Cognitive Radio Research@WINLAB Roy Yates WINLAB Rutgers - PowerPoint PPT Presentation

Cognitive Radio Research@WINLAB Roy Yates WINLAB Rutgers University December 10, 2008 ryates@winlab.rutgers.edu www.winlab.rutgers.edu 1 WINLAB

Cognitive Radio Research A Multidimensional Activity � Theory and Algorithms � Spectrum Policy � Fundamental Limits � Economics � Information & Coding Theory � Regulation � Cooperative Communications � Legal � Game Theory & Microeconomics � Business � Hardware/ Software Platforms & Prototyping � Programmable agile radios � GNU platforms � Cognitive Radio Network Testbed 2 WINLAB

FCC Spectrum Management False Scarcity Maximum Amplitudes Heavy Use Heavy Use Less than 6% Occupancy Less than 6% Occupancy Amplidue (dBm) Sparse Use Medium Use Frequency (MHz) � Most bands are actually � Most bands are actually � All bands are allocated, � All bands are allocated, largely unused. largely unused. many for multiple many for multiple purposes. purposes. 3 WINLAB

Spectrum Policy Debate � Property Rights � Triumph of Economics � [ Coase, Hazlett, Faulhaber+ Farber] � Owners can buy/ sell/ allocate spectrum � A spectrum market will yield an efficient solution � Open Access and Com m ons � Triumph of Technology � [ Noam, Benkler, Shepard, Reed] � Agile radios to dynamically share common spectrum � Open Access: Strict technology needs- sensing, interference � Commons: Distributed protocol followed in system 4 WINLAB

The Spectrum Debate & Cognitive Radio � What everyone agrees on now: ☺ � Spectrum use is inefficient � FCC licensing has yielded false scarcity � Possible middle ground? � Dynamic spectrum access � Short-term property rights � Spectrum use driven by both technology and market forces � Cognitive Radios with ability to incorporate market forces? � Microeconomics based approaches to spectrum sharing � “Dynamic Spectrum Access Models: Towards an Engineering Perspective in the Spectrum Debate” by Ileri & Mandayam, IEEE Communications Magazine, Jan 2008. 6 WINLAB

Cognitive Approaches: Outlook � Lots of netw ork ( and channel) state inform ation needed to enable efficient � Discovery End-to-end routed path � Self-organization From A to F Bootstrapped PHY & � Cooperation Techniques control link C B B D D PHY A E PHY C A PHY B A Multi-mode radio PHY Ad-Hoc Discovery F & Routing Capability Control (e.g. CSCC) Functionality can be quite challenging! 7 WINLAB

Cognitive Radios need information � Reactive schemes (without explicit coordination protocols) have limitations. � Interference is a receiver property. � Alternative: Infrastructure-based coordination � Examples of coordination mechanisms: � Information aids � “Spectrum Coordination Channel” to enable spectrum sharing � Network architectures � “Spectrum Servers” to advise/ mediate sharing 8 WINLAB

Common Spectrum Coordination Channel (CSCC) [Jing, Raychaudhuri] CSCC can coordinate radios with incompatible PHY � � CSCC � Employs an out-of-band etiquette channel & protocol � Periodic TX of radio parameters on CSCC � TX at higher power to reach hidden nodes � Local contention resolved via protocol-independent etiquette policies � Also supports ad-hoc multi-hop routing associations CSCC Ad- RX hoc Ad- range net B hoc for X net A X Maste r CSCC Y Node Ad-hoc RX range Piconet for Y 9 Jing, Raychaudhuri WINLAB

802.11 & 16 Co-Existence: Reactive vs. CSCC-based Power Control [] 802.11b Hotspot D SS-AP 100m Single 802.11 Hot Spot Case AP SS 1km BS 802.16a Cell Average Link Throughput (Mbps) 1.2 0.8 Average Link Throughput (Mbps) 0.7 1.0 0.6 0.8 0.5 802.16a DL 802.16a DL with CSCC 0.6 0.4 802.11 link 802.11 link with CSCC 0.3 0.4 Average No Coordination No Coordination Average with CSCC 0.2 CSCC frequency adaptation 0.2 0.1 0.0 0.0 802.16 DL 802.11 link 0 200 400 600 800 1000 802.16 DL and 802.11 link Distance between 802.16 SS and 802.11 hotspot (meters) CSCC frequency adaptation when Throughputs vs. D SS-AP by using CSCC D SS-AP = 200m and traffic load 2Mbps pow er adaptation and traffic load 2Mbps 10 WINLAB

What can a Spectrum Policy Server do? (Anything & Everything) Spectrum server Implicit CSCC Secondary AP sensors Home networking (( )) Wireless world is not flat! Wireless LANs TV broadcasting multihopping Multiaccess/variable rate orthogonal & non-orthogonal Interference channel / transmission schemes wideband transmissions multiplexing schemes 11 WINLAB

