Design and Experimental Evaluation
of Multi-User Beamforming in
Wireless LANs
Ehsan Aryafar
Rice
Narendra Anand
Rice
Theodoros Salonidis Edward Knightly
Technicolor
Rice
ACM MobiCom 2010
MIMO LANs
• MIMO increases throughput with antenna arrays at transmitter and
We present the design and experimental
receiver
• However,
real world client
devices
fewer antennas
thanfor
APs due to
evaluation
of the
firsthave
MUBF
platform
cost and space
WLANs
• MUBF allows for APs to leverage
antennas belonging to group of nodes
Xirrus 16 ant AP
Tx
Rx
Tx
Rx
Ehsan Aryafar
Rx Rice
Rx Networks
Rx RxGroup
Ehsan Aryafar
Rice Networks Group
Crash Course on Beamforming
AP
p12
• Omni
– Fixed vs ant
selection
Ehsan Aryafar
Rice Networks Group
Crash Course on Beamforming
AP
p12
• Omni
– Fixed vs ant
selection
Ehsan Aryafar
• Adaptive Beam (SUBF)
– Higher coverage
– Higher SNR
Rice Networks Group
Multi-User Beamforming:
Throughput Increase
desired signal
w1
w2
AP
w1s1 + w 2 s2
s1
w'1 s1 + w' 2 s2 s2


y1  h1 w1 s1  h1 w2s2  noise
h1
inter-user interference


• MUBF sends the contents to both receivers at the same time
• Each user’s data stream is weighted at the transmitter
• Appropriate weights can reduce or eliminate the amount of
inter-user interference
Ehsan Aryafar
Rice Networks Group
Multi-User Beamforming:
Throughput Increase
desired signal
w1s1
w 2 s2
+
w'1 s1 w' 2 s2
y1  h1 w1 s1  h2 w2s2  z1
AP
s1
s2
inter-user interference


0
• Zero-Forcing beamforming (ZFBF)
– weights are selected such that the amount of
inter-user interference is zero
Ehsan Aryafar
Rice Networks Group
Multi-User Beamforming:
Interference Reduction
Client 2
Client 1
User affected by AP’s
interference
AP
Channel Information
• A user can obtain an interference-free channel
by sharing its channel information
Ehsan Aryafar
Rice Networks Group
Outline
• Background
• System Implementation
• Experimental Evaluation
• Conclusion
Ehsan Aryafar
Rice Networks Group
Methodology
• Unified Implementation Platform
– First Implementation and experimental evaluation of
different beamforming algorithms on a common platform
• Experimental Characterization of System Performance
– Compare against single-user TDMA schemes
– Use repeatable controlled channels and
– Real-time indoor channels
• Evaluation Metric
– SNR or the corresponding Shannon capacity
Ehsan Aryafar
Rice Networks Group
WARPLab Research Framework
• WARP is clean-slate MAC and PHY
– Off-the-shelf platforms: Limited
programmability/observability
• WARPLab brings together WARP and
MATLAB
– Manage network communication of up to
16 WARP nodes
– Baseband signals are generated in
MATLAB and downloaded to WARP nodes
– WARP nodes send/receive the RF signals
Ehsan Aryafar
Virtex-II Pro FPGA
Rice Networks Group
Implementation
1
Training
BF Weights
5
( )
H Matrix
and Weight
Calculation
4
Log RSSI Data
(End of Cycle)
8
For more information about our testbed and
7
implementation please attend our demo!
Rx
Rx
Rx
Tx
3
6 MUBF Data (OTA)
2
Training (OTA)
Ehsan Aryafar
Rice Networks Group
Experimental Design
• Multiplexing Gain
– Receiver separation distance
– User selection algorithm
– User population size
• Channel Variation
– Environmental variation
– User mobility
• Spatial Reuse
– Location based interference
– Multi-point interference reduction
– Network throughput
Ehsan Aryafar
Rice Networks Group
Impact of Receiver Separation
• Issue: How does receiver
separation distance affect
spatial multiplexing gain?
Ehsan Aryafar
Rice Networks Group
Impact of Receiver Separation
• Issue: How does receiver
separation distance affect
spatial multiplexing gain?
8
λ
9
-4
λ/4 - 6
10
7
λ/2 - 5
R11
3
11
R221
TX
2.85m
2.85m
Ehsan Aryafar
Rice Networks Group
Impact of Receiver Separation
• Similar capacity up to λ/2
Separation distance
• ZFBF at λ/4:
– 6 dB decrease in per-link
SNR
Ehsan Aryafar
Omni
SUBF
ZFBF
8
6
4
2
0
Location ID: 2
30
Per-link SNR (dB)
• ZFBF doubles capacity
compared to Omni
Sum Capacity
(bps/Hz)
• Issue: How does receiver
separation distance affect
spatial multiplexing gain?
10
3
4(λ) 5(λ/2) 6(λ/4) 7
3
4(λ) 5(λ/2) 6(λ/4) 7
20
10
0
Location ID: 2
Rice Networks Group
Experimental Design
• Multiplexing Gain
– Receiver separation distance
– User selection algorithm
– User population size
• Channel Variation
– Environmental variation
– User mobility
• Spatial Reuse
– Location based interference
– Multi-point interference reduction
– Network throughput
Ehsan Aryafar
Rice Networks Group
User Mobility
• Issue: Evaluate impact of
outdated channel
information due to user
mobility
h'2

