mute: sounding inhibition for mu-mimo wlansob4/files/secon2014_bejarano.pdfcommunications magazine,...

Bejarano SECON 2014

MUTE: Sounding Inhibition for MU-MIMO WLANs

Oscar Bejarano

July 1st, 2014

1

Edward W. Knightly

Eugenio Magistretti

Omer Gurewitz

Rice University Rice University

Rice UniversityBen Gurion University

Bejarano

Single-Antenna Systems (Downlink)Motivation

2

802.11-Based Networks

AP

SECON 2014

Bejarano


3


AP

SECON 2014

Bejarano


4


AP

SECON 2014

Bejarano


5


AP

SECON 2014

Bejarano


6


AP

SECON 2014

Bejarano

Multi-Antenna Systems (Downlink)Motivation

7


AP

SECON 2014

Bejarano


8


AP

SECON 2014

Bejarano


Similarly to SU-MIMO

‣ Increases spectral efficiency

9

In contrast to SU-MIMO

‣ As many users as antennas at AP ‣ Multiplexing gain at AP even with minimal

number of antennas in usersantennas at AP!antennas at user

MkNk

Multi-User MIMO

‣ Spatial multiplexing ‣ Simultaneous spatial sharing of medium by

multiple users!

‣ Sum capacity scales with min(M,X

Nk)

Spencer, Q. H., Peel, C. B., Swindlehurst, A. L., & Haardt, M. (2004). An introduction to the multi-user MIMO downlink. Communications Magazine, IEEE, 42(10), 60-67.

AP

SECON 2014

Bejarano

Multi-User MIMOMotivation

10

Theoretical

Venkatesan’03 Viswanathan’03 

Caire’03  Jindal’04  

Spencer’04  Sharif ’05  Yoo’06 

Gesbert’07  Caire’10  

Experimental

Aryafar’10 Rahul’12  Balan’12 

Shepard’12 Shen’12  Yang’13 

Zhang’13 Chen’13  

SECON 2014

Extensive body of literature (theoretical and recent experimental work) has demonstrated vast capacity gains

‣ Large PHY gains!‣ MAC not considered

Bejarano

The ProblemMotivation

Costly overhead

11SECON 2014

Bejarano


12

Costly overhead

AP

Signal precoding enables MU-MIMO transmissions

‣ Channel State Information at Transmitter (CSIT) necessary !‣ Beam-steering weight computation (for inter-stream interference cancellation)!‣ Obtained via sounding 

time

AP

U1

U2

U3

U4

SECON 2014

Bejarano


13SECON 2014

We demonstrate that the costs required to enable MU-MIMO can outweigh the benefits

‣ Sounding process in current MU-MIMO systems is expensive and inefficient

‣ MAC enhancements necessary!‣ Large gap between innovative theoretical tools and protocol design

To provide a protocol-based framework that guarantees the benefits of MU-MIMO outweigh costs, with the goal of realizing PHY gains at the

system level

Our Objective

Bejarano

Proposed SolutionMUTE

14SECON 2014

We propose MUTE

MUTE addresses the issue of overhead associated with channel sounding

‣ Temporarily inhibits sounding based on channel stability!‣ Leverages presence of static users and epochs characterized by slowly

moving channels!

‣ Best case: MU-MIMO transmissions without preceding channel sounding!‣ Worst case: Basic 802.11ac behavior!!

Bejarano

RoadmapRoadmap

‣ Motivation!‣ Sounding process in MU-MIMO!‣ Sounding overhead reduction via sounding inhibition!

‣ Design of MUTE!‣ Evaluation of MUTE!

‣ Conclusion!

