1 MAC Protocols for Wireless Networks: Interaction between Physical Layer and MAC Nitin H. Vaidya University of Illinois at Urbana-Champaign .

1

MAC Protocols for Wireless Networks:Interaction between Physical Layer and

MAC

Nitin H. VaidyaUniversity of Illinois at Urbana-Champaign

www.crhc.uiuc.edu/wireless

© 2004 Vaidya

2

Joint work with

Jason FuemmelerXue Yang

Romit RoyChoudhuryJungmin So

Pradeep KyasanurVenugopal Veeravalli

Funded in part by

National Science FoundationMotorola Center for Communications

NSF graduate fellowshipVodafone graduate fellowshipMotorola graduate fellowship

3

Our Research

MAC/Routing/Transport protocols for wireless

Distributed algorithms (leader election, clock sync, ...)

Misbehavior in wireless networks

4

Wireless Ad Hoc Networks

Formed by wireless hosts that may be mobile

Without necessarily using infrastructure

Routes between nodes may potentially contain multiple hops

5

Ad Hoc Networks

EA

B CD

X

Z

Ad hoc connectivity

6

Hybrid Environments

Infrastructure + Ad hoc connectivity

EA

B CD

BS1 BS2

X

Z

infrastructure

Ad hoc connectivity

7

Wireless Capacity

Wireless capacity limited

In dense environments, performance suffers

How to improve performance ?

8

Improving Wireless Capacity

Exploit physical resources

Exploit diversity

Examples ...

9

Add Spectrum

More bandwidth

Example: Multiple channels in IEEE 802.11

10

Improve Spatial ReusePower/Rate Control

A B C D

A B C D

11

Exploit Infrastructure

Infrastructure provides a tunnel to forward packets

EA

B CD

BS1 BS2

X

Z

infrastructure

Ad hoc connectivity

12

Exploit Antennas

Diversity antenna

Steered beam directional antenna

13

Path Diversity

Multiple paths to a destination

Multiple next-hops to a destination

14

Exploiting Diversity

Exploiting physical layerrequires suitable protocols

Routing

Medium access control (MAC)

Link

Network

Transport

PhysicalLayer

Upper layers

15

Medium Access Control (MAC)

MAC protocols coordinate wireless channel access

May be centralized or distributed

Distributed protocols suit ad hoc & hybrid networks

16

MAC Protocols

Need to design MAC protocols to exploitphysical layer capabilities

Proof by example …

17

Outline

CSMA Protocols:

Warning: Work-in-progress

Directional antennas & Multiple channels:

Brief discussion, time permitting

18

Carrier Sense Multiple Access (CSMA)

Listen-before-you-talk

A host may transmit only if the channel is idle

19

Carrier Sense Multiple Access (CSMA)

Implementation using Carrier Sense (CS) threshold

If received power < CS threshold Channel idle

Else channel busy

20

Carrier Sense Multiple Access (CSMA)

D perceives idle channel although A is transmitting

AB C

D

distance

po

we

r

D’s CS Threshold

21

Carrier Sense Multiple Access (CSMA)

D perceives busy channel when A transmits

AB C

D

distance

po

we

r D’s CS Threshold

22

Transmission Reliability

Reliability depends on SIR = S / I (ignoring noise)

SIR requirement depends on modulation scheme (rate)

AB C

D

distance

po

we

r

SI

23

Warning

SIR and SINR used in this talk interchangeably

SINR = S / (I + N) S = signal I =

interference N = noise

SIR = S / I (when Noise ignored)

24

Packet Transmission

The transmitter must choose Transmit rate Transmit power CS Threshold

Prior work typically focuses on choosing 1 parameter,

keeping others fixed

Jointly adapting 2-of-3 potentially better strategy

25

Joint Adaptation2-of-3

Which 2-of-3 ?

Practical considerations might eventually prove one combination superior

Our on-going work

Transmit Rate & CS threshold Transmit Power & CS threshold

26

Joint Control:

Transmit Power and CS Threshold

27

Assumed Protocol Requirements

Distributed decision-making

Choose transmit power to meet desired SINR (despite future transmitters)

Not destroy on-going transmissions

E to F

A to B

C to D

28

Observation 1

AB

distance

po

we

r

Higher transmit power increasesreceived signal power S at the intended receiver

Low transmit power

S

29

Observation 1

AB

distance

po

we

r

High transmit power

Higher transmit power increasesreceived signal power S at the intended receiver

S

30

Observation 2

AB J

distance

po

we

r

For a fixed CS threshold,high transmit power closest interferer farther

Decreases interference I

Low transmit power

31

Observation 2

AB J

distance

po

we

r

For a fixed CS threshold,high transmit power closest interferer farther

Decreases interference I

High transmit power

32

Observations 1 & 2

Higher transmit powerincreases signal S,decreases interference Iat the intended receiver

SIR = S/I improves

Double dipping!

