Medium Access and Interference Cancellation: Protocol and Evaluation - - PowerPoint PPT Presentation

Medium Access and Interference Cancellation:

Protocol and Evaluation

Abishek Sankararaman and François Baccelli

1

2

Introduction

• Focus: Medium Access problem in Ad-hoc networks.
• Aim: Propose simple implementable protocols by

incorporating observations and results from Information Theory.

3

Motivation

• Some key features of emerging wireless networks

Dense Decentralized Control

D2D Communication Vehicular Communication (802.11p)

4

Motivation

• Some key features of emerging wireless networks

Dense Decentralized Control

D2D Communication Vehicular Communication (802.11p)

Managing Interference is a key challenge - primarily handled through Medium Access Control algorithms in ad-hoc networks.

5

Popular Medium Access Solution

• CSMA (Carrier Sense Multiple Access) - 802.11 standards
• ‘Interference as Noise’ (IAN) paradigm.

6

Popular Medium Access Solution

• CSMA (Carrier Sense Multiple Access) - 802.11 standards
• ‘Interference as Noise’ (IAN) paradigm.

Rxr No Interfering Transmitters

CSMA/CA Schematic

Txr T R Guard Zone around a scheduled receiver

• Simple Distributed Implementation (RTS/CTS)

7

Ad-hoc Network - Interference Channel

• Capacity and achievability is unknown in general.

T1 T2 R1 R2 1 1 a a

2 user interference channel

8

Results from Information Theory

• Capacity and achievability is unknown in general.
• , IAN is optimal.
• , SIC (Successive Interference Cancellation)

decoding is optimal. (Receivers treat the transmitters as a MAC channel).

T1 T2 R1 R2 1 1 a a

2 user interference channel a → 0 a → ∞

9

Successive Interference Cancellation

T1 T2 T3 Rxr

Received Powers Pi, Rates Ri

C ⇣

Pi N0+P3

j=i+1 Pj

⌘ ≥ Ri , i ∈ {1, 2, 3}.

Pi > Pj ∀i < j

where , C(x) = 1

2 log2(1 + x)

Gaussian Codebook

10

Successive Interference Cancellation

• Separation of Powers needed to ensure decodability !

T1 T2 T3 Rxr

Received Powers Pi, Rates Ri

C ⇣

Pi N0+P3

j=i+1 Pj

⌘ ≥ Ri , i ∈ {1, 2, 3}.

Pi > Pj ∀i < j

where ,

Gaussian Codebook

C(x) = 1

2 log2(1 + x)

11

SIC - Separation of Powers

T1 T2 T3 Rxr

Received Powers Pi, Symmetric Rate R

Pi > Pj ∀i < j

Pi N0+I+Pk

j=i+1 Pj ≥ Q

12

SIC - Separation of Powers

• Separation of Powers needed to ensure decodability !

T1 T2 T3 Rxr

Received Powers Pi, Symmetric Rate R

Pi > Pj ∀i < j

,

Pi N0+I+Pk

j=i+1 Pj ≥ Q

Pi

Pi+1 needsto be significantly larger than

13

Main Idea of an Improved Protocol

General capacity region is unknown

T1 T2 T3 T4 T5 R1 R2 R3 R4 R5

14

Main Idea of an Improved Protocol

Any pair of links form a 2 user interference channel.

General capacity region is unknown

T1 T2 T3 T4 T5 R1 R2 R3 R4 R5

a15 a51 a11 a55

15

Main Idea of an Improved Protocol

Any pair of links form a 2 user interference channel.

General capacity region is unknown

T1 T2 T3 T4 T5 R1 R2 R3 R4 R5

a15 a51 a11 a55

If and , then

• CSMA/CA will schedule at most one link.

a15 >> a55 a51 >> a11

16

Main Idea of an Improved Protocol

Any pair of links form a 2 user interference channel.

General capacity region is unknown

T1 T2 T3 T4 T5 R2 R3 R4 R5

a15 a51 a11 a55

R1

If and , then

• CSMA/CA will schedule at most one link.
• However if the receivers can perform SIC, then both links could

potentially be scheduled.

a51 >> a11

a15 >> a55

17

Main Idea of an Improved Protocol

Any pair of links form a 2 user interference channel.

General capacity region is unknown

T1 T2 T3 T4 T5 R1 R2 R3 R4 R5

a15 a51 a11 a55

Need to define when a cross interference is ‘strong’.