15SECON 2014

Bejarano

IEEE 802.11ac Sounding TimelineSounding Overhead

16

AP

U1

U2

U3

Null Data Packet Announcement: inform which

users will be served next

AP

NDPA

time

SECON 2014

Bejarano


17

Null Data Packet: sound users (training

sequences)

AP

U1

U2

U3

AP

SIFS

!NDPNDPA

time

SECON 2014

Bejarano


18

Compressed Beamforming Report:

channel information for each user

AP

U1

U2

U3

Compressed !BF Report

AP

SIFS

!NDPNDPA SI

FS

time

SECON 2014

Bejarano


19

Report poll: request beamforming report

from each user

AP

U1

U2

U3

AP

SIFS

!NDPNDPA SI

FS Poll

SIFS


time

SECON 2014

Bejarano


20

Repeat for the remaining users

AP

U1

U2

U3

AP

SIFS

!NDPNDPA SI

FS Poll

SIFS



SIFS Poll

SIFS

SIFS


time

SECON 2014

Bejarano

IEEE 802.11ac Sounding Overhead AnalysisSounding Overhead

21

!We demonstrate sounding overhead has a significant impact

on the overall system performance

SECON 2014

Bejarano


22

Metric ! Fraction of airtime consumed by sounding overhead

time

Sounding Data transmission

Sounding time

Sounding time + Data transmission time

Parameters Maximum subcarrier grouping!! Minimum quantization bits!! Packet Size 1500 bytes

Lower-Bound

SECON 2014

QPSK 16−QAM 64−QAM 256−QAM0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Frac

tion

of a

irtim

e co

nsum

edby

sou

ndin

g ov

erhe

ad

20 MHz ChannelQPSK 16−QAM 64−QAM 256−QAM

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Frac

tion

of a

irtim

e co

nsum

edby

sou

ndin

g ov

erhe

ad

80 MHz Channel

Bejarano


23


0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

20 MHz Channel

Frac

tion

of a

irtim

e co

nsum

edby

sou

ndin

g ov

erhe

ad

2 Users 96KB4 Users 96KB2 Users 18KB4 Users 18KB2 Users 1.5KB4 Users 1.5KB


0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

80 MHz Channel

>70%

No frame aggregation

SECON 2014

Bejarano


24


0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

20 MHz Channel

Frac

tion

of a

irtim

e co

nsum

edby

sou

ndin

g ov

erhe

ad



0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

80 MHz ChannelQPSK 16−QAM 64−QAM 256−QAM0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Frac

tion

of a

irtim

e co

nsum

edby

sou

ndin

g ov

erhe

ad


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Frac

tion

of a

irtim

e co

nsum

edby

sou

ndin

g ov

erhe

ad

80 MHz Channel

>40%

K.Tan, et al., “Fine-Grained Channel Access in Wireless LANs.” In Proc. of ACM SIGCOMM, 2010.

Frame Aggregation† ‣ Larger packets by aggregating frames to amortize overhead!‣ However, depends on traffic demands, contention, delay and channel stability

18 kB aggregation

SECON 2014

Bejarano


25


0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

20 MHz Channel

Frac

tion

of a

irtim

e co

nsum

edby

sou

ndin

g ov

erhe

ad



0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

80 MHz Channel

96 kB aggregation


0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Frac

tion

of a

irtim

e co

nsum

edby

sou

ndin

g ov

erhe

ad


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Frac

tion

of a

irtim

e co

nsum

edby

sou

ndin

g ov

erhe

ad

80 MHz Channel

>10%

K.Tan, et al., “Fine-Grained Channel Access in Wireless LANs.” In Proc. of ACM SIGCOMM, 2010.

Frame Aggregation† ‣ Larger packets by aggregating frames to amortize overhead!‣ However, depends on traffic demands, contention, delay and channel stability

SECON 2014

Bejarano


26

Metric Goodput

# of bits transmitted


AP

AP

0 10 20 30 40 50 600

50

100

150

200

250

300

350

400

Per−

Use

r Goo

dput

(Mbp

s)

Frame Aggreggation (Frames)

80 MHz

SECON 2014

Bejarano


27

Metric Goodput



AP

0 10 20 30 40 50 600

50

100

150

200

250

300

350

400

Per−

User

Goo

dput

(Mbp

s)


80 MHz

2 Users MU−MIMO2 Users SISO

AP

SECON 2014

Bejarano


28

Metric Goodput



AP

0 10 20 30 40 50 600

50

100

150

200

250

300

350

400

Per−

User

Goo

dput

(Mbp

s)


80 MHz

2 Users MU−MIMO2 Users SISO

AP

SECON 2014

Costs can outweigh the benefits

Bejarano

RoadmapRoadmap



‣ Conclusion!

