Heather Zheng Department of Computer Science p p University of - - PowerPoint PPT Presentation
Heather Zheng Department of Computer Science p p University of California, Santa Barbara CS201 UCLA, May 22, 2008 1 Explosion of wireless networks and devices Static spectrum assignments are inefficient p g Under utilization +
Heather Zheng Department of Computer Science
1
p p University of California, Santa Barbara CS201 UCLA, May 22, 2008
Explosion of wireless networks and devices Static spectrum assignments are inefficient
p g
Under‐utilization + over‐allocation Artificial spectrum scarcity
Solution: Migrate from long‐term static spectrum Solution: Migrate from long‐term static spectrum
assignment to dynamic spectrum access
2
Scalability and speed Scalability and speed Support a large number of nodes
Adapt to time‐varying demands Efficiency + Fairness
Maximize spectrum utilization Avoid conflict
Reliability Provide QoS
3 Manhattan (Courtesy of Wigle.net)
Introduction Dynamic Spectrum Management Distributed spectrum coordination for fast
Interference‐aware admission control to provide
Conclusion and ongoing work
4
G l G l Goal Goal: Allocate spectrum to maximize system
utility
Assumption Assumption: 100% willingness to collaborate Node Collaboration Node Collaboration Action: Iterative Explicit Coordination Action: Iterative Explicit Coordination Action: Iterative Explicit Coordination Action: Iterative Explicit Coordination
l l
no local improvement can improve utility
Cao & Zheng, SECON 2005, Crowncom07, JSAC08, MONET08
5
Isolation
Limited neighbor coordination to
Isolation between groups
Limited neighbor coordination to reduce complexity
g g
Self-contained group coordination to prevent group conflicts
Restricted modifications Isolated bargaining group
6
Each local improvement will improve the global system utility
Fast Convergence Fast Convergence: The system converges after at most O(N2) local Node Collaboration Node Collaboration adjustments, N= network size Guaranteed Spectrum Allocation Guaranteed Spectrum Allocation: Each Guaranteed Spectrum Allocation Guaranteed Spectrum Allocation: Each node n’s allocated spectrum A(n) ≥ Poverty Line PL(n)
⎥ ⎦ ⎥ ⎢ ⎣ ⎢ + = 1 ) ( ) ( ) ( n D n L n PL
Total usable spectrum
⎦ ⎣ +1 ) (n D
Conflict degree
Cao & Zheng, SECON 2005
7
20% nces
ances ances
15% ge of instanc
5% 10% Percentage
centage centage
1 2 3 4 5 5% P
Perc Perc
1 2 3 4 5 Node utility / lower bound
A(n)/PL(n) A(n)/PL(n)
8
Each channel i has a weight of B (n) Each channel i has a weight of Bi(n) Each node’s spectrum allocation
i( ) i( )
Extended poverty line
A(n) > PL(n)
i i i i
9
Cao & Zheng, Crowncom07
Each infrastructure node n supports tn users Maximize end‐user fairness
Each infrastructure node’s spectrum has a
k n n
10
= ) (n N k
11
Use poverty line to initiate coordination Enable multiple parallel coordination events
Minimize adaptation delay
12
B i i d
Enabled Bargaining
Bargaining ends; Bargaining timer expires
Enabled Bargaining
Send request; Receive request;
Request
Disable timer expires Receiver ACK/disable message
ACK Disable Disabled
13
We can also regulate the coordination format to avoid disabling neighbors
# of Local coordination scales Adaptation delay flattens out linearly with the # of APs p y because of parallelism.
14
1Mbps Wireless Backhaul running CSMA/CA among APs
Quick adaptation to local Minimum disturbance to Quick adaptation to local dynamics neighbors
15
1.
Centralized, topology based
1.
Centralized, topology based
Graph Coloring
2.
Decentralized device coordination
negotiation‐based local coordination
based independent adjustment
Focus on AP based scenarios
16
Determine how many and which channels to use at each AP
Manhattan (Courtesy of Wigle.net)
Spectrum Channels
time
Unreliable spectrum access
How can we regulate nodes’
17
spectrum demand to maintain reliability?
17
Introduction Dynamic Spectrum Management
Distributed spectrum coordination for fast adaptation Interference‐aware statistical admission control to
Conclusion and Future Work
18
Admission based on statistical traffic information
… …
D namic spectr m allocation
time
19
19
Dynamic spectrum allocation based on short‐term demand
Cao & Zheng, INFOCOM08
Guarantee: Prob(total traffic exceeds C)<
γ −
e
3) , ( s f γ
C
1 2Router capacity = C
20
Stricter than the original interference constraints
Original interference constraints g Simplified linear interference constraints p
21
Our strategy: Uniform S
Based on the following analytical observations
Uniform S is optimal under uniform traffic statistics Under non‐uniform traffic statistics, the use of uniform S
has bounded degradation g(s)
Under non‐uniform traffic statistics, the optimal uniform S
is bounded
22
No Adm SPARTA O tage < 2% Outage = 60% Outage < 2% Peak Rate Outage =0% g
SPARTA almost doubles utilization from PRA SPARTA loses up to 40% utilization but greatly reduces
23
SPARTA loses up to 40% utilization but greatly reduces
Issues in Dynamic Spectrum Systems
y p y
Utilization, fairness
Our contributions : Collaborative
Theoretical models Deployment
Our contributions : Collaborative
spectrum sharing for large‐scale networks
Security
Incentives
Distributed coordination for fast
system convergence
Rule regulated self‐adjustment for
Spectrum allocation
Routing
Rule regulated self‐adjustment for
simple deployment
Providing reliability and efficiency via
p Device coordination
statistical admission control
Much more to be done
24
Hardware: software radios