.g. [3].
2. DIRECTIONAL ANTENNAS
We are motivated by the commercial availability of miniature
passive directional antennas that can be employed on mobile
devices. Microstrip antennas are good examples. The microstrip
antennas used in our prototype have a patch area of 3.23.2 cm2
[4]. This allows the placement of these antennas on front and
back sides of smartphones. Also theoretically, the width of the
patch can be even smaller while maintaining 5dBi or 8dBi peak
3. CHARACTERIZING SMARTPHONE
ROTATION
We first characterize the orientation and rotation of smartphones
during wireless access with data collected from field usage. It is
important to note that the relative direction of a device with its
access point (base station) is determined by not only device orientation but also location. We do not include location in our characterization because mobile client location and its changes have
been extensively studied as mobility. Moreover, change in relative direction due to mobility is much slower than that due to
rotation thanks to the large distance between a mobile device and
its access point or base station.
z Sky
1
Device X axis
Device Z axis
y Magnetic
North
Ground
0.5
0
0.1s
1s
10s
0.5
0.5
0
-4
10
-2
10
10
Rotational speed(/s)
0.1s
1s
10s
0
-4
10
10
-2
10
10
-4
10
Rotational speed(/s)
10
-2
10
10
10
Rotational speed(/s)
Figure 2: Cumulative Distribution Function (CDF) of rotation speed for three Euler angles, calculated for 0.1s, 1s,
and 10s intervals
Zero order
First order
()
Device Y axis Y
0.1s
1s
10s
100
1
0.01
10ms 100ms
1s
10s
Prediction Interv.(s)
Zero order
First order
Zero order
First order
100
100
1
0.01
0.01
10ms 100ms
1s
10s
Prediction Interv.(s)
10ms 100ms
1s
10s
Prediction Interv.(s)
4. CHARACTERIZING DIRECTIONAL
PROPAGATION
In this section, we report experimental characterization of the
quality of directional links under various propagation environments, orientation, and rotation. While a few recent works have
reported NLOS indoor performance of directional links [1, 2], no
one has characterized NLOS indoor directional propagation under
device rotation.
Dir-Client
Dir-AP
Omni-Client
Omni-AP
RSS(dBm)
RSS(dBm)
-20
-30
-40
-50
-60
0
data packet, and the motor-driven node records the RSSI for
every ACK packet sent back from the AP.
All indoor experiments are carried out on the third floor of Duncan Hall, an office building at Rice University. Outdoor experiments are carried out in an open space outside Duncan Hall. During the experiments, the client and AP nodes are separated as
allowed by the WARP nodes, ~15 meters in most cases. For indoor, experiments are carried out at various locations in the building with and without light-of-sight propagation paths, or LOS
indoor and NLOS indoor. The distance between the AP and client
in our indoor experiments is comparable to such distance in an
enterprise 802.11 network. Indoor environments have been
known to be much more challenging for directional communication than outdoor LOS propagation.
-40
Dir-Client
Dir-AP
Omni-Client
Omni-AP
-50
-60
0
RSS(dBm)
RSS(dBm)
Figure 4: Rotation platform for directional channel characterization. The motor can replay field-collected traces or
orient the WARP board toward any given direction in the
azimuth plane
-30
Direction()
Direction()
-20
Dir-Client
Dir-AP
Omni-Client
Omni-AP
Direction()
-30
-40
Dir-Client
Dir-AP
Omni-Client
Omni-AP
-50
-60
0
Direction()
5dBi
30
5dBi
100
10
20
15
10
Error(dB)
15
Error(dB)
20
8dBi
100
Zero order
First order
25
total time(%)
total time(%)
25
0.01
Zero order
First order
0.01
5
0
8dBi
30
[0,0.1)
[0.1,1)
[1,10)
[10,inf)
superiority intervals(s)
[0,0.1)
[0.1,1)
[1,10)
[10,inf)
superiority intervals(s)
10ms
100ms
100ms
1s
Prediction Intervals(s)
Omni-directional antenna
Antenna switch
Transceiver
Directional
antennas
...
10ms
Figure 7: RSS prediction error for the 5dBi and 8dBi directional antennas for 10ms to 10s intervals
1s
Prediction Intervals(s)
RSSI
Antenna selection
Best
Yes
No
Safe
(2) Should
we assess?
No
Assessantennas
Yes
The key problem is to decide whether the client should assess the
antennas to find out the best one or simply use the omni directional antenna. Our solution to this problem is based on two intuitive observations:
In the Best mode, is the last identified best antenna still the
best? The client considers the current antenna no longer the
best if the predicted RSSI of current antenna, avg, is lower
than that of any other antennas assessed in the last 100ms.