29SECON 2014

Bejarano

Sounding Inhibition (MUTE)MUTE

30SECON 2014

Understanding MUTE

MUTE evaluates two key tradeoffs

Tradeoff 1:!

! Extensive channel knowledge at the AP VS increased sounding overhead

Tradeoff 2:!

! High channel estimate accuracy VS increased sounding overhead

*NOTE:  Notice, this correlation happens in signal space

!

‣ Sound and serve 2 users (possibly correlated*), OR !

‣ sound all users and serve the rate maximizing group (orthogonal/semi-orthogonal)

‣ However, prohibitive to sound more than 4 users

Bejarano Ph.D. Thesis Proposal

Sounding Inhibition (MUTE)Sounding Overhead

31

AP

Tradeoff 1:!

! Extensive channel knowledge at the AP VS increased sounding overhead

Leverage User Diversity

Bejarano


32SECON 2014

User Diversity - Find users with orthogonal/semi-orthogonal channels

Topology!

One 4-antenna AP!

30 single-antenna users!

!

Experiment!

1) Sound N random users (uniform distr.)!

2) Choose the K users that maximize rate!

!

Bejarano


33SECON 2014


Topology!

One 4-antenna AP!


!

Experiment!



!

K = 4!N = 4 to 10

4x10

5

10

15

20

25

30

35

40

45

50

55

Aggr

egat

e R

ate

(bps

/Hz)

4x40

5

10

15

20

25

30

35

40

45

50

55

Aggr

egat

e R

ate

(bps

/Hz)

1 Sounded Users2345678910Exhaustive Search 4x1

0

5

10

15

20

25

30

35

40

45

50

55

Aggr

egat

e R

ate

(bps

/Hz)

4x40

5

10

15

20

25

30

35

40

45

50

55

Aggr

egat

e R

ate

(bps

/Hz)

1 Sounded Users2345678910Exhaustive Search

N

Bejarano


34SECON 2014


Topology!

One 4-antenna AP!


!

Experiment!



!

K = 4!N = 4 to 10

4x10

5

10

15

20

25

30

35

40

45

50

55

Aggr

egat

e R

ate

(bps

/Hz)

4x40

5

10

15

20

25

30

35

40

45

50

55

Aggr

egat

e R

ate

(bps

/Hz)

1 Sounded Users2345678910Exhaustive Search 4x1

0

5

10

15

20

25

30

35

40

45

50

55

Aggr

egat

e R

ate

(bps

/Hz)

4x40

5

10

15

20

25

30

35

40

45

50

55

Aggr

egat

e R

ate

(bps

/Hz)

1 Sounded Users2345678910Exhaustive Search

Clear benefit from an increased number !of monitored/sounded users

N

Bejarano


35SECON 2014

Accuracy - Channel estimates degrade with time, specially in highly mobile environments

Sounding Overhead - Costly to sound every time before a transmission

VS

Tradeoff 2:!

! High channel estimate accuracy VS increased sounding overhead

Bejarano


36SECON 2014

MUTE strives to use !the most accurate information available !

from as many users as possible !while minimizing sounding overhead !to guarantee a net throughput gain

However, a fundamental change in traditional systems is needed



37

Traditional system - sounding user set selection coupled with transmission user set selection

Who to sound?

Who to serve?



38

AP Schedule transmissions for 4 users




39

Sound all same 4 users

High overhead


AP



40

Serve all same 4 users

Poor user grouping


AP



41

In contrast…



42

AP Schedule transmissions for 4 users


MUTE decoupled from



43

APSound a subset of those 4 users,  

e.g., 2 in this case


MUTE decoupled from

Low overhead



44

AP

Serve a new set of users.  Leverage user diversity based on

historical information the AP already has


MUTE decoupled from

Rate Maximization

Use previously collected channel statistics to select!a better option



45

AP

Serve a new set of users.  Leverage user diversity based on

historical information the AP already has


MUTE decoupled from

Rate Maximization

Use previously collected channel statistics

Bejarano


46SECON 2014

Therefore, a decoupled system:!!

