Fundamentals of Cognitive Radio Technology of Cognitive Radio Technology ... (Dr. Mark McHenry) ... Conceptual. Clustering. Set Cover Using. Generalization &

Fundamentals of Cognitive Radio Technology

Seminari Fondazione Ugo BordoniNew Frontiers in the Management of Radio Spectrum

June 2009

Dr. Joseph Mitola IIIDistinguished Professor, School of Engineering and Science,

Distinguished Professor, School of Systems and Enterprises, andVice President for the Research Enterprise

Stevens Institute of Technology

Domain-Independent Systems Engineering

http://www.stevens.edu/sit/2

Overview• Foundations

– Saturated Commercial Peak Loading– Low Average Spectrum Occupancy

• Dynamic Spectrum Research and Rule Making– WiFi – Fiber Proliferation– Regulatory Perspectives: European Workshops– FCC R&O

• Innovation in Spectrum Management– Advanced Services: 3D, Multisensory– Key Challenge: Rule Complexity and Cost

3

GSM, UMTS Saturated

Spectrum Fully Allocated, Saturated at Off-Peak Hours

4

Low Average Utilization

2 GHz

500 MHz

[© UC Berkeley, 2004, Used with Permission] 5

Policy Lanes in the RoadPolicies

1. Stay between lines

2. Do not hit other cars

3. Don’t run over things

Cars

If Cars and Responder Vehicles Can Share the Road, Why Not the Spectrum?

First Responders

Stay in lanes

Behaviors

Can block lanes

6

A Spectrum Pool Etiquette

© 1997-2007 Dr. Joseph Mitola III

7

8

Z m2

Cell Area

Frequency Reuse Creates Spectrum

System = N Cells

Physics of Spectrum Pooling

Public SafetyCell Area

Cost Includes Backhaul

Pooled Spectrum EtiquetteAdvertise (NB-CDMA)

Backoff Politely to Legacy Users

© 1997-2007 Dr. Joseph Mitola III 9

Ideal Cognitive Radio Architecture

User

Environment

Antenna RF Modem Baseband User Interface

Hardware

INFOSEC

Protocol Stack, ControlModem, Equalizer, etc.

Software RadioSoftware Modules …..

KnownSelf

RXML = Radio eXtensible Markup Language

Cognition CycleCognition

EnvironmentSpace-TimePropagationNetwork

Meta-SoftwareAntenna RF Modem

INFOSECBasebandUser Interface

Equalizer Model

Variable Bindings

...

User(s)IdentityNeedsBehaviors ...

RXML FramesOntologiesDescriptorsEffectors

Adapted from Cognitive Radio Architecture (Wiley 2006) with permission

© 2006 Dr. Joseph Mitola III Used with Permission 10

Dynamic Spectrum Research and Rule Making

FCC Report and OrderCognitive RadioTV Whitespace

Spectrum Management

SDR Forum Workshop: Regulators Address “Social Contracts”FCC: Rules Must Balance Economics with Public Interest

Regulatory Perspectives: Japan NICT/ARIB, UK Ofcom, Germany

12

Power Density vs. Occupancy

© 2005 Shared Spectrum Company (Dr. Mark McHenry) Used with Permission – Overlay © 2007 Dr. Joseph Mitola III

Processing Gain

Antenna Gain

Duty Cycle

13

Dynamic Spectrum

Primary (Licensed) Spectrum

Cellular 1G 2G CDMA 3GPP

Underlay: Ultra Wide Band (UWB)

Noise Temperature

Cordless Telephones: Analog, DECT …WLAN: 802.11 a/b/b WiFi …WPAN: 802.15, BlueTooth, HomeRF

Unlicensed Spectrum (Part 15)

Space-time UseageConsiderations

© 2006 Dr. Joseph Mitola III Used with Permission

Exciting Research and Development

Overlay: OFDM Water Filling

Measurements © 2006 Shared Spectrum Company, Used with Permission

14

Dynamic SpectrumOFDM “Water”

Multiple Input Multiple Output

(MIMO)

Beamforming

How Fast Will the

Water Rise?

