Dynamic Rate Adaptation in IEEE 802.11 WLANs SongYiLin@ICT August 10, 2008.

Dynamic Rate Adaptation in IEEE 802.11 WLANs

SongYiLin@ICT

August 10, 2008

References

[1] On the Performance Characteristics of WLANs: Revisited (SIGMETRICS 2005)

[2] CARA: Collision-Aware Rate Adaptation for

IEEE 802.11 WLANs (INFOCOM 2006)

[3] Robust Rate Adaptation for 802.11 Wireless Networks (MOBICOM 2006)

[4] IEEE 802.11 Rate Adaptation: A Practical Approach

(MSWiM 2004)

[5] Link Adaptation Strategy for IEEE 802.11 WLAN via Received Signal Strength Measurement (ICC 2003)

Outline

Introduction of Rate Adaptation in IEEE 802.11 WLANs

Existing ChallengesDevelopment of Rate Adaptation Al

gorithmsPresentative old onesNew Kids in this areaMy Opinion

Outline




802.11 MAC (with RTS/CTS on)

[2]

RTS: 20 bytes in mac

CTS: 14 bytes in mac

What is “Rate Adaptation” ?

The 802.11 a/b/g/n standards allow the use of multiple transmission rates

802.11b, 4 rate options (1,2,5.5,11Mbps) 802.11a, 8 rate options (6,9,12,18,24,36,48,54

Mbps) 802.11g, 12 rate options (11a set + 11b set)

The method to select the transmission rate in real time is called “Rate Adaptation”

Rate adaptation is important yet unspecified by the 802.11 standards

Why do we need “Rate Adaptation” ?

AccessPoint MN


AccessPoint MN


AccessPoint MN

Distance Effects :attenuation

fading interference

SRN


[5]

BPSK 1MbpsQPSK 2Mbps

BPSK/QPSK/CCKDifferent

modulation schemes

Best throughput

How to adjust the rate ?

Rate adaptation plays a critical role to the throughput performance

Ideally, the transmission rate should be adjusted according to the ……

throughput decreases

Rate too high

Loss ratio increases

Rate too low

under-utilize the capacity

channel condition

How to estimate channel condition ?

SNR of the channel SNR of receiver…… no feedback in 802.11…… fluctuation of SNR……

Gauging how well the currently chosen rate performs (Statistics: transmission loss/success) not timely affected by random collisions (unnecessary downshift)easy to implement

receiversender

Modified frame

Symmetric link

Outline




Channel Dynamics

Wireless channel exhibits rich channel dynamics in practical scenarios

Random channel errorMobility-induced changeCollisions induced by

Hidden-terminals Multiple contending clients

Random collision congestion

Channel Dynamics

When should the transmission rate be updated?

too quick: perform bad when channel conditions fluctuate acutely

too slow: not in time (base on a relative long history)

The algorithm should be adaptive…

Equidistant Distribution

AccessPoint MN2

MN1

MN4

MN3

Node Contention Effects:

Collisions induced by: random backoff hidden terminal

Equidistant Distribution- random collisions

[1]

means:unnecessary

downshift

Equidistant Distribution- hidden terminal

[3]

means:unnecessary

downshift

Hidden terminal broadcast packets at a mild rate of 0.379Mbps continuously while other nodes begin with 11Mbps

Non-equidistant Distribution

AccessPoint

MN3

Nodes Diversity Effects:

collisionshidden terminalchannel diversity link capture

MN1

MN2

MN4

fairness

Non-equidistant Distribution- fairness

The poor-channel flows would consume more time and system resources

Different kinds of fairness: throughput fairness time-share fairness

In single-rate networks: equivalent

In multi-rate networks: time-share fairness is the one to be concerned

Outline




Rate Adaptation Algorithms List

1997 ARF1998199920002001 RBAR2002 OAR 2003 LA 2004 AARF MultiRateRetry AMRR 2005 ONOE LD-ARF SampleRate2006 CARA RRAA2007 AORA CHARMTBC ……

Classification of existing algorithms

Open-loop

Local channel estimation

Closed-loop

Packet transmissionsituation

RTSCTS

Beaconprobe response

Information channel estimation depends on:

LACHARM

AORA (idle slots)

ARFAARFMMR

AMRRONOE

LD-ARF(NACK)SampleRate

CARARRAA

RBAROARCARARRAA

CHARM(noise)

Classification of existing algorithms

Estimation (channel conditions)

Which layer to use

Action (how to adjust)

Which messages to use

How to estimate

sequentialrate

adjustment

bestrate

adjustment

PHYMAC hybrid datasignal

probe Noprobe

mappingcalculate

deterministicstatistical

Rate Adaptation Processing:

