1

15-744: Computer Networking

L-19 Cognitive Wireless Networks

Cognitive Wireless Network

• Optimize wireless networks based context information

• Assigned reading• White spaces• Online Estimation of Interference (3 sections)• DIRC: Optimizing Directional Antennas (2

sections)• Optional reading

• Centaur: Hybrid optimization

2

Overview

• Background on cognitive wireless

• Building conflict graphs

• DIRC: directional antennas

• White space networks

3

Cognitive Wireless Networks

• Performance of wireless network depends on deployment environment and use• Try for any complex system, but especially for

wireless given interactions with physical world• Node density, mobility, physical infrastructure,

traffic load, wireless technologies, ..• Often not practical to hand tune the system

• Ok on campus: fairly predictable, strong control• Home, hotspots, vehicular, industrial, ..

4

2

Generic Cognitive System

System

SensingMeasurement

ModelingAnalysis

OptimizationManagement

ActuationControl

AbstractionsIntent‐basedManagement

Stability and manageability

Scale/EfficiencyHeterogeneity

How about Cognitive Wireless?

• Sensing and measurement• Limited by capabilities of wireless NICs• Low resolution, lack of calibration• Cost of exchanging measurements

• Modeling and analysis• Abstraction of link, network• Minimizing cost in building, maintaining model• Basis for “CS” optimization, global

management

Cognitive Wireless continued

• Optimization and management• Focus is often network capacity plus fairness,

but can consider other factors• Often uses heuristics (always exponential)

• Actuation and control• Execute the plan devise above• Minimize overhead in distributing instructions,

coordination in execution• Low resolution control, lack of calibration

7

Cognitive Wireless Networking

• Very common for tuning of low level PHY and MAC parameters• Optimization if local – one or two nodes• Example: rate adaptation

• More global optimizations, e.g. dealing directly with interference• Use a conflict graph as the abstraction

• Dynamic spectrum access• Adapt to presence of primary users

8

3

9

Rate Adaptation Goal

1 Mbps

2 Mbps5.5 Mbps11 Mbps18 Mbps24 Mbps36 Mbps48 Mbps

54 Mbps

10

Existing Approaches

• Trial-and-error Probing• Use recent packet loss history to pick rate• Short history oversensitive to loss• Long history slow to react to changes• How to differentiate poor SINR from collision?

• SINR: Signal to Interference + Noise Ratio• Need to know SINR at the receiver• Use RTS/CTS to learn channel quality

• Overhead RTS/CTS rarely used in practice

RTS

Data

CTST R

What are the models used?

11

CHARM• Channel-aware rate selection algorithm• Transmitter passively determines SINR at

receiver by leveraging channel reciprocity• Determines SINR without the overhead of active

probing (RTS/CTS)• Select best transmission rate using rate table

• Table is updated (slowly) based on history• Needed to accommodate diversity in hardware

• Jointly considers problem of transmit antenna selection

12

SINR: Noise and Interference

• Noise• Thermal background radiation• Device inherent

• Dominated by low noise amplifier noise figure

• ~Constant• Interference

• Mitigated by CSMA/CA• Reported as “noise” by NIC

SINR = RSS

Noise + ∑ Interference

4

• Leverage reciprocity to obtain path loss• Compute path loss for

each host• On transmit:

• Predict path loss based on history

• Select rate & antenna• Update rate thresholds

14

Time

1 Mbps

2 Mbps

5.5 Mbps

11 Mbps

SINR

CHARM: Channel-aware Rate Selection

TR

Per-node History

Time

Some Examples

• More global optimizations requires more complex models

• Routing and opportunistic forwarding based on signal propagation conditions

• Maximizing spatial reused based on transmit power adaptation, directional antennas, carrier sense tuning, …

• Packet scheduling as a replacement for carrier sense

15

Overview

• Background on cognitive wireless

• Building conflict graphs

• DIRC: directional antennas

• White space networks

16

What is a Conflict Graph?

• A graph that summarizes the interference between wireless links

• Nodes are wireless links and edge indicate that the links interfere

• Many variants depending on the nature of the network and optimization• Definition of interference, e.g. MAC vs PHY• Directed versus undirected edges• Annotations of the edges

17

5

18

Example Conflict Graph

19

Constructing CGs

• Passive techniques• Monitors to collect traces• Interference inferencing algorithms• E.g., Jigsaw, WiT

• Active techniques• Send broadcast traffic in parallel• Use Broadcast Interference Ratio (BIR) metric• Measurements of bandwidth• Typically off line

20

Tradeoffs

� Active: better accuracy, captures weak interference

� Passive: low overhead, no downtime, no client mods

� Micro-probing: combine the best of both

21

Micro-Probing Architecture

6

22

Basics of Micro-Probing

• Testing for CS interference• AP i initiates broadcast transmissions at some set of

times• AP j initiates transmission at same times with offset• Use MAC service time (MST) estimates

• Testing for collision interference• Initiate transmission between AP i and client• AP j sends broadcast frame at same instant

Example: Transmit Power Control

• Adjust transmit power for each transmission to maximize spatial reuse• Maximize simultaneous transmission

• What model should we use?• “Circle model”: each transmitter has a

transmission and interference range• Widely used in (bad) simulators

• SINR model: packet reception depends on SINR at the receiver

23

Transmit Power Control – Take 1

• Idea: reduce transmit range to minimum needed to reach destination so interference is minimized

• Right?

