Information Theory for Mobile Ad-Hoc Networks (ITMANET): The FLoWS Project

FLoWS Overview and Progress

Andrea Goldsmith

DARPA’s ITMANET Challenge

• Develop and exploit a more powerful information theory for mobile wireless networks.

• Anticipated byproducts include new separation theorems to inform wireless network "layering" as well as new protocol ideas.

Hypothesis: A better understanding of MANET capacity limits will lead to better network design and deployment.

MANET Capacity: What we don’t know

• Capacity of large dynamic networks

• Capacity of basic network building blocks

Xi

Yi-1

p(yi,si|xi,si-1)

Si-1 Si D

YiTx Rx

• Capacity of time-varying links (with/without feedback)

Shannon and Wireless Networks

• Shannon’s capacity definition based on infinite delay and asymptotically small error was brilliant

• Much progress in finding the Shannon capacity limits of wireless single and multiuser channels

• Little known about these limits for mobile wireless networks, even for simple (canonical) models– Few fundamental design insights have emerged.

– Queuing and capacity theory incompatible if capacity implies infinite delay and no error.

– How should capacity be defined for wireless networks?

Limitations in theory of MANETs today

– Shannon capacity pessimistic for wireless channels and intractable for large networks

WirelessInformation

Theory

Optimization Theory

B. Hajek and A. Ephremides, “Information theory and communicationsnetworks: An unconsummated union,” IEEE Trans. Inf. Theory, Oct. 1998.

WirelessNetworkTheory

– Large body of wireless (and wired) network theory that is ad-hoc, lacks a basis in fundamentals, and lacks an objective success criteria.

– Little cross-disciplinary work spanning these fields

– Optimization techniques applied to given network models, which rarely take into account fundamental network capacity or dynamics

Our Approach: Consummating Unions

• When capacity is not the only metric, a new theory is needed to deal with nonasymptopia (i.e. delay, random traffic) and application requirements– Shannon theory generally breaks down when delay, error, or user/traffic dynamics must be considered

• Fundamental limits are needed outside asymptotic regimes

• Optimization provides the missing link to address these issues

WirelessInformation

Theory

WirelessNetworkTheory

OptimizationTheory

Menage a Trois

FLoWS Program Objectives

• Develop tractable and insightful metrics and models for MANET information theory.

• Define fundamental performance limits for MANETs in terms of desired objective metrics.

• Obtain upper and lower performance bounds for these metrics for a given set of MANET models.

• Define the negotiation between the application and network for resource allocation and performance optimization of our given metrics

• Bound the cost of using our set of metrics as the interface between the network and applications.

Capacity Delay

Power

Upper Bound

Lower Bound

Capacity and Fundamental Limits

Application Metrics

Capacity

Delay

Power

Utility=U(C,D,E)

Application andNetwork Optimization

(C*,D*,E*)

Constraints

Degrees ofFreedom

Models andDynamics

Application Metrics and

Network Performance

LayerlessDynamic Networks

New Paradigmsfor Upper

Bounds

Models

New MANET Theory

MANET Metrics

Metrics

Fundamental Limitsof Wireless Systems

Thrust Objectives and Rationale• Models and Metrics (Leads: Effros,Goldsmith,Medard):

– Objective: Develop a set of metrics for dynamic networks that capture requirements of current and future applications

– Rationale: Models for MANETs are needed that are tractable yet lead to general design and performance insights

• New Paradigms for Upper Bounds (Leads: Effros,Koetter,Medard)– Objective: Obtain bounds on a diversity of objectively-defined metrics for complex

interconnected systems.– Rationale: A comprehensive theory for upper bounding the performance limits of

MANETs will help guide design

• Layerless Dynamic Networks (Lead: Zheng)– Objective: Design of networking strategies as a single dynamic probabilistic

mapping, without pre-assigned layered structure – Rationale Remove layering and statics from MANET theory.