Ehsan Aryafar
h1
h2
h'1
 

Rice Networks Group
User Mobility
bps/Hz
• Issue: Evaluate impact of
outdated channel
information due to user
mobility
• Repeatable channel
Similar
conditionsexperiments
Aggregate Capacity
– Channel must be updated
at (λ/8) movement
– Equal to 10 msec update
rate for a typical pedestrian
speed (3 mph)
Ehsan Aryafar
SNR (dB)
can be done for static
– 802.11n Task(in
Group
receivers
paper). The required channel
channel model
rate for a typical residential environment is
Per-link SNR
• Required channel update
100 msec.
rate
Rice Networks Group
Experimental Design
• Multiplexing Gain
– Receiver separation distance
– User selection algorithm
– User population size
• Channel Variation
– Environmental variation
– User mobility
• Spatial Reuse
– Location based interference
– Multi-point interference reduction
– Network throughput
Ehsan Aryafar
Rice Networks Group
Multi-Point Interference Reduction
• Issue: Evaluate a sender’s
ability to reduce
transmission footprint at
multiple locations
p1
– Interference reduction at
unintended receivers
– Impact on the QoS of the
served user
Interference
Reduction Points
Ehsan Aryafar
Rice Networks Group
Multi-Point Interference Reduction
• Issue: Evaluate a sender’s
ability to reduce
transmission footprint at
multiple locations
• Interference Reduction:
– Interference reduction
capability does not depend
on the location/number of
unintended receivers
Ehsan Aryafar
Rice Networks Group
Multi-Point Interference Reduction
• Issue: Evaluate a sender’s
ability to reduce
transmission footprint at
multiple locations
SNR difference at the intended receiver
• Interference Reduction:
– Interference reduction
capability does not depend
on the location/number of
unintended receivers
• Increase in number of
unintended receivers, can
significantly drop the QoS of
the currently served users
Ehsan Aryafar
Rice Networks Group
Prior Work
• Theoretical Work on MU-MIMO
– DPC (Costa’83) and its optimality (CS’03)
– ZFBF (YG’06 and WES’08)
• Practical Protocols
– IAC (GPK’09) and SAM (TLFWZCV’09)
We present the design and experimental evaluation
of a MUBF platform for wireless LANs
Ehsan Aryafar
Rice Networks Group
In Summary
• Design and implementation of the first MUBF platform for WLANs and
found via experimental evaluation:
• Users can simultaneously receive data down to a half of wavelength from
one another
• ZFBF can tolerate channel variations due to environmental variation,
however, is strongly affected by user mobility
• ZFBF can efficiently eliminate interference at undesired locations. This
does not depend on the location/number of unintended receivers,
however, can significantly reduce the QoS for the currently served users
WARP: http://warp.rice.edu
RNG: http://networks.rice.edu
Ehsan Aryafar
Rice Networks Group
Back Up
Ehsan Aryafar
Rice Networks Group
iburst
Patented technology
for concurrent
transmission
Suitable for outdoor
channels
Ehsan Aryafar
Rice Networks Group
Crash Course on Beamforming
AP
p12
h11
h21

• Omni
– Fixed vs ant
selection
Ehsan Aryafar
• Switched Beam • Adaptive Beam
– Fixed beam
– High coverage
– Higher range
– SUBF
Rice Networks Group
Weight Selection Algorithms
• Zero-Forcing beamforming (ZFBF)
– Condition: h w  0  j k => Pkhk wk sk  z
k j
k
– Heterogeneous link qualities through power
allocation
• Regularized Channel Inversion