R1

If and , then

• CSMA/CA will schedule at most one link.
• However if the receivers can perform SIC, then both links could

potentially be scheduled.

a51 >> a11

a15 >> a55

18

CSMA 1-SIC Protocol

Schematic of CSMA/CA

T1 R1 T2 T3 R3 R2

Guard Zone around a receiver

T1 R1 T2 T3 R3 R2

Guard Zone around a receiver

Schematic of proposed CSMA 1-SIC protocol.

T4 R4

19

CSMA 1-SIC Protocol

Schematic of CSMA/CA

T1 R1 T2 T3 R3 R2

Guard Zone around a receiver

T1 R1 T2 T3 R3 R2

Guard Zone around a receiver

Schematic of proposed CSMA 1-SIC protocol.

T4 R4

Separation of Received Powers - Donut Shaped Guard Zone.

20

CSMA 1-SIC Signaling

Tx Rx

Assume time-slotted system.

21

CSMA 1-SIC Signaling

Tx Rx

Each link (Tx) samples a Random Timer Value in say [0,1]

t0 t3 t1 t6 t2 t4 t5

Tx ‘senses’ channel till timer expires.

22

CSMA 1-SIC Signaling

Tx Rx

t0 t3 t1 t6 t2 t4 t5

Tx ‘senses’ channel till timer expires.

Send RTS

23

CSMA 1-SIC Signaling

Tx Rx

t0 t3 t1 t6 t2 t4 t5

Rx ‘senses’ to hear a RTS.

Send CTS

24

CSMA 1-SIC Signaling

Tx Rx

t0 t3 t1 t2 t4 t5

Rx ‘senses’ to hear a RTS.

Send CTS

25

CSMA 1-SIC Signaling

Tx Rx

t0 t3 t1 t2 t4 t5

Tx broadcasts ‘Established’ to silence nearby receivers

Send Established

26

CSMA 1-SIC Signaling

Tx Rx

t0 t3 t1 t2 t4 t5

Send Established

Tx broadcasts ‘Established’ to silence nearby receivers

27

CSMA 1-SIC Signaling

Tx Rx

t0 t3 t1 t2 t4 t5

Tx transmits ‘Established’ signal

Established

28

CSMA 1-SIC Signaling

Tx Rx

t0 t3 t1 t2 t4 t5

Rx transmits ‘Blocked’ signal to silence all other strong interferers

29

CSMA 1-SIC Signaling

Summary

• Randomized Protocol (Timers Chosen randomly).
• 2 parameters to tune.
• Guarantees to any scheduled receiver that there will be at-

most one ‘strong’ interfering transmitter.

30

CSMA k-SIC Protocol

• The separation of powers leads to 2k parameter protocol.
• One can then develop a similar signaling algorithm.

One$Interfering$Transmi/er$Allowed$ No$Interfering$Transmi/er$ r1$ r2$ r3$ r4$

31

CSMA /CA Versus CSMA 1-SIC

Non-Monotonicity

t1 t2 t4 t3

CSMA/CA

t1 t2 t4 t3 t1 t2 t4 t3

CSMA 1-SIC

32

CSMA /CA Versus CSMA 1-SIC

Non-Monotonicity

t1 t2 t4 t3

CSMA/CA

t1 t2 t4 t3 t1 t2 t4 t3

CSMA 1-SIC

• Averaged over timer values however, CSMA 1-SIC schedules

more links.

33

CSMA /CA Versus CSMA 1-SIC

Non-Monotonicity

t1 t2 t4 t3

CSMA/CA

t1 t2 t4 t3 t1 t2 t4 t3

CSMA 1-SIC

• Averaged over timer values, CSMA 1-SIC schedules more links.
• This also means, that the interference levels are higher.

34

Performance Evaluation - Setup

• A Stochastic Network Model to compare the gains in adopting the

protocol.

Dipole Network Model -

Each Tx has an unique Rx. Tx form a PPP and the corresponding Rx is located at an uniform and independent angle away.

35

Performance Evaluation - Setup

• A Stochastic Network Model to compare the gains in adopting the

protocol.

Dipole Network Model -

Each Tx has an unique Rx. Tx form a PPP and the corresponding Rx is located at an uniform and independent angle away.