RBAROAR

ARF/AARFMMR/AMMR

LD-ARFSamleRate

CARA 、 RRAAAORA 、 ONOE

LA 、 CHARM

RBAROARCARA

LD-ARFRRAA

CHARM

ARF/AARFMMR/AMMR

LD-ARFSampleRate

AORACHARM

RRAA

RBAROARLA

CHARM

ARFAARF

LD-ARFMRRONOECARA

AMRRSampleRate

RRAAAORA

CHARM

ARFAARFMRR

AMRRONOE

LD-ARFCARARRAAAORA

RBARLA

SampleRateCHARM

Trend of rate adaptation algorithms

open-loop & statistics based: ARFclose-loop & SNR based (rts/cts): RBAR/OARstatistics & SNR based & adaptive: LAstatistics based & adaptive: AARFstatistics based & adaptive & estimate transimission time for differe

nt rates: SampleRatestatistics based & collisions avoid/detect: CARA 、 RRAA

• easy to implement

• can not react on the real time channel situation

• suffer from random collisions

• not compliant with current 802.11 networks

• The SNR is obtained based on” Symmetric link”, which is not accurate…

• improve the upshift performance

• Still suffer from random collision

• One probe every 10 frames

• Can react quickly to mobility

• Too sensitive to probe failure

• Differentiate the reasons for packet loss…

• Increasing load at some level…

Each algorithm has its own Achille’s heel…

Outline




Presentative old ones ARF RBAR OAR LA AARF SampleRate

ARF- How does it work ?

Use packet transmission situation to estimate the channel condition:

If two consecutive ACK frames are not received correctly, the second retry and subsequent transmissions are done at a lower rate and a timer is started.

When the number of successfully received ACKs reaches 10 or the timer goes off, a probe frame is sent at the next higher rate. However, if an ACK is NOT received for this frame, the rate is lowered back and the timer is restarted.

ARF- Does it work well ?

Advantages: Compliant with 802.11 All things can be done by the sender Easy to implement

Disadvantages: Suffer from random collisions and hidden ter

minals Constantly upshift try when channel conditio

n is stable Rate can only be adjusted step by step

RBAR- How does it work ?

Receivers control sender’s transmission rate:

RTS and CTS are modified to contain info on size and rate.

Uses analysis of RTS reception to estimate SNR and send choice back to sender in CTS.

Receiver picks rate based on pre-defined SNR thresholds.

RBAR- Does it work well ?

Advantages: Rate can be adjusted according to the real time

channel condition Do not need to adjust the rate step by step Will not suffer from random collisions and hidd

en terminalsDisadvantages:

not 802.11 compatible (modified RTS/CTS) The rate-SNR table is obtained based on a priori

channel mode SNR is not easy to get (most WLAN cards only ha

ve RSSI) RTS/CTS is seldom used (only when the frame is

too large…) RTS/CTS introduce extra load

OAR- How does it work ?

Make full use of coherence times, provide time-share fairness:

Improvement based on RBAR.

Coherence times are durations for which mobile stations have better-than-average channels.

Grant the user during a coherence time a channel access time that allows multiple packet transmissions (Fragment

mechanism in 802.11).

The poor-channel flows would consume more time and system resources

OAR- Does it work well ?

Advantages: Nodes with good channels can se

nd more packets while providing time-share fairness to all the nodes

The same as RBAR…Disadvantages:

The same as RBAR…

LA- How does it work ?

Assume that the channel is symmetric: Use RSSI to approximate SNR. Use SNR of the sender to approximate

SNR of the receiver. Each node maintains 12 dynamic RSS

thresholds. The thresholds are updated depending on

whether the transmission is successful.

Rate selection is based on both the RSS thresholds and number of retransmission attempts.

LA- Does it work well ?

Advantages: 802.11 compatible. RSSI is much more easy to get. Rate can be adjusted according to the

real time channel condition. Do not need to adjust the rate step by

step. Can adjust the rate during network

congestion.

Disadvantages: The symmetric assumption is dubitable. RSSI is quite different from SNR. suffer from random collisions and hidden

terminals.

AARF- How does it work ?

Dynamic adjust the upshift threshold of ARF:

Improvement based on ARF. To fix one existing problem of ARF (Constantly u

pshift try when channel condition is stable)

ARF

AARF

AARF- Does it work well ?

Advantages: Compliant with 802.11 All things can be done by the sender Easy to implement

Disadvantages: Suffer from random collisions and

hidden terminals Rate can only be adjusted step by step When channel condition gets better

quickly…it can not react quickly on it

SampleRate- How does it work ?Select rate by statistic information about the tr

ansmission time for each rate: maintain the expected transmission time for each ra

te and update it after each transmission. (wnd=1s--10s)

A frame is transmitted at the rate that currently has the smallest expected transmission time.

sends one probe packet at another randomly selected rate every 10 frames.