– Increase
system performance
– Does not easily allow for heterogeneous link
qualities due to non-zero inter-user interference
Ehsan Aryafar
Rice Networks Group
Multi-Point Interference Reduction
• Issue: Evaluate a sender’s
ability to reduce transmission
footprint at multiple locations
• Interference Reduction:
– SUBF’s interference could be
significantly higher/lower
than Omni
– ZFBF’s interference reduction
capability does not depend
on the location/number of
unintended receivers
Ehsan Aryafar
Rice Networks Group
Weight Selection
Zero Forcing Beamforming (ZFBF)
• Assume 4 Tx Antennas and 3 single-antenna receivers
 hA1
H   hB1
hC1
hA 2
hA3
hB 2
hC 2
hB 3
hC 3
hA 4 
hB 4 
hC 4 
hk's – H for each recv.
• Calculate weights with pseudo-inverse
 wA1 wA 2

*
* 1
W  H HH
  wB1 wB 2
 wC1 wC 2


• “Zero Interference” Condition
hk (wj )  0, k  j
T
wA3
wB 3
wC 3
wA 4 

wB 4 
wC 4 
wj's
Implementation - WARPLab
• All baseband processing performed on Host PC
• Processed signals are downloaded to buffers in FPGA on
transmitting WARP node
• HostPC sends Transmit/Receive trigger signals to WARP nodes
• Data is transmitted over the air, stored in buffers on receiving
node’s FPGA
• Data/RSSI readings uploaded to HostPC for data
processing/logging
User Population Size
Tx/Rx
1
4
6
5
Rx
Door Rx
Door
2
Tx/Rx
Tx/Rx
Door
2.85m
3
Rx
• Q: How does the number of
concurrently served users affect
performance?
Door
Door
Door
• A: Capacity increases and saturates
while per-user SINR drops significantly.
2.85m
Aggregate Capacity
SUBF
ZFBF
Omni
14
12
10
8
6
4
2
0
SUBF
ZFBF
25
20
SINR (db)
bps/hz
Omni
Average Per-User SINR
15
10
5
0
2
3
# of Receivers
4
2
3
# of Receivers
4
User Selection (Link Quality Difference)
Tx/Rx
1
4
6
5
Rx
Door Rx
Door
2
2.85m
• Q: How do link quality differences between
receivers affect system performance?
Tx/Rx
Tx/Rx
Door
2.85m
3
Rx
Door
Door
Door
• A: Link quality differences between
concurrently served users do not affect each
user’s SINR.
Environmental Variation
Aggregate Capacity
Omni-T
SUBF-T
ZFBF-T
Omni-R
SUBF-R
ZFBF-R
100
Omni-T
Omni-R
Time
(ms)
SUBF-T
SUBF-R
9
8
7
bps/hz
-802.11n Task Group model for indoor
residential environment
- (T) : Typical –Fading rate of 1.157 Hz
- (R) : Rapid –Fading rate of 2.778 Hz
6
5
4
3
2
• Q: How does performance vary
with channel update rate in
typically/rapidly varying channels?
10
50
25
500
Average Per-User SINR
ZFBF-T
ZFBF-R
• A: Assuming a link can suffer up to
a 3dB decrease in SNR below Omni,
100ms and 50ms update rates are
necessary for typically/rapidly
varying channels, respectively.
SINR (dB)
20
15
10
5
0
10
50
100
500
Time (ms)
Door
Stairs
λ/2
6
Door
10
9
4
5
7
TX
R
• Q: How does MUBF’s interference reduction
capability vary with the location of the
unintended receiver?
3
8
Door
Door
Interference Reduction (Location)
2
W
• A: The location of the unintended receiver
does not affect the interference reduction
performance of MUBF (when #Rx < DOF).
1
2.85m
Door
Door
Door
Interference at W
Omni
2.85m
ZFBF
Interference (db)
30
25
20
15
10
5
0
1
2
3
4
5
6
Location ID
7
8
9
10
Channel Variation
Ehsan Aryafar
Rice Networks Group
Testbed
Ehsan Aryafar
Rice Networks Group
Channel Estimation
Ehsan Aryafar
Rice Networks Group
Network Throughput
Ehsan Aryafar
Rice Networks Group