© 1997-2007 Dr. Joseph Mitola III 15

Complexity of Space-Time-RF

© 2006 Dr. Joseph Mitola III Used with Permission

© 2000 RF-CAD

Higher Fidelity Sharing by Primary and Secondary Users

16

17

Horizontal and Vertical Artifacts Strongly Influence Path Loss

Signal Subspace Dynamics

© 1997-2007 Dr. Joseph Mitola III © IEEE Academic Use

Complexity of Multipath

Multiple Input Multiple Output (MIMO)

y(t)

x(t-τ2)

1st Bit

x(t)

2nd Bit

x(t-τ1)

3rd Bit

© 1997-2007 Dr. Joseph Mitola III

© 2007 Stephen Fast, Remcom

18

Multiple Element Antenna ArraysPatented by ATT as BLASTDeveloped by Europe as MIMOMultipath Enemy Friend3x to 10x Capacity of ShannonBring MIPS

© MIT Lincoln Laboratory

FCC R&O Provisions

• Power Limits– 100 mW– 40 mW Adjacent Channel

• Web Database Access– Database indicates prohibitions– Whitespace devices must access database– Daily access for use– No access (to database) ⇒ No access (to spectrum)

19

IEEE P1900.5 Policy Language

[ XG Policy Language R. Krishnan et al, BBN]

20

Innovation in Spectrum Management

Policies, Policy Languages, and Behavior Modeling for Autonomous

Conformance to Policy Intent

Cognitive Radio Evolution

• Cognitive Radio Foundation Era (1990-2005)– New radio spectrum with new RF technology– Grow markets via services (GSM) and products (WiFi)– Cognitive radio introduced as far term idea (1998)– Cognitive radio R&D: DySPAN, CROWNcom, XG…

• Cognitive Radio Evolution Era (2005-2020)– Dynamic spectrum approved for low power devices– Behavioral policy languages for spectrum management– Cost of conforming to policies/exceptions is growing– Sentient spaces paradigm establishes expectations[J. Mitola III, “Evolution of Cognitive Radio Architecture” Proceedings of the IEEE, 2009]

22

Use Case Evolution

Use Case Parameters Foundation era (1990-2005) Evolution era (2005-2020) Core wireless use cases Towards ubiquitous access Towards integrated services Profit margins High (handsets-infrastructure)

then handset profits declining Low (handsets-infrastructure) to high for differentiated services

Value proposition QoS (Connectivity, data rate) QoI (User is the 8th OSI layer) PSTN integration SS7[1], SDH[2] IP-SIP [3], Mobile IP, or IPv6[4] Reconfigurable HW Not worth the cost vs chipset Transitioning to mainstream? Location awareness Niche applications Ubiquitous Multimedia Infeasible to feasible Strong differentiator Spectrum awareness Within allocated band Across multiple bands Spectrum Auctions Large blocks for long term Small space-time holes short term Public safety Distinct markets Integration with agility Data rate framework Stationary, walking, vehicle Hot spot, traveling, emergency Sentient Spaces Video surveillance markets Elder care and home robotics

23

Interactive TV

HDTV

Computing: Moore’s Law

Cognitive Radio Domains

Wireless 1G, 2G GSM-PHS, 3G CDMA, B3G OFDM

Wireless Web

WiFi, WiMAX, OFDM

Land Mobile, Emergency, TETRA …

B3G

MIMO

> 100 Mbps

???

Security, Privacy

Broadcast TV

Cognitive Radio

Dynamic Spectrum

Blue tooth

time

complexity

24

Transportation Domain View

Wireless 1G, 2G GSM-PHS, 3G CDMA, B3G OFDM

Wireless Web WiFi

Land Mobile TETRA …

B3G

MIMO

> 100 Mbps

???Cognitive Radio Networks and Users

Blue tooth

Highways

EZ-Pass Tolls

Vehicle Enablers

- As Base Station

Interactive TVHDTV

Security, Privacy

Emergencies

Law Enforcement

Vehicular Enablers

- Integration

- As Base Station

Ports

Container ID

Shipboard AIS

- Robotic Ports

OnStar

25

Security Often Comes Last• Radio Security Research @ Stevens

– Compromising emanations, weak codes and ciphers• Radio biometrics (turning TX nuances into trustable IFF)

– Inherently Trustable Comput-ications Technology– Post-quantum Cryptography– Password-reset Guessing from Google & Psychology

• Evaluate military operations security (OPSEC)• Foundations

– Discrete mathematics, Galois fields, problems– Cryptographic algorithms (AES) and systems (Kerberos)– Attacks fall mainly outside of “crypto” per se– Large scale cybersecurity (Secure Systems curriculum)