No Power Control. All scheduled Tx transmit at unit power. Fading - Channel between any pair of devices is random and symmetric Path loss - l(r) = r-4

Fxyl(||x − y||)

36

Performance Evaluation - Metrics

• The metrics

MAP - (Medium Access Probability (pa) )

37

Performance Evaluation - Metrics

• The metrics

MAP - (Medium Access Probability (pa) ) Success Density - (Fraction of scheduled links successful (ps) )

SINR > Q

38

Performance Evaluation - Metrics

• The metrics

MAP - (Medium Access Probability (pa) ) Success Density - (Fraction of scheduled links successful (ps) ) Throughput - (Fraction of links that get scheduled and are successful)

SINR > Q

39

Performance Evaluation - Metrics

• The metrics

MAP - (Medium Access Probability (pa) ) Success Density - (Fraction of scheduled links successful (ps) ) Throughput - (Fraction of links that get scheduled and are successful)

SINR > Q

λpspa Throughput =

40

Performance Evaluation - MAP

0.2 0.4 0.6 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Rayleigh Fading γ MAP 0.2 0.4 0.6 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 No fading γ MAP CSMA IAN CSMA 1−SIC CSMA IAN CSMA 1−SIC

In large random networks, more links get scheduled on average.

0.2 0.4 0.6 0.8 0.5 0.6 0.7 0.8 0.9 Rayleigh Fading γ Success Probability 0.2 0.4 0.6 0.8 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 No Fading γ Success Probability CSMA IAN CSMA 1−SIC CSMA IAN CSMA 1−SIC

41

Performance Evaluation - Success Probability

CSMA 1-SIC has higher interference since it schedules aggressively !

42

Performance Evaluation - Throughput

Nonetheless, CSMA 1-SIC has higher throughput !

0.5 1 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 Rayleigh Fading γ Throughput 0.5 1 0.16 0.18 0.2 0.22 0.24 0.26 0.28 No Fading γ Throughput CSMA IAN CSMA 1−SIC CSMA IAN CSMA 1−SIC

0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55

λ Optimal Throughput CSMA IAN CSMA 1−SIC CSMA 1−SIC (Constrained) Aloha + 1−SIC

43

Throughput Optimization

λpspa Throughput =

Donut shaped Guard zone is indeed required

Performing SIC improves throughput.

0.6 0.8 1 1.2 0.12 0.14 0.16 0.18 0.2 0.22 0.24

Rayleigh Fade Threshold Optimal Throughput

0.6 0.8 1 1.2 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3 0.32 0.34

No Fade Threshold Optimal Throughput

CSMA IAN CSMA 1−SIC CSMA 1−SIC (Constrained) CSMA IAN CSMA 1−SIC CSMA 1−SIC (Constrained)

44

Throughput Optimization

λpspa Throughput =

Performing SIC improves throughput.

Donut shaped Guard zone is indeed required

45

Higher order SIC

• 2k parameters to choose and optimize over.
• Expect some form of ‘diminishing returns’ by increasing k.
• No clean performance comparisons with CSMA IAN yet.

One$Interfering$Transmi/er$Allowed$ No$Interfering$Transmi/er$ r1$ r2$ r3$ r4$

46

Open Problems - Computational

Computation of densities (MAP)

• Exact computation is hard for even regular CSMA/CA.
• Matérn like approximation
• An incoming link must compete with all other links having a smaller timer value

regardless of whether they were scheduled.

47

Open Problems - Computational

Computation of densities (MAP)

• Exact computation is hard for even regular CSMA/CA.
• Matérn like approximation
• An incoming link must compete with all other links having a smaller timer value

regardless of whether they were scheduled.

• Hard to compute even under this approximation !
• A link is scheduled only if a transmitter does not ‘kill’ a receiver with smaller

timer value.

• Extremal shot noise of the point process formed from all possible k+1 tuples of

the points of a PPP is needed.

48

Open Problems - Physical

5 10 15 20 25 30 5 10 15 20 25 30

IAN Scheduling Outcome

5 10 15 20 25 30 5 10 15 20 25 30

1−SIC Scheduling Outcome

‘Jamming Regime’

49

Summary and Conclusions

• Looked at an improved paradigm for designing protocols.

Implementable distributed protocols from simple observations.

• A more fundamental question - ‘What is a good protocol’ ?

‘Fairness Efficiency’ tradeoff for spatial wireless resource.