‣Allows the AP to sound only the users that need to be sounded!!

‣Enables the AP to serve only the set of users that maximizes the aggregate rate

Bejarano


47

In MUTE, the AP relies on channel statistics to decide which users to sound

SECON 2014

We enable the AP to assess the tradeoff between predicted channel volatility and rate penalty due to inaccurate CSIT

Two Empirical Observations

‣ Variation in most recent samples provides insights into near-future samples (e.g., [1-5])!

‣ Collected samples in static channels degrade similarly with time (e.g., coherence time)

[1] Shen, Zukang, Jeffrey G. Andrews, and Brian L. Evans. "Short range wireless channel prediction using local information." Signals, Systems and Computers, 2004. Conference Record of the Thirty-Seventh Asilomar Conference on. Vol. 1. IEEE, 2003.[2] Duel-Hallen, Alexandra. "Fading channel prediction for mobile radio adaptive transmission systems." Proceedings of the IEEE 95.12 (2007): 2299-2313.

[3] Gesbert, David, et al. "Outdoor MIMO wireless channels: Models and performance prediction." Communications, IEEE Transactions on 50.12 (2002): 1926-1934.[4] Halperin, Daniel, et al. "Predictable 802.11 packet delivery from wireless channel measurements." ACM SIGCOMM Computer Communication Review 41.4 (2011): 159-170.[5] Phillips, Caleb, Douglas Sicker, and Dirk Grunwald. "A survey of wireless path loss prediction and coverage mapping methods." Communications Surveys & Tutorials, IEEE 15.1 (2013): 255-270.

Bejarano


48

time0 0.5 1 1.5 2 2.5 3 3.5 400.001

0.002

0.003

0.004

0.005

0.006

0.007

0.008

0.009

0.01

Time (sec)

|H|

1subcarrier−1transmitter

most recent

chan

nel m

agni

tude

t1 t2

age

Two Empirical Observations:!

Variation in most recent samples provides insights into near-future samples !

Collected samples in static channels degrade similarly with time

age: analogous to concept of coherence time

SECON 2014

Bejarano


49

‣ AP computes magnitude and phase change between each collected sample and selects relevant samples!

‣ Most recent samples!‣ Samples within certain age

In principle, MUTE determines how much the last collected sample is expected to vary

time0 0.5 1 1.5 2 2.5 3 3.5 400.001

0.002

0.003

0.004

0.005

0.006

0.007

0.008

0.009

0.01

Time (sec)

|H|

1subcarrier−1transmitter

most recent

chan

nel m

agni

tude

t1 t2

age

!magnitude

change - sample variance

0 50 100 150 2000

0.005

0.01

0.015

0.02

0.025STATIC−NLOS

Age (Time Diff. Between Two Consecutive Samples

Mag

nitu

de C

hang

e (A

bsol

ute

Diffe

renc

e)

0 50 100 150 2000

0.05

0.1

0.15

0.2

0.25

0.3

0.35STATIC−NLOS

Age (Time Diff. Between Two Consecutive Samples

Chan

ge in

Ang

le (A

bsol

ute

Diffe

renc

e)

Abs

olut

e M

agni

tude

Cha

nge

Age

sample variance

‣ Compute sample variance‣ If variance above threshold,  

sound user (per-user threshold)

SECON 2014

Bejarano

RoadmapRoadmap



‣ Conclusion!

50SECON 2014

Bejarano

Experimental Evaluation of MUTEMUTE

51SECON 2014

Can MUTE strike a balance between overhead suppression and rate penalty due to inaccurate channel estimates?

Our evaluation answers the following question:

Bejarano


52

‣ Trace-driven emulation!‣ Complete downlink zero-forcing beamforming system!‣ Flexible system, replay channels for different schemes

the first one in order to collect their reports.