33

Towards a Protocol

Node A transmits a packet to B

Interference sources:1.Transmitters already active before A begins

talking2.Transmitters that transmit after A begins talking

A B C

D

distance y

x

z

E

F

34

Interference from “Already Active” Sources

E is transmitting before A starts transmitting Such interference by design less than CS

threshold

A B C

D

E

F

35

Interference from Future Sources

C begins transmitting after A starts its transmission

C must have sensed idle channel despite A’s transmission

A B C

D

E

F

36

Transmit Power & CS Threshold Selection[Fuemmeler04]

How to ensure that a transmission will be completed reliably ?

Analysis suggest that transmit power and CS threshold should be adapted together

37

Notation

g(x) = gain at distance x from the transmitter

Received power = transmit power * gain

Distance x

G

ain

g(x

)

38

Notation

Pt = transmit power Pcs = Carrier-sense threshold

SINR threshold for reliability = noise

k = estimated number of interferers

39

Power Control & CS ThresholdTwo Constraints

= design constant

40

Sensitivity Constraint

Loud transmitters should be sensitive Whisperers may be insensitive

Limits interference to ongoing transmissions

41

Interference Constraint

CS threshold = interference margin per interferer

Initial interference < CS Threshold it can be modeled as an equivalent future interferer (alternatively it can be added to η)

42

Interference Constraint

Chosen transmit power also serves as an indication of tolerable interference to the corresponding transmission

Chosen transmit power serves dual roles of limiting interference to others, and limiting interference to self

43

Performance Evaluation

Static Protocol

44

Static Protocol

Pick a fixed value of k

Solve the two constraints to obtain transmit power and CS threshold

45

Ring Topology

N nodes spaced equally around a ring

Each node transmits to its immediate counterclockwise neighbor

Symmetric topology

46k = number of interfererers

Ag

gre

gat

e t

hro

ug

hp

ut

(kb

ps)

N = 128

N = 8

Ring Topology(Static Protocol: k fixed)

Single flow throughput = 767 kbps SINR > 10 dB

47

Random Topology

x = Txo = Rx

48

Random Topology(Fixed Receive Power)

CS Threshold (Watts)

Ag

gre

gat

e t

hro

ug

hp

ut

(kb

ps)

x = Topology 1o = Topology 2

49

Random Topology(Fixed Transmit Power)

CS Threshold (Watts)

Ag

gre

gat

e t

hro

ug

hp

ut

(kb

ps)

x = Topology 1o = Topology 2

50

Static Protocol(fixed k)

Ag

gre

gat

e t

hro

ug

hp

ut

(kb

ps)

x = Topology 1o = Topology 2

k

51

Failure Rate(Static Protocol: fixed k)

Pa

cke

t fa

ilure

ra

tex = Topology 1o = Topology 2

k

52

Performance Evaluation

Dynamic Protocol

53

Dynamic Protocol

Estimate k = number of interferers

Solve the two constraints to determinetransmit power and CS threshold

54

Parameter k

k : the tunable knob

Co-location approximation introduces error: Error handled by increasing dynamic range of

k

Intuitively, k is a measure of total interference as a multiple of that from a “worst-case interferer”

55

Estimation of Optimal k

Preliminary protocol : Details omitted here

Obtain an initial estimate on “reasonable” non-zero k starting at k = 0:

Increase k on packet failure Decrease k on success

Gradient-descent search: Motivated by the throughput and failure rate

versus k curves

56

Throughput Comparison (kbps)Random Topologies

Protocol Topology 1 Topology 2

Fixed receive power

3071 3710

Fixed transmit power

3627 3944

Static k 4780 5218

Dynamic k 4955 5365

57

Dynamic Protocol

Allows each node to use different k,transmit power and CS threshold

Can improve performance compared toglobally constant parameters

58

Ongoing Work

Current dynamic protocol slow to adapt,may not work well in presence of rapid mobility

Small scale effects (fading) not accounted for May potentially be accounted for by adding

interference margin

Need to co-exist with other dynamic mechanisms such as backoff in 802.11

Fairness issues

59

Related Work

Power-Controlled Multiple Access (PCMA):Monks et al. Receiver-based mechanism using busy-tone (BT) Data power & BT power chosen independently We use data for both purpose: only 1 power to choose

Zhu et al. (Intel): Globally constant CS threshold

Muqattash et al.: Need separate control channel, or a priori coordination

Congnitive radios: Sahai et al. (Allerton’04)

60

Summary

Joint adaptation of CS threshold & transmit power beneficial

Need further work to make protocol more robust

61

Impact of Protocols Overheads onCS Threshold

Skip?

62

Goal

To identify how protocol overheads affect optimal carrier-sense threshold

Assume fixed transmit power

63

Observation

AB

po

we

r

Higher CS Threshold brings closest interferer closer

Higher interferences at B Lower SIR Lower transmit rate

I J

CS threshold

64

Small Carrier Sense Threshold(Sensitive Nodes)

Sensitive transmitters must be far away

With fixed transmit power, interference reduces

Transmit rate can be increased

65

Small Carrier Sense Threshold(Sensitive Nodes)

AB X

Y

PQ

Smaller CS threshold Fewer simultaneous transmissions, but at higher rate

66

Large Carrier Sense Threshold(Insensitive Nodes)