Antenna
training packet
Regular packet
SEL
RSS(dBm)
-45
ACK
Dir3
-50
Omni
-55
Dir1
Dir2
Antenna
training symbols
-60
0
ACK SEL
(a)
10
15
time(second)
20
(b)
8. EVALUATION
The method is inspired by the 802.11 Physical Layer Convergence Protocol (PLCP), in particular the 802.11n MIMO PLCP.
The 802.11 PLCP employs training symbols, for the receiver to
estimate the channel and decode the packet. We employ training
symbols similar to the 802.11n MIMO PLCP and existing antenna selection methods for MIMO, e.g. [14]. Each antenna takes
turn to transmit the training period (8s in 802.11). Knowing the
number of training symbols per antenna, the receiver detects the
symbols that belong to each antenna, and estimates the RSSI for
that antenna. The receiver then notifies the transmitter of the antenna with the strongest channel. This feedback from the receiver
to the transmitter makes the transmitter antenna selection different from and more challenging than antenna selection on the receiver side.
We have implemented MiDAS with up to three directional antennas on the WARP platform. Using this implementation, we are
able to collect traces for a detailed simulation study and also to
perform real-time experiments to demonstrate the effectiveness of
the packet-based and symbol-based antenna selection methods.
6
5
4
3
2
Upper bound
Symbol-based
Packet-based
6
5
4
3
2
2
1
100ms
1s
10s
NLOS-ind. LOS-ind.
outd.
Environment
1
10ms
Upper bound
Symbol-based
Packet-based
Gain(dB)
Upper bound
Symbol-based
Packet-based
Gain(dB)
Gain(dB)
Gain(dB)
Upper bound
Symbol-based
Packet-based
4
3
2
1
three
two-opp
two-adj
one
Antenna Configuration
5dBi
8dBi
Antenna Gain
(a)
(b)
(c)
(d)
Figure 12: Impact of system settings on the performance of the multi-antenna system. Each bar presents the median, 10th and
90th percentile gain for three antenna selection cases: 1) upper bound assuming the best antenna is always used; 2) symbol-based
antenna selection described in Section 7; and 3) packet-based antenna selection described in Section 6
There is one difference though. That is, four antennas, three directional and one omni directional, are used in the experiments.
Therefore, the prototype repeatedly transmits a training period of
128 symbols from each of the four antennas in turn as fast as
possible. The AP node measures and records the RSSI of all of
the antennas. As a result, we are able to measure the RSSI of each
antenna approximately every 10ms. We collect traces for three
5dBi and three 8dBi directional antennas, each in various propagation environments, including NLOS indoor, LOS indoor, and
outdoor, as described in Section 4.1. The same transmit power of
18dBm is used for all experiments. Figure 11 (b) presents a segment of the RSSI traces of the four antennas for NLOS indoor.
The RSSI traces allow us to simulate the impact of various aspects of the antenna selection methods in a controlled manner. In
the following, the simulation default assumes 1) the traffic is
Poisson with the mean packet interval of 10ms; 2) the configuration of the multi-antenna system is three identical directional
antennas and one omni directional antenna; 3) directional antennas have 5dBi peak gain; and 4) the environment is NLOS indoor.
By changing each of these four aspects, we are able to reveal their
impact on the system performance, reported in terms of RSS gain
at the AP over that of the omni directional antenna case. That is,
we report the difference between the RSS achieved by the proposed multi-antenna system and that by a traditional omni directional system. For each evaluation, we show the system performance for three antenna selection methods: the genie-aided upper
bound that knows the best antenna for every packet, the packetbased and symbol-based antenna selection methods according to
Sections 6 and 7, respectively. Figure 12 summarizes the results.
Each bar in the graphs shows the median gain (in dB) of MiDAS
antenna selection algorithms over an omni system in different
scenarios. We also include the 10th and 90th percentile values of
this gain in each bar. We make the following observations.