AP

U1

U2

U3

SIFS

SIFS

SIFS

SIFS

SIFS

SIFSNDPA NDP

Compressed Beamforming



Beamforming Report Poll

Beamforming Report Poll

AP

U1

U2

U3

Multi-User DataU1

U2U3 U3

PADPAD

BARSIFS

SIFS

Block ACK

BARSIFS

SIFS BARSIFS

SIFS

Block ACK

Block ACK

(a)

(b)

Monday, March 10, 14

Figure 2: 802.11ac Timeline: Sounding and feed-back process

Modifying the 802.11ac Sounding Structure inCUF. To enable MMSE-SIC, the preambles and beam-forming report data need to be aligned so that pream-bles are sent in a staggered manner whereas data trans-missions begin simultaneously. CUF leverages the basicsounding structure in 802.11ac to do this alignment bytriggering each user’s transmission accordingly. Thatis, upon reception of the NDP, all users that were ad-dressed in the NDPA packet compute their channel vec-tors to the AP. Then, based on the order specified inthe NDPA, they align their preambles and beamform-ing reports. Figure 3 shows the proposed alignment forthe case of three di↵erent users. By zero-padding be-tween preambles and payload carrying the beamformingreports, we can align the reports from all users with-out requiring any additional synchronization. This al-lows the AP to receive each preamble from the di↵erentstreams in a clean manner. Notice that for interop-erability with legacy devices, the zero-padding can bedone at the beginning of the data packet therefore con-serving the original 802.11ac preamble structure.

IEEE P802.11ac/D5.0, January 2013

305 Copyright © 2013 IEEE. All rights reserved.This is an unapproved IEEE Standards Draft, subject to change.

123456789

1011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465

NSTS in VHT-SIG-A indicates that there are zero space-time streams for the STA in the PPDU, then the STAmay elect to not process the remainder of the PPDU.

22.3.12 VHT preamble format for sounding PPDUs

NDP is the only VHT sounding format.

The format of a VHT NDP PPDU is shown in Figure 22-26.

NOTE—The number of VHT-LTF symbols in the NDP is determined by the SU NSTS field in VHT-SIG-A.

The VHT NDP PPDU has the following properties:— uses the VHT PPDU format but without the Data field— is a VHT SU PPDU as indicated by the VHT-SIG-A field— has the data bits of the VHT-SIG-B field set to a fixed bit pattern (see 22.3.8.3.6 (VHT-SIG-B defi-

nition))

22.3.13 Regulatory requirements

Wireless LANs (WLANs) implemented in accordance with this standard are subject to equipment certifica-tion and operating requirements established by regional and national regulatory administrations. The PHYspecification establishes minimum technical requirements for interoperability, based upon established regu-lations at the time this standard was issued. These regulations are subject to revision or may be superseded.Requirements that are subject to local geographic regulations are annotated within the PHY specification.Regulatory requirements that do not affect interoperability are not addressed in this standard. Implementersare referred to the regulatory sources in Annex D for further information. Operation in countries withindefined regulatory domains might be subject to additional or alternative national regulations.

22.3.14 Channelization

A VHT channel is specified by the four PLME MIB fields specified in Table 22-22 (Fields to specify VHT

Figure 22-26—VHT NDP format

L-STF L-LTF L-SIG VHT-SIG-AVHT-STF VHT-LTF

8 µs 8 µs 8 µs4 µs 4 µs 4 µs per VHT-LTF symbol

VHT-SIG-B

4 µs

AP

U1

U2

U3

Preamble

SIFS

Nc IndexThis index describes the number of columns in the feedback matrix, with one col‐umn for each spatial stream. As a three-bit field it can take on eight values, whichmatches the eight streams supported by 802.11ac. This field is set to the number ofspatial streams minus one.

Figure 4-7. NDP Announcement frame format (single-user)

NDP frameUpon transmission of the NDP Announcement frame, the beamformer next transmitsa Null Data Packet frame, which is shown in Figure 4-8. The reason for the name “nulldata packet” should be obvious in looking at the frame; Figure 4-8 shows a PLCP framewith no data field, so there is no 802.11 MAC frame. Channel sounding can be carriedout by analyzing the received training symbols in the PLCP header, so no MAC data isrequired in an NDP. Within an NDP there is one VHT Long Training Field (VHT-LTF)for each spatial stream used in transmission, and hence in the beamformed datatransmission.