• End-to-End Metrics and Performance (Leads: Ozdaglar,Shah)– Objective: Provide an interface between application metrics and network

performance– Rationale: A theory of generalized rate distortion, separation, and network

optimization will improve application performance

StructuredCoding

Thrust Synergies and New Intellectual Tools

New BoundingTechniques

Optimization

Code Construction

Combinatorial Tools

DynamicNetwork IT

Thrust 1

Thrust 2

Game Theory

Thrust 3

Optimization

StochasticNetworkAnalysis

CSI, Feedback,and Robustness

Thrust 0 Recent Achievements

Metrics

Effros, Goldsmith: Expectation and Outage in Capacity and Distortion

Models

Goldsmith: Diversity/multiplexing/delay tradeoffs

Effros: networks with side information

Shah: multicast capacity

Moulin: Mobility

Medard: delay/energy minimization

Zheng: UEP

Medard, Zheng: Diversity-distortion tradeoff

Coleman, Effros, Goldsmith, Medard, Zheng: Channels and Networks with Feedback

Goldsmith: Cognitive Nodes

New bounding techniques

Thrust 1 Recent Achievements

Code constructionNetwork information theory

Networkingand OptimizationCombinatorial Tools

Medard: coded time-division duplex for delay or energy minimization

Goldsmith: multiple sender interference channel

Moulin: converse for Gelfand-Pinsker MAC

Effros: networks with side information

Effros: effect of feedback in networks

Shah: multicast capacity of large wireless networks

Koetter, Medard: joint coding and scheduling in wireless networks

Zheng: unequal error protection converse

Moulin: mobility for interference mitigation

Effros: game-theoretic approaches to network coding

Koetter, Medard: Distributed scheduling for network coded multicast

Dynamic Network Information Theory

CSI, feedback, and robustness Structured coding

Thrust 2 Recent Achievements

Thrust 3 Recent Achievements

Shah, Medard: Queueing analysis for coded networks with feedback

Johari: Fluid limits for gossip

Meyn: Relaxation techniques for network optimization

Goldsmith, Johari: Oblivious equilibrium for stochastic games with concave utility

Boyd, Goldsmith: Wireless network utility maximization as stochastic optimal control

Optimization

Stochastic Network Analysis Game Theory

Ozdaglar: Distributed second order methods for network optimization

Ozdaglar: Noncooperative scheduling with correlated channel states

Effros: Noncooperative network coding

Moulin: Interference mitigating mobility

Progress on what we don’t know

Xi

Yi-1

p(yi,zi,si|xi,si-1)

Si-1 Si D

YiTx Rx

• Capacity of time-varying links (with/without feedback)

• Capacity of finite-state Markov channels with feedback• Converses under unequal error protection• Multiplexing-diversity-delay-distortion tradeoffs in MIMO • Generalized capacity and separation

• Capacity of basic network building blocks

• Capacity region/bounds for Z channel and interference channels• Capacity bounds for cognitive interference/MIMO channels

• Upper bounds and converses for interference channels with a relay (via interference and message forwarding)

• Capacity of dynamic networks

• Network equivalence• Scaling laws for arbitrary node placement and demand• Multicast capacity• Effect of feedback and side information• Dynamic/multiperiod network utility maximization• Generalized Max-Weight policies• Game-theoretic approaches• Mobility for interference mitigation• Delay or energy minimization• Distributed optimization

FLoWS progress since last September

• New breakthroughs in upper bounds, feedback and CSI, cognitive techniques, interference forwarding, multicast traffic, and dynamic/distributed network optimization,

• New synergies within and between our thrust areas

• New and ongoing collaborations among PIs

• Overview paper for Scientific American– Co-authors: Effros, Goldsmith, Medard– Paper accepted for publication

• Magazine paper with overview of FLoWS– Paper near completion – will be submitted within 3 weeks

• All Phase 2 progress criteria have been met

• Website updated with Sept. PI meeting slides, recent publications, and recent results.