24

Using SINR – Take 2

• Decreasing transmit power reduces interference for other nodes but makes transmission more vulnerable to interference!

• Should increase transmit power to maximize chances of reception

• Right?• But you cannot have it both ways?

SINR =Signal

Noise + ∑ Interference

7

Automatic Power Control

• Any transmission creates interference on all other links• Must consider pair-wise interference between all links

• Concurrent transmission is possible ifSINR1+SINR2 ≥ 2*SINRthreshold

• How use this observation for a distributed power control algorithm?

0 0.5 1 1.50

0.2

0.4

0.6

0.8

1

CD

F of

Lin

k Th

roug

hput

Link Throughput (Mbps)

Iterative, 0.72Min Power, 0.62Equal Power, 0.46

n2

AP1

n1

AP2

L11 L22

L12

L21

Improving Spatial Reuse

• Create pairwise conflict graph for “links” in the wireless network• Edges defined using the SINR model• Uses path loss – very efficient

• Remove links by adjusting transmit power• Based on equation of previous slide

27

Distributed Automatic Power Control

• Distribute algorithm by having each node operate on a “local” conflict graph• Includes transmitters up to two hops away• Direct observations + gossiping of signal strength

• Final step is to adjust the CCA threshold to allow simultaneous transmissions (messy)

• Implementation is very tricky• Calibration of registers, limited control, deafness,

mobility, etc.

28Me Peer Potential

InterfererPeerPotential

Interferer

Overview

• Background on cognitive wireless

• Building conflict graphs

• DIRC: directional antennas• Presentation Richard

• White space networks

29

8

Overview

• Background on cognitive wireless

• Building conflict graphs

• DIRC: directional antennas• Presentation Richard

• White space networks• Based on slides by authors

45

dbm

Frequency

‐60

‐100

“White spaces”

470 MHz 700 MHz

What are White Spaces?

46

0 MHz

7000 MHz

TV ISM (Wi-Fi)

700470 2400 51802500 5300

are Unoccupied TV ChannelsWhite Spaces

54-90 170-216

Wireless Mic

TV Stations in America

•50 TV Channels

•Each channel is 6 MHz wide

•FCC Regulations*•Sense TV stations and Mics •Recent ruling allows use of databases•Portable devices on channels 21 - 51

47

Hig

her

Freq

uenc

y

Wi-Fi (ISM)

Broadcast TV

The Promise of White Spaces

48

0 MHz

7000 MHz

TV ISM (Wi-Fi)

700

470

2400

5180

2500

5300

54-90 174-216

Wireless Mic

More Spectrum

Longer Range

Up to 3x of 802.11g

at least 3 - 4x of Wi-Fi

9

White Spaces Spectrum AvailabilityDifferences from ISM(Wi-Fi)

49

FragmentationVariable channel widths

1 2 3 4 51 2 3 4 5

Each TV Channel is 6 MHz wide Use multiple channels for more bandwidthSpectrum is Fragmented

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

1 2 3 4 5 6 >6

Frac

tion

of S

pect

rum

Se

gmen

ts

# Contiguous Channels

Urban

Suburban

Rural

White Spaces Spectrum Availability

Differences from ISM(Wi-Fi)

50

FragmentationVariable channel widths

1 2 3 4 5

Location impacts spectrum availability Spectrum exhibits spatial variation

Cannot assume same channel free everywhere

1 2 3 4 5

Spatial Variation

TVTower

White Spaces Spectrum Availability

Differences from ISM(Wi-Fi)

51

FragmentationVariable channel widths

Incumbents appear/disappear over time Must reconfigure after disconnection

Spatial VariationCannot assume same channel free everywhere

1 2 3 4 5 1 2 3 4 5Temporal Variation

Same Channel will not always be free

Any connection can bedisrupted any time

Channel Assignment in Wi-Fi

52

Fixed Width Channels Optimize which channel to use

1 6 11 1 6 11

10

Spectrum Assignment in WhiteFi

53

1 2 3 4 5

Spatial Variation BS must use channel iff free at clientFragmentation Optimize for both, center channel and width

1 2 3 4 5

Spectrum Assignment Problem

Goal Maximize Throughput

Include Spectrum at clients

Assign Center Channel

Width&

Accounting for Spatial Variation

54

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

=1 2 3 4 5 1 2 3 4 51 2 3 4 51 2 3 4 5

Intuition

55

BSUse widest possible channelIntuition

1 3 4 52Limited by most busy channelBut

Carrier Sense Across All Channels

All channels must be freeρBS(2 and 3 are free) = ρBS(2 is free) x ρBS(3 is free)

Tradeoff between wider channel widths and opportunity to transmit on each channel

Discovering a Base Station

56

Can we optimize this discovery time?

1 2 3 4 5

Discovery Time = (B x W)

1 2 3 4 5

How does the new client discover channels used by the BS?

BS and Clients must use same channelsFragmentation Try different center channel and widths

11

SIFT, by example

57

ADC SIFT

Time

Am

plitu

de

10 MHz5 MHz

SIFT

Pattern match in time domain

Does not decode packets

Data ACK

SIFS