Cognitive Radio: Spectrum Policy Server Internet-based Spectrum Policy Server (a “Google for spectrum”) � IP-based CSCC � Coarse location information � SPS provides centralized local spectrum coordination � Spectrum Internet Policy Server Internet w w w .spectrum.net AP1: type, loc, freq, pwr Etiquette AP2: type, loc, freq, pwr Protocol BT MN: type, loc, freq, pwr AP1 Access Point (AP2) WLAN operator A WLAN operator B Master Node Wide-area Cellular data service Ad-hoc Bluetooth Piconet 12 WINLAB

Cognitive Radio Networks Cross Layer Scheduling via a spectrum server Scheduling Physical links - � achievable rates depend on PHY (( )) layer TX & RX Routing – decisions based on � network/ application layer metrics Constraints � Link Rate > ∑ flows on link Perform ance bounds via � Flow 1 centralized scheduling Flow 2 Low -com plexity scheduling � schem es Randomized Distributed scheduling � (RDS) Column generation methods for � choosing good scheduling modes 13 WINLAB [ Raman, Mandayam, Yates]

Two Tier Dynamic Spectrum Access Spectrum Policy Server/ Regulator/ Clearinghouse Level I Service Providers SPs obtain ( SP) compete spectrum from SPS Level II End Users: Adapt End users obtain rate, power, spectrum spectrum from use for max net utility SPs Examples: 802.22 Service Providers OFDM tone allocation to end users DimsumNet 14 WINLAB

Two Tier DSA Properties � Develop engineering models for shaping spectrum policy � Features: � Dynam ic Spectrum Access: Short term allocation of spectrum resources � Tem porary Exclusive Usage: Parties do not suffer interference � Market Based Allocation: Supply and demand determines who gets how much bandwidth 15 WINLAB

Two-Tier Spectrum Access Mechanisms � D-Pass (Dynamic Property- Rights Spectrum Access) � Allocation based charges � SPs pay for spectrum allocation � SPs then compete for users via simultaneous auctions � D-CPass (Dynamic-Commons Property-Rights Spectrum Access) � Usage based charges � Clearinghouse mediates bidding among users � SPs only pay for spectrum actually used � D-CPass yields better spectrum utilization 16 Ileri, Mandayam WINLAB

Two Tier DSA User and Service Provider Heterogeneity Spectrum Level I Clearinghouse SPs buy/ lease spectrum from Service Providers clearinghouse ( SP) compete to maximize profits Level II Level II End users lease End users lease End Users: Adapt rate, spectrum from spectrum from power, spectrum use SPs SPs for max net utility 19 Acharya, Yates WINLAB

Cost Model for the SP � Spectrum Cost � C(X) = CX, X: sum of spectrum from all users � Constant C set by clearinghouse � Depends on Geographical region, urban/ rural � Power Cost � Transmit power = ν X � F( ν ,X) = T ν X � Constant T may depend on � Presence of other providers in band ‘X’ 20 WINLAB

User j: spectrum & rates � h j = downlink gain SP TX: fixed power spectral density ν � � Spectral efficiency K j = log(1+ ν h j / N 0 ) x j = allocated spectrum � Rate R j = K j x j , � Higher Spectral Efficiency K j � user j = Better Radio Technology 21 WINLAB

User j: utility functions � Logarithmic � Exponential U(R j ) = log(1+ R j ) ) = Γ j / Γ j U(R j [ 1-exp(-R j )] � Elastic data application � Application with target rate Γ j (large file download) U(R j ) U(R j ) Γ j R j R j 22 WINLAB

Objective of the SP and Users � Clearinghouse set spectrum price � SP maximizes its net revenue � Expense: Spectrum purchase and transmit power � Income: Charges the users � Users maximize their utility minus cost � Expense: Charge paid to the SP � Gain: Increase in utility due to spectrum 23 WINLAB

Elasticity of Demand Ratio of % change in demand to % change in price � Logarithmic Utilities: Elastic demand ( ε > 1) always � When price is increased, % fall in demand is higher � SP can’t arbitrarily overprice spectrum � Exponential Utilities: Inelastic demand ( ε < 1) for low μ � Low μ : Enough spectrum for users to be rate saturated � Price changes in this regime, % change in demand is less � 24 WINLAB

SP Profit vs Spectrum Cost C Exponential Utilities 10 ν = 50 dBm/MHz Target Rate = 1 Mbps ν = 30 dBm/MHz ν = 20 dBm/MHz Power Cost, T= 10 ν = 10 dBm/MHz 8 Total cost C e = C+ T ν Profit of SP � High C 6 C e = C+ T ν ~ C User Utility ↑ as ν ↑ SP incentive: High ν 4 � Low C 2 C e = C+ T ν ~ T ν SP incentive: Low ν 0 0 0.5 1 1.5 2 2.5 3 spectrum cost (C) $/MHz 25 WINLAB