Figure 4-8. NDP format

VHT Compressed Beamforming Action frameFollowing receipt of the NDP, the beamformee responds with a feedback matrix. Thefeedback matrix tells the beamformer how the training symbols in the NDP were re‐ceived, and therefore how the beamformer should steer the frame to the beamformee.

Single-User (SU) Beamforming | 71NDP

Preamble

Preamble

Pad

Pad

SIFS

Multi-User DataU1

U2U3 U3

PadPad

AP

U1

U2

U3

BF Preamble

SIFS

Pad

Pad

BF Report

BF Report

BF Report

BF Report

BF Report

BF Report

BF Preamble

BF Preamble

Pad

Pad

Multi-User DataU1

U2U3 U3

PadPad

Staggered Preambles + SDMA-Based Reports

Staggered Beamformed Preambles

Stream for User 1

Stream for User 2

Stream for User 3

U1Multi-User Data

U2

U3 U3 Pad

Pad

} SDMA

Friday, February 21, 14

Figure 3: CUF Staggered Feedback Format

User Synchronization and RF Mismatch De-tection and Correction. In contrast to single-userMIMO architectures in which all transmitted streamsoriginate from the same device with RF interfaces shar-ing a common clock and a fully synchronized transmis-sion trigger, in uplink MU-MIMO (or SDMA), trans-missions from di↵erent users do not have these charac-teristics. This leads to the following key synchroniza-tion challenge. The di↵erence between transmit chainsin all users leads to a di↵erent carrier frequency o↵-set (CFO) for each user. Although CFO can be eas-ily estimated and corrected in the single-user MIMO

case, this estimation represents a significant challengein the multi-user case. In particular, the estimation foreach user needs to be performed over a clean preamblewithout interference from other streams. As previouslymentioned, CUF achieves this by staggering the pream-bles. However, correcting for the CFO of each streambecomes more di�cult since the received signal is a lin-ear combination of multiple independent streams.

In addition to CFO estimation, CUF proposes a methodfor correcting the o↵set from each stream. In general,the method works as follows. After CFO estimation forall streams, the compound signal is passed through theMMSE-SIC receiver. Once the AP knows which streamto decode first, it applies the CFO correction for thatspecific stream to the entire compound signal. Next,the MMSE-SIC algorithm decodes this first stream, re-moves the applied CFO correction, and removes the de-coded signal component from the original compoundsignal. The process is repeated until all streams are de-coded. We provide an analytical derivation to demon-strate the feasibility of this method. Let x

i

(n) be thetime domain symbol n transmitted by the ith user, andh

i

(n) be impulse response of the ith channel (for eachantenna at the AP). Consequently, after passing thesignal of the ith user through the channel we obtains

i

(n) = xi

(n) ⇤ hi

(n). Ignoring the noise terms, thecompound received baseband signal is given by

r(n) =KX

i=1

si

(n)ej2⇡�fin

where K is the number of users transmitting simulta-neously to the AP, and �f

i

denotes the ith user’s CFOnormalized by symbol period. Assuming that the APattempts to decode the signal from user 1 first, it willcorrect the CFO in time domain by multiplying r(n)with the term e�j2⇡�f1n. Thus, in the first MMSE-SICiteration the compound signal obtained after correctingthe CFO of the first stream is given by

y(n) = r(n)e�j2⇡�f1n

= s1(n)(1) + s2(n)ej2⇡n(�f2��f1) + · · ·

+ sK

(n)ej2⇡n(�fK��f1)

Notice that the MMSE-SIC receiver can now start de-coding the component of the first stream. After thefirst stream has been decoded, the compound signal ismultiplied by ej2⇡�f1n to remove the CFO componentfrom this first stream. This yields a compound signalcomprised of all original streams except the first one,where each of them can be decoded in the same way.

3.2 Preventing Decoding Errors Due to SignalCompounding

The performance of the decoding process in CUF canbe limited by the superposition of signals with very

4

‣ Comprehensive channel measurement collection (indoor static, dynamic, and mobile environments)

802.11ac timings

SECON 2014

Bejarano


53

‣ Benchmark!‣ Always sound!‣ Most updated information!