First, both symbol-based and packet-based antenna selection methods achieve very good gain (close to the upper bound) when
traffic is intensive, i.e. average packet interval is below 100ms, as
is apparent from Figure 12 (a). The packet-based method is
slightly worse than the symbol-based because the packet-based
method is much coarse-grained and has higher overhead in antenna assessment as discussed in Sections 6 and 7. Moreover, neither
of the two antenna selection methods ever falls below the omni
directional antenna case and therefore the safety belt provided by
the omni directional antenna does work. The packet-based method provides little gain when the traffic become sparser (packet
Third, our results indicate that a design with two directional antennas will provide much of the benefit of the directionality, as is
apparent from Figure 12 (c). It also shows that the opposite pair
(two-opp) provides more gain than the adjacent pair (two-adj) as
illustrated in Figure 11. Because two-opp covers a larger angular
range than two-adj, we do expect that two-opp outperforms twoadj averagely, although it is possible that two-adj outperforms
two-opp under certain propagation environments. Moreover, MiDAS provides decent gain even with only one directional antenna. Note that all these configurations also have one omni directional antenna in addition to their directional antennas.
Finally, our results indicate a more focused antenna may not necessarily provide higher gain. As observed from Figure 12 (d),
MiDAS with three 5dBi antennas achieves better gain than that
with three 8dBi antennas, even for the upper bound case. Comparing Figure 12 (c) and Figure 12 (d), we can see even two 5dBi
antennas outperform three 8dBi antennas. Recall Figure 6 from
Section 4. There we observed that the 5dBi antenna provides
longer superiority intervals than the 8dBi one. As a result, MiDAS with 5dBi antennas requires less frequent antenna assessment and antenna changes.
Gain(dB)
Upper bound
Symbol-based
Packet-based
4
3
2
1
0
one
two-opp
three
four
five
six
seven
eight
1
2
3
4
5
6
7
s.t.
pick
else
pick the base rate (
1)
pick
0
end
25
150
MiDAS-5dBi
MiDAS-8dBi
omni
GG-Upper bound
GG-MiDAS
PR-Upper bound
PR-MiDAS
100
20
15
50
10
5
0
-10
10
20
Omni SNR(dB)
30
0
0
10
20
30
Omni SNR(dB)
(a)
(b)
Figure 14: (a) Goodput of MiDAS+RA/PC (three 5dBi and
three 8dBi) and Omni+RA in NLOS indoor; (b) Goodput
Gain (GG) and transmit Power Reduction (PR) of MiDAS+RA/PC (three 5dBi) versus omni+RA in NLOS indoor
transmit power is already included in the corresponding omni link
SNR in the x axis of both graphs in Figure 14.
Figure 14 (a) shows the goodput of MiDAS+RA/PC with three
5dBi and three 8dBi antennas. The same figure also shows the
goodput for the baseline omni+RA system. We make the following observations: First, three 5dBi antennas outperform three
8dBi antennas. Second, there are some SNR values for which
omni system has no connection or its goodput is low (below 3.3
Mbps), but MiDAS+RA/PC is capable of maintaining a much
higher goodput. We refer to these SNR values as Weak or no
connection region. Third, MiDAS+RA/PC always outperforms
the omni system. Figure 14 (b) shows the performance of MiDAS+RA/PC in terms of goodput improvement and power reduction for three 5dBi antenna compared to the omni system. We
have not plotted these graphs for Weak or no connection region, because the goodput gain can be infinite in this regime. The
first point in Figure 14 (b) corresponds to the SNR value in which
omni system has 3.3Mbps goodput.
We have also computed the genie-aided upper bound for the two
evaluated metrics with the assumption that the node has perfect
knowledge of channel conditions in the future. The results show
that MiDAS+RA/PC has a performance very close to the upper
bound. The results show that, for omni-SNR = 3 dB, where omni
system has a goodput of 3.3Mbps, MiDAS+RA/PC increases the
goodput by 132% while reducing the transmit power by 7%. Also, for omni-SNR = 30 dB, where omni system has a goodput of
34.4 Mbps, MiDAS+RA/PC maintains the same goodput while
reducing the transmit power by 48%
14
Goodput (Mbps)
30
Goodput (Mbps)
35
12
10
BPSK
QPSK
QAM16
8
6
4
2
0
10
20
30
40
SNR (dB)
transmit power over all possible values. For each transmit power,
we send packets as fast as possible for 30 seconds. The size of the
data part of the packet is 1400 bytes for QAM16 modulation, and
is modified accordingly for other modulations to have equal airtime. Figure 15 shows the results. We observe that QPSK modulation always outperforms BPSK over all possible SNR values
in terms of link goodput. Therefore, an SNR-triggered rate adaptation algorithm will not choose BPSK at any input SNR. This
means that we can only benefit from the other two modulations in
our experiments.