Focus Talks and Posters

• Thrust 1:– Moulin: Towards strong converses for MANETs

• Thrust 2:– Coleman: A dynamic approach to feedback coding

strategies in MANETs

• Thrusts 2 and 3:– Effros: The cost of selfishness in network coding capacity

• Thrust 3:– Boyd and Goldsmith: Optimizing wireless network

performance in stochastic environments

• Posters on all recent results

Progress Criteria: Phase 2– Evolve results in all thrust areas to examine more complex models,

robustness/security, more challenging dynamics, and larger networks. • Network equivalence• Multihop networks with interference forwarding• Interference mitigation via mobility• Unequal error protection• Control principles for feedback designs• Capacity scaling laws for arbitrary node placement and arbitrary demand• Distributed second order methods for network optimization • Generalized Max-Weight policies with optimal distributed implementations• Local dynamics for topology formation• Oblivous equilibrium

– Demonstrate synergies between thrust areas: compare and tighten upper bounds and achievability results for specific models and metrics; apply generalized theory of distortion and utility based on performance regions developed in Thrusts 1-2.

• Capacity of finite-state broadcast channels w/wout feedback and Z channel (1-2)• Bounds on capacity of cognitive radio and interference channel w/relay (1-2)• New approach to optimize coding and scheduling (1-3 )• New game-theoretic approach to network coding (1-3 )• Multiperiod NUM applicable to any underlying network capacity region (1-2-3)

Progress Criteria: Phase 2– Demonstrate that key synergies between information theory, network theory,

and optimization/control lead to at least an order of magnitude performance gain for key metrics.

• Coding-based TDD has order of magnitude throughput increase versus uncoded• Unbounded gain in network capacity with feedback • Order of magnitude throughput gain in hyperarc scheduling• Orders of magnitude BER reduction in wireless NUM• Order of magnitude utility improvement for generalized Maxweight

– Pose clearly defined community challenges related to evolving our theory that inspires other researchers to collectively make breakthrough progress.

• Community challenges posed in plenary talks and tutorials, as well as invited papers and vision papers. Proposed book will define FLoWS ITMANET theory.

– Publish 2 vision papers, one for the community (e.g. in the IEEE Wireless Communications Magazine) and one for the broader technical community (e.g. in Nature or Science) illuminating our ideas, results, and their potential impact

• Scientific American paper accepted. Iteration with publisher ongoing. • Magazine paper near completion, will be submitted within 3 weeks. • Book proposal has been discussed with Cambridge University Press.

Project Impact To Date• Plenary Talks

– Boyd: Dysco’07, S. Stevun Lecture’08, CNLS’08, ETH’08– Effros: ISIT’07 – Goldsmith: ACC’07, Gomachtech’08, ISWPC’08, Infocom’08, RAWC’09, WCNC’09– Koetter: ITW’07, WiOPT’08– Medard: Gretsi’07, CISS’07, Gilbreth Lecture (NAE)’07, IT Winter School’08, UIUC Student

Conference’08, Wireless Network Coding’08, – Meyn: Erlang Centennial’09– Ozdaglar: ACC 2009, NecSys'09 , ASMD’08– Johari: World Congress of the Game Theory Society’08– Shah: Plenary talk at the Interperf/Valuetools 2007

• Recent Tutorials– Boyd: MOCCS’08, WOSP’08– Ozdaglar: ISIT’08, CDC’08– Medard/Koetter: PIMRC’08– Meyn: Mathematics of OR’08– Shah: ACM Sigmetrics/Performance ’08, CDC’09,

• Conference Session/Program Chairs– ISIT’07, CTW’08, WiOpt’08, CTW’09, ITW’09, ITW’10, ISMP’09, INFORMS’09

• Invited journal papers– “Breaking spectrum gridlock through cognitive radios: an information-theoretic approach”, IEEE

Proc’09 (w/ Jafar, Maric, and Srinivasa)

Publications to date

• 16 accepted journal papers, 13 more submitted

• 89 conference papers (published or to appear)

• SciAM paper to appear– Magazine paper to be submitted shortly– Book on FLoWS vision and results under development

• Publications website:– http://www.stanford.edu/~adlakha/ITMANET/flows_publications.htm

Summary

• Significant progress on all thrust areas

• Significant progress on synergies between thrust areas

• Ongoing and fruitful collaborations between PIs

• Phase 2 goals met

• Significant impact of FLoWS research on the broader research community (IT, communications, networking, and control/optimization)