!‣ MUTE - Two tolerance levels!‣ Set threshold to allow close to 2 bps/Hz loss!‣ Set threshold to allow close to 1 bps/Hz loss!‣ Tradeoff: overhead reduction vs rate loss!

!‣Environments!‣ Static - Static users and static environments!‣ Dynamic - Static users and dynamic environments

SECON 2014

Setup

Bejarano


54




SECON 2014

Setup

Bejarano


55




SECON 2014

Setup

Bejarano


56

‣ Setup!‣ Overhead not considered!‣ 4x4 system!‣ 30-user topology!‣ Compute rate loss!

‣ Rate penalty inversely proportional to overhead reduction!

‣ Smaller penalty in dynamic scenarios — more conservative!

‣ Higher penalty in static scenarios  — less conservative!

‣ Accurately tune how much we are willing to sacrifice in terms of rate performance

SECON 2014

Evaluation of rate penalty due to infrequent sounding

Bejarano


57SECON 2014



‣ Rate penalty inversely proportional to overhead reduction!

‣ Smaller penalty in dynamic scenarios — more conservative!

‣ Higher penalty in static scenarios  — less conservative!


Static Dynamic0

2

4

6

8

10

Per−

Use

r Rat

e(b

ps/H

z)

MUTE 2bps/Hz toleranceMUTE 1bps/Hz toleranceBenchmark

Bejarano


58

Static Dynamic0

2

4

6

8

10

Per−

Use

r Rat

e(b

ps/H

z)



‣ Penalty inversely proportional to overhead reduction!

‣ Smaller penalty in dynamic  — more conservative!

‣ Higher penalty in static  — less conservative!


SECON 2014


Bejarano


59SECON 2014

Evaluation of overall throughput performance

‣ Setup!‣ Overhead considered!‣ 4x4 system!‣ 30-user topology!‣ 1.5 to 18 kBytes aggregation!‣ Compute xput gain compared to

benchmark!

‣ Significant gains in different environments!‣ Near 70% gains, and 30% even with

large frame aggregation!

‣ Close to 50% gains in dynamic!‣ MUTE adapts to provide balance between

overhead and estimate accuracy

Bejarano


60SECON 2014

Static Dynamic Combined0

20

40

60

80

Thro

ughp

utG

ain

(%)

1.5 KBytes 3 KBytes 18 KBytes


20

40

60

80

Thro

ughp

utG

ain

(%)


Evaluation of overall throughput performance‣ Setup!

‣ Overhead considered!‣ 4x4 system!‣ 30-user topology!‣ 1.5 to 18 kBytes aggregation!‣ Compute xput gain compared to

benchmark!





Bejarano


61SECON 2014


20

40

60

80

Thro

ughp

utG

ain

(%)



20

40

60

80

Thro

ughp

utG

ain

(%)


Evaluation of overall throughput performance‣ Setup!

‣ Overhead considered!‣ 4x4 system!‣ 30-user topology!‣ 1.5 to 18 kBytes aggregation!‣ Compute xput gain compared to

benchmark!





Bejarano


62

Static Dynamic0

2

4

6

8

10

Per−

Use

r Rat

e(b

ps/H

z)


MUTE attains a net throughput gain  

‣ Gains originated from sounding overhead reduction dominate the losses incurred due to inaccurate channel estimates

SECON 2014


20

40

60

80

Thro

ughp

utG

ain

(%)



20

40

60

80

Thro

ughp

utG

ain

(%)


Bejarano

ConclusionMUTE

63

Costs to enable MU-MIMO can outweigh the benefits !

Even without considering losses due to inter-stream interference!!

Sounding overhead in 802.11ac can be detrimental to MU-MIMO performance

We demonstrate the feasibility of sounding inhibition in MU-MIMO networks

!MUTE strikes a balance between overhead reduction and rate penalty due to inaccurate

channel estimates

SECON 2014

Bejarano

ConclusionMUTE

64

Thank you!

SECON 2014

mute: sounding inhibition for mu-mimo wlansob4/files/secon2014_bejarano.pdfcommunications magazine,...

Documents