AB

Higher CS threshold More simultaneous transmissions, but at lower rate

M

N

R

S

67

Optimal Carrier Sense Threshold

CS Threshold = 0 Only one transmission at a time

CS Threshold = ∞ All hosts transmit together

Optimal somewhere in between

Protocol overhead affect the optimal CS threshold …

68

Impact of Protocol Overheads

Two components

Rate-independent overhead: Propagation delay, backoff slot in 802.11

Rate-dependent overhead: Collision cost

69

Impact of Protocol Overhead

Rate-independent

Rate-dependent

Single rate Rtransmissionat a time

Two rate R/2transmissionsat a time

70

Impact of Protocol Overhead

Can be beneficial to havemore simultaneous transmission atsmaller transmit rate

One approach: Divide bandwidth into multiple channels Previously investigated by others

Our approach: Increase “density” of transmitters

71

Impact of Protocol Overheads [Yang05]

Increase density of transmitters,while decreasing transmission rate

Reduces rate-independent overhead

Conclusion:

Optimal CS threshold larger whenrate-independent overhead is considered

72

Outline

CSMA Protocols

Directional Antennas & Multiple Channels

73

Antenna Capabilities

Fixed beam antennas Omnidirectional

antennas, directional antennas with a fixed beam pattern

Movable beam antennas Switched, steered,

reconfigurable, adaptive, smart …

74

Antenna Capabilities

Protocols designed for fixed beam antennas inadequate with movable beam antennas

Movable directional beams introduce additional benefits and challenges Deafness Neighbor discovery Utilizing long links

75

Deafness [RoyChoudhury04]

S transmitting to D A cannot carrier-sense S A sends packet to S, but gets no response

D

A

S

76

Deafness

Two difficult-to-distinguish possibilities Collision at S S is looking in a different direction

D

A

S

77

Deafness

Two difficult-to-distinguish possibilities Collision at S S is looking in a different direction

Need different responses to the two events

Similarities to the problem of distinguishing between packet losses due to congestion and corruption in context of TCP-over-wireless.

78

Deafness

Deafness arises due to directionality& the ability to change beam directions

Need to adapt MAC protocol to handle deafness

79

Multi-Channel Environments

Skip

80

Multi-Channel Environments

Multiple Channels available in IEEE 802.11 3 channels in 802.11b 12 channels in 802.11a

Utilizing multiple channels may improve throughput

1

defer

1

2

Single channel Multiple Channels

81

Issues

Using k channels does not necessarily translate into proportional throughput improvement

Hosts limited by number of transceivers

Nodes on different channels cannot talk to each other

1 2

82

Multi-Channel Hidden Terminals

Consider the following naïve protocol

Static channel assignment

Communication takes place on receiver’s channel

Sender switches to receiver’s channel to transmit

83

Multi-Channel Hidden Terminals

A B C

Channel 1

Channel 2

A transmit on channel 1

C is on channel 2

84

Multi-Channel Hidden Terminals

A B C

C switches to channel 1 and transmits

Channel 1

Channel 2

Collision occurs at B

85

Multi-Channel Environments[So04,So03,Kyasanur04]

Protocols need to be designed to maximize channel utilization despite

Transceiver limitations

Multi-channel hidden terminals

Switching delays

86

Conclusion

87

Conclusion

Newer chipsets & radios are allowing more flexibility

Timescale over which parameters can be changed shrinking

•Ideal scenario: Per-packet switching

Makes it possible to exert more control on the physical layer

88

Conclusion

Necessary to exploit physical layer capabilities to maximize performance

Need suitable MAC and Routing protocols

Many physical layer characteristics interesting: Transmit power, transmit rate, carrier-sense

threshold Antenna Channel diversity Switching delays Multi-user diversity

89

Conclusion

State-of-the-Art:

Protocol mechanisms designed for particular physical layer capability

Holy grail:

Adaptive protocols that learn physical layer characteristics and adapt to them

90

Thanks!

nhv@uiuc.edu

www.crhc.uiuc.edu/wireless

91

Thanks!

nhv@uiuc.edu

www.crhc.uiuc.edu/wireless

92

Additional Slides

93

Physical Layer ParametersAffecting MAC

Transmit rate Transmit power Antenna Single or multi-channel Single or multi-band Switching delays …

94

Analysis (details omitted)[Fuemmeler04]

Approximations to facilitate analysis:

Noise known

Sender & receiver effectively co-located

Sender senses same power level as the receiver

Approximation error handled by adaptation mechanism in the protocol

95

Interference

From already on-going transmissions From future interferers

Interference from on-going transmissions can be modeled explicitly or as an equivalent future interferer

We choose the latter option, which fortuitously works out (but former can be used as well)

96

Illinois Wireless Wind Tunnel

New project : To develop an environment for repeatable experiments on wireless networks

To be built in an anechoic chamber

“Scaling” of wireless network and environment to account for differences in mobility and obstacle size in the real network and experimental network

97

Other Research

Energy efficient protocols for wireless networks

TCP over wireless networks

Handling misbehavior in wireless networks

98

Ring Topology(Transmit Power fixed)

CS Threshold (Watts)

Ag

gre

gat

e t

hro

ug

hp

ut

(kb

ps)

N = 128

N = 8