We performed two experiments using three 5dBi directional antennas. The client is placed on the motor platform which rotates
according to the traces, and the AP is stationary, and each experiment lasts for 20 minutes. The packets are transmitted as fast as
possible using the selected modulations. In the first experiment,
the omni antenna experiences a low SNR which is not enough to
use the QAM16 modulation. In the second experiment, the omni
link is strong enough to switch to the higher modulation. The
results show that, in the first experiment, the gain from MiDAS is
enough to switch to the higher modulation for most of the packets, and therefore the goodput is increased by 85%, using the
same transmit power as the omni transmitter. In the second experiment, this gain from MiDAS is used to save transmit power by
51% while increasing the goodput by 7%.
11. CONCLUSIONS
We make the following conclusions following the characterization, simulation, and experimentation reported in this work.
Through MiDAS and its antenna selection methods, directional antennas can be effectively employed to improve the
transmission gain of smartphone-like mobile devices by almost 3dB under various propagation environments and realistic rotation. Such gain can be achieved without any change
to the deployed network infrastructure.
Three 5dBi antennas, placed on the adjacent sides of a mobile device, can provide 3dB gain of MiDAS.
ACKNOWLEDGEMENTS
The work was supported in part by NSF awards ECCS/IHCS
0925942, CNS/NeTS-WN 0721894, CNS/CRI 0751173, CNS0551692, CNS-0551692, and support from the TI Leadership
University program. The authors would like to thank the anonymous reviewers for their constructive comments and thank Dr.
Dina Papagiannaki for shepherding the camera ready.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
tenna Selection," in IEEE Transactions on Vehicular Technology, vol. 58, issue 3, 2009.
J. Camp and E. Knightly, "Modulation rate adaptation in urban
and vehicular environments: cross-layer implementation and
experimental evaluation," in Proc. ACM MobiCom, September
2008.
D. Qiao, S. Choi, and K. Shin, "Goodput analysis and link
adaptation for IEEE 802.11 a wireless LANs," in IEEE Transactions on Mobile Computing, vol. 1, issue 4, 2002.
M. Youssef, M. Yosef, and M. El-Derini, "GAC: EnergyEfficient Hybrid GPS-Accelerometer-Compass GSM Localization," in Arxiv preprint arXiv:1004.3174.
K. Kunze, P. Lukowicz, K. Partridge, and B. Begole, "Which
way am I facing: Inferring horizontal device orientation from
an accelerometer signal," in Proc. IEEE Int. Symp. on Wearable Computers (ISWC), Septemeber 2009.
P. Mohan, V. Padmanabhan, and R. Ramjee, "Nericell: Rich
monitoring of road and traffic conditions using mobile smartphones," in Proc. ACM SenSys, November 2008.
J. C. Liberti and T. S. Rappaport, Smart Antennas for Wireless
Communications: IS-95 and Third Generation CDMA Applications: Prentice Hall PTR Upper Saddle River, 1999.
V. Navda, A. P. Subramanian, K. Dhanasekaran, A. TimmGiel, and S. Das, "MobiSteer: using steerable beam directional
antenna for vehicular network access," in Proc. ACM MobiSys,
June 2007.
C.-L. Yang and W. J. Chappell, "Angular Diversity Prominent for Compact Devices of Wireless Sensor Network in Indoor Environments," in Proc. IEEE Antennas and Propagation
Society Int. Symp. (APS), June 2007.
A. Amiri Sani, H. Dumanli, L. Zhong, and A. Sabharwal,
"Power-efficient directional wireless communication on small
form-factor mobile devices," in Proc. ACM/IEEE Int. Symp.
Low Power Electronics and Design (ISLPED), August 2010.
O. Norklit, P. D. Teal, and R. G. Vaughan, "Measurement and
evaluation of multi-antenna handsets in indoor mobile communication," in IEEE Transactions on Antennas and Propagation,
vol. 49, issue 3, 2001.
J. S. Colburn, Y. Rahmat-Samii, M. A. Jensen, and G. J. Pottie,
"Evaluation of personal communications dual-antenna handset
diversity performance," in IEEE Transactions on Vehicular
Technology, vol. 47, issue 3, 1998.
M. Rogers, K. Hrovat, K. McPherson, M. Moskowitz, and T.
Reckart, "Accelerometer data analysis and presentation techniques," Technical Report, National Aeronautics and Space
Administration, Lewis Research Center, 1997.
Static
Accelerated
Avg.
Max.
Avg.
Max.
0.9
1.5
1.1
1.8
3.8
9.5
1.5
3.3
11.7
21.1
4.6
9.5
Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.
Alternative Proxies: