Transcript Document

Smart-Radio-TechnologyEnable Opportunistic
Spectrum Access
Univeristy Of California Davis
PI: Xin Liu (CS)
NSF NeTS Workshop
2006@UCLA
Project Goals and Scope

What are the impacts and properties of the white space
and how can we quantify them?



Q: one experiment shows 62% of white space in spectrum under
3GHz at a certain location. Is exploiting this white space
equivalent to gaining 0.63*3GHz bandwidth?
A: It depends.
How should secondary users share the white space
dynamically and efficiently?


To develop a framework and performance metrics to evaluate
sharing mechanisms
To study new protocols and to identify the suitable solutions for
different application scenarios.
Characterizing Spectrum-Agile
Networks

A new metric, Equivalent Non-Opportunistic Bandwidth, to quantify
 Spatial diversity gain
 Statistical multiplexing gain


The effects of spectrum availability pattern, network topologies, and other
factors are being studied
Inherent benefits of heterogeneity between primary and secondary users
 TV stations and WLAN devices
 if we allow WLAN to operate in TV service contour when TV station is silent , statistical
multiplexing gain
 If not, we still have spatial diversity gain!


Investigating analytical models to capture the spatial and temporal
characteristics of white space and their impact on spectrum-agile networks
X. Liu and W. Wang, "On the Characteristics of Spectrum-Agile
Communication Networks", IEEE Symposium on New Frontiers in Dynamic
Spectrum Access Networks (DySPAN), Baltimore, MD, Nov. 8-11, 2005.

X. Liu, “Characterizing Spectrum-Agile Networks”, under submission.
Dynamic Spectrum Sharing

Two unique characteristics: location-dependency and time-variance
 Location-dependency: list-coloring
 Time-variance: allocation algorithms have to work under scenarios with limited
information exchange from neighbors due to time-variance


Channel allocation formulated as list-coloring problem
Algorithms proposed:
 Optimal Solutions: Centralized brute force search, served as Benchmark
 Distributed Greedy: Assign channel one by one, maximize allocation for each
channel
 Distributed Fair: To achieve max-min fairness by taking the link degree and
channel degree into account
 Distributed Randomized: Balanced between utilization and fairness, smallest
complexity

W. Wang, X. Liu, and Hong Xiao, "Exploring Opportunistic Spectrum
Availability in Wireless Communication Networks", IEEE VTC Fall 2005,
Dallas, TX, September 25-28, 2005
Traffic Information Uncertainty &
Robust Resource Allocation
 Accurate traffic information is hardly available
 Traffic varies over time and difficult to measure
 Dissemination of traffic information may incur delay and
overhead
 On the other hand, coarse estimation is possible
 Source-destination pairs & range of the traffic demands
 Developed a routing and scheduling scheme that works well for a
range of traffic conditions
 Achieve the best worst-case performance
 Extended to topology control – topology control must take into
account traffic demand and be performed infrequently
 To study uncertainty in Spectrum-Agile networks.
 W. Wang and X. Liu, “Robust routing-scheduling in multihop
wireless networks”, under submission
Current and Future
Research Emphasis
 To capture the spatial and temporal characteristics
of white space and to quantify their impact on
spectrum-agile networks
 To develop centralized and decentralized
algorithms with different degrees of information
exchange among primary and secondary users
 To consider fairness and power/interference
constraints
 To study the impact of dynamic spectrum
utilization on QoS and to propose appropriate
admission control schemes
Links to other projects











Xin Liu (University of California, Davis) CAREER: Smart-Radio-Technology-Enabled
Opportunistic Spectrum Utilization
Dirk Grunwald, Doug Sicker, John Black (University of Colorado), NeTS-ProWIN: Topology And
Routing With Steerable Antennas
Uf Turelli, Kevin Ryan (Stevens Institute of Tech), Milind M. Buddhikot, Scott Miller (Lucent Bell
Lab), Dynamic Intelligent Management of Spectrum for Ubiquitous Mobile Network
(DIMSUMnet)
Kang G. Shin, University of Michigan, Efficient Wireless Spectrum Utilization with Adaptive
Sensing and Spectral Agility
Qing Zhao, UC Davis, An Integrated Approach to Opportunistic Spectrum Access
Randall Berry, Michael Honig and Rakesh Vohra, Northwestern University, Smart Markets for
Smart Radios
Mario Gerla, Stefano Soatto, Michael Fitz, Giovanni Pau, UCLA, Emergency Ad Hoc Networking
Using Programmable Radios and Intelligent Swarms
Saswati Sarkar, University of Pennsylvania, Dynamic Spectrum MAC with Multiparty Support in
Adhoc Networks
Marwan Krunz, Shuguang Cui, University of Arizona Resource Management and Distributed
Protocols for Heterogeneous Cognitive-Radio Networks
Dennis Roberson, Cindy Hood, Joe LoCicero, Don Ucci (Illionis Institute of Technology), Uf Tureli
(Stevens Institute of Technology) Wireless Interference and Characterization on Network
Performance
Narayan Mandayam, Christopher Rose, Predrag Spasojevic, Roy Yates, WINLAB Rutgers
University, Cognitive Radios for Open Access to Spectrum
Links to other projects
 Platform/Testbed projects
 Dirk Grunwald (U. Colorado), John Chapin (Vanu, Inc), Joe Carey
(Fidelity Comtech) A Programmable Wireless Platform For Spectral,
Temporal and Spatial Spectrum Management
 Jeffrey H. Reed, William H. Tranter, and R. Michael Buehrer, Virginia
Tech, An Open Systems Approach for Rapid Prototyping Waveforms
for Software Defined Radio
 D. Raychaudhuri (WINLAB, Rutgers University) ORBIT: Open Access
Research Testbed for Next-Generation Wireless Networks
 B. Ackland, I. Seskar & D. Raychaudhuri, (WINLAB, Rutgers
University), T. Sizer (Lucent Technologies), J. Laskar(GA Tech) High
Performance Cognitive Radio Platform with Integrated Physical and
Network Layer Capabilities
 Babak Daneshrad, University of California, Los Angeles,
Programmable/Versatile Radio Platforms for the Networking
Research Community
 Prasant Mohapatra, University of California, Davis, Quail Ridge
Wireless Mesh Networks: A Wide Area Test-bed
Additional Information
NSF NeTS Workshop
ENOB: Effective NonOpportunistic Bandwidth
 Equivalent non-opportunistic bandwidth
required to achieve the same throughput
vector as in the case of opportunistic
spectrum availability.
 Non-opportunistic band: always available to
the users as in the traditional command-andcontrol manner.
 Depends on channel availability
correlations of secondary users
 A metric to quantify the impact of diversity
A Naïve Example
 Two secondary nodes opportunistically
access a primary channel
 Observes independent channel availability
with prob. p.
 They interfere with each other
 Assume one unit of throughput per unit of
bw.
A Naïve Example Cont’d
 Total throughput:
 W(p*p*1+2p(1-p)*1+(1-p)(1-p)*0)=Wp(2-p)
 ENOB = Wp(2-p)
 62% white space under 3G
 W= 3GHz, p= 0.62
 ENOB = 2.76 GHz
 Instead of Wp=3*0.62=1.86GHz
Intuitions
 Spectrum is not being “created” by secondary users.
 Exploit spectrum holes created by primary users.
 Different secondary users have diff. availability
 Spectrum opportunity and its properties are determined by
primary users
 ENOB: a metric to quantify the degree of spatial reuse and
statistical multiplexing between primary and secondary
users.
 Analogy: effective bandwidth used to capture statistic
multiplexing gain.
 Depends on correlations of channel availability among users
 Depends on sharing criterion
ENOB of a Chain Topology
…
1
2
3
N
 Consider the dependency of channel availability
among users
 Evenly spaced nodes
 p0: prob. a node observes the channel avail.
 pc: prob. node i observes given a neighbor does
A Chain Topology
Different Schemes
 Node 1 interferes with all
others
 Nodes observe channel
availability independently
 Objectives:
 maxsum
 maxmin
 maxT1
5
2
1
4
3
ENOB cont’d
ENOB Summary
 A metric to quantify the effect of
opportunistic channel availability
 Its value depends on
 Topology, traffic pattern of primary, etc.
 Channel availability dependency
 Channel allocation algorithm/objective
 Heterogeneous network
 Implications on resource management
Why traffic-aware topology control?
Topology at the maximum power
Topology with minimum power and interference
 Two traffic patterns
 Local: every node sends to its right
neighbor
 Single-sink: every nodes sends to the nth
node
An Example (cont’d)
n-3
n-3
Topology at the maximum power
n-1
n-2
n-2
n
n-1
Topology with minimum power and interference
Local
Single-sink
Clique
1/(n-1)
1/(n-1)
Chain
1/3
< 1/(3n-6)
Observation: Minimizing interference/power is not necessarily optimal.
Motivations
 Topology control must take into account traffic.
 Accurate traffic information is hardly available
 Traffic varies over time
 Difficult to measure
 Dissemination of traffic information may incur excessive
overhead
 Topology control should be infrequent to avoid frequent
service disruptions
 On the other hand, coarse estimation on the traffic
pattern/demand is possible
 Source-destination pairs (e.g., single-sink)
 Range of the traffic demands (e.g., 200K – 1Mbps)
Traffic-Oblivious Routing and
Scheduling
 Objective: to design a routing and
scheduling that works well for a range
of traffic conditions
 To achieve the optimal worst-case
performance in the range of traffic
conditions being considered
 The problem can be solved using a
single LP with an infinite number of
constraints.
Competitive Analysis
Congestion
Minimum congestion level
Competitive ratio
Oblivious ratio
Formulation
Objective
Problem formulation
Non-linear
Formulation
Master LP
All traffic patterns
Infinite #
Formulation
Slave LP (to check the constraint of the
master LP)
Formulation
 The above formulation has finite number
of variables, but infinite number of
constraints.
 To further reduce the complexity
 Convert the slave LP to its dual form
 Combine the master and the dual of the slave
to form a single LP
What have we learned?
 Well-designed multipath is desirable.
 Spatial reuse
 Load balancing
 Robust performance
 Low oblivious ratio
 Close to ideal performance with perfect
information
 Robust even under faulty information