Downshift its rate every 4 consecutive transmission failure.

SampleRate- Does it work well ?Advantages:

React quickly to mobility Do not always need to adjust the rate

step by step Compliant with 802.11

Disadvantages: suffer from random collisions and

hidden terminals (time window) Can be very sensitive to probe failure

Outline




New Kids in this area

CARA - Collision Aware Rate Adaptation

RRAA – Robust Rate Adaptation Algorithm

CARA- the key idea

The primary contribution is to differentiate frame collisions from frame transmission failures caused by channel error.

focus on rate down only, rate up is the same as that of ARF

Two methods for identifying collisions:• Adaptive RTS probing• Identifying collision via CCA detection

CARA- Adaptive RTS Probing

Assumptions:All RTS transmission failures are

due to collisions.Transmission failure after RTS/CTS

must be due to channel errors.To reduce the signaling overhead,

RTS probing that enables an RTS/CTS exchange ONLY when a data frame transmission fails.

CARA-with default RTS Probing

Data frame transmitted without RTS/CTS.

If the transmission fails, RTS/CTS exchange is activated for the next retransmission. If this one fails again, then the rate is lowered.

If retransmission is successful, stay at the same rate and send next frame without RTS/CTS.

CARA –Identifying collision via CCA detection

If the wireless channel is busy while the expected ACK reception dose not start, the station conclude that a collision has just happened to its data transmission.

[2] A transmission failure detected by CCA to be a collision will not cause a RTS Probing

CARA- Performance evaluation 1

[2]

CARA-1: CARA with only default RTS Probing RTS/CTS: ARF scheme using RTS/CTS all the time

CARA- Performance evaluation 2

[2]

CARA-1: with only default RTS Probing CARA-2: with both default RTS Probing and CCA detecion

CARA-existing problems

When the channel condition is so bad that even the RTS can not be sent… CARA will be stuck there…

The network congestion can not be sensed, in which situation the rate should be downshifted…

when hidden terminals exist, it suffers from the drawback of RTS oscillation, which alternates on and off for RTS.

RRAA- the key idea

Short-term statistics to handle

• random loss• mobility• drastic changes

Adaptive RTS to handle

• collision

RRAA- loss estimation

Instead of single probe frame.Uses a loss estimation window and comp

utes the estimated loss ratio over the window (20-100ms).

Uses upper and lower loss threshold for each rate and estimated loss ratio to decide when to switch rates.

Otherwise, retain the current rate and continue sliding window

RRAA- Critical Loss Ratio (P*)

For any rate R, let the next lower rate be R_ and the next higher rate be R+.

With a loss ratio of P*, the throughput at R becomes the same as the loss-free throughput at R_.

RRAA- PMTL and PORI

We set PMTL = αP*(R), α ≥ 1• Decrease the rate when expected throughput

is more than PMTL • α = 1.25>1, to anticipate certain level of losses at

R_

PORI = PMTL(R+) / 2• increase the rate when expected throughput

is less than PORI • The loss ratio at the current rate R has to be

small enough such that the rate increase not quickly jump back to R

RRAA- Rate change

[3]

loss ratio thresholds:

RRAA- Adaptive RTS Filter

Selective use of RTS/CTS Tradeoff between overhead and benefits

of RTS

RTSwnd (RTScounter) RTSwnd is initially set to be 0. Window is increased by one when last frame lo

st without RTS (potentially due to a collision) When the last frame was lost with RTS or succe

eded without RTS, RTSwnd is halved (assume no collision involved).

RRAA- A daptive RTS Example

Collision may

occur

Nocollisio

n

More packets are sent with RTS if

the collision level is high

RRAA- Performance evaluation

[3]

Compared with ARF 、 AARF 、 SampleRateStatic client case:

• Throughput gains 0.3% ~ 67.4%

Mobile client case• Throughput gains 10.0% ~ 27.6%

Hidden-station case• Throughput gains 74% ~ 101%

RRAA- existing problems

[3]

How to decide the window size under different situation

The network congestion can not be sensed, when turn on RTS/CTS

Outline




My Opinion The ability to differentiate the reasons

for a frame loss. Link error (SNR) Random collisions/Hidden terminals Network congestion (available bandwidth?

CCA? Idle slots?) The speed of the adaptation (be

adaptive in different situation? What happens when introduce handoffs here?)

The interaction between link-layer rate adaptation and high-layer (TCP) rate adaptation

An Evaluation model

Thanks……

Questions?

Dynamic Rate Adaptation in IEEE 802.11 WLANs SongYiLin@ICT August 10, 2008.

Documents

robust rate adaptation

dynamic rate adaptation

rate options

chosen rate

mild rate

mbps slide

opinion slide

terminal slide