Post on 20-Feb-2022
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