26

A Port as a Secure Resilient Space

Spectrum Management Allowing Device Autonomy

27

Cognition Based Services

Axis 207W

© 1997-2007 Dr. Joseph Mitola III

Ambient Intelligence Challenges Spectrum Management

28

Multisensory Perception

Characters

Tokens/Words

Phrases

Dialogs

Scenes

Obs

erve

Pha

se H

iera

rchy

of R

einf

orce

d Se

quen

ces

Acoustic Sensors

Phonemes

Words

Phrases

Dialogs

Scenes

Keyboard/email Speech Vision

Scene

Place

Time

Vision Sensors

Texture

Objects, Flows

Expressions

Text ScanPHY

Protocols

Applications

RF-Location

RF Environment29

GSM Saturated but…

Dynamic Spectrum Available

[Riva Networks GSM Basestation]

Low Power, Building Shadowing, Creates GSM Reuse Opportunities

30

Building Shadows, Multipath and Reflections

WiFi 1 Mbps VoIP Coverage (Green)

31

Virginia Tech

AT&T

Polytechnic

Ground Truth

Polytechnic

908 MHzRosslyn, VA

Building Down

Penn State - < 2 dB Average Error

2 ½ D RSSI Estimation3D/4D CAD

MIMO PerformancePrecision Etiquette

© 1998 Penn State ARL

Contemporary 2 ½ D RSSI Estimation

RSSI – Received Signal Strength Intensity

32

RF Spatial Energy Distribution

© 2008 Remcom

© 2008 Remcom

33

Wireless LAN (WiFi) Proliferation

© 2006 Scientific American used with permission

DHCP-SIP On Steroids

© 2006 Scientific American used with permission

© 2006 Scientific American, Used with Permission

MIMO-Mobile

34

Cognitive Radio Economics Domains• Transportation – Emergency Response

– Next round of innovations• Agriculture

– Robotic on-farm carts, packaging machines– Reduce Sushi-Sashimi Time Lines

• Elder Care– Video-speech-RF scene integration

• Military and Coalition Partners– Higher risk R&D for leap-ahead innovation– Domains overlap all commercial domains

35

Radio Architecture Evolution

Observe

OrientEstablish Priority

Plan

Decide

ActOutsideWorld

Send a Message

Receive a Message

Set Display

Read Buttons Save Global States

Allocate Resources

Initiate Process(es)

Generate Alternatives

Evaluate AlternativesParse

Pre-process

Infer on Context Hierarchy

UrgentImmediate

Normal

Register to Current Time

PriorStates

NewStates

Learn

iCR Arcitecture

Haykin Radio-Centric

36

Machine LearningR

epre

sent

atio

n Sp

ace

Numeric

Symbolic

Learning StrategySupervised Unsupervised

ArtificialNeuralNetworksPowerful GeneralizationPerformance DegradesWhen IrrelevantFeatures are Present

ConceptualClusteringSet Cover UsingGeneralization &Specialization Decision Trees

Prod

uctio

n R

ules

Feat

ure

Vect

ors

Pred

icat

e C

alcu

lus

FeatureClusteringSet PropertyReinforcementEstimation overMeasurements,DocumentsSVM, Rough Sets

AbductiveInferenceOccam’s RazorOver StructuredFeature Spaces

Wor

d Ve

ctor

s

N-G

ram

s GeneticAlgorithmsBlind Learning, RobustSlow, Massively ParallelConstrained by the Coding of Chromosomes

Case-BasedStorage of ExamplesMemory BasedNearest-NeighborInductive RetrievalAdapt Pre-StoredSolutions to CurrentSituation(Does not requirea-priori model of the solution space)

Knowledge-BasedStructure backgroundknowledge in Rule BaseAcquires New RulesMay Use Certainty Calculus

EntropyNetworkLogic TreeTransformedto Neural Net(N-0.5, 0.5)

HiddenMarkovModels

Concept-BasedAcquires NewPredicates Tieresias

CLARION

© 2006 Dr. Joseph Mitola III Used with Permission 37

Secondary Use Concepts

38

Policy Language Foundations

Semantic Web

39

Problem with Policy Languages

• Policy Language = Computer Language– Symbols and rules (define grammar)– Semantic web uses relationships for meaning

• Policy Makers Use of Language– Lawyers, economists, businesses– Language refers to “best practices”– Language of Prior Experience

≠ Computer Language

• Why?

40

Spectrum Management via Cognitive Linguistics

< Bayi-Balan/>1

<Person/>

<Woman/><Frau/>

<Man/><Mensch/>

<Li/>

<Giraffe/>

<Animal/>

<Chair/>

Neonatal DevelopmentThings are earliest fundamental gestalts

Mother-self differentiation, reinforcementPlaces (stomach, home, box) containment

<Teenage Bride/>

Analytic

<Mommy/>

Experiential

Semantic WebCYC, SUO, SUMO, FOAF

1. In Dyirbal culture, the concepts male and female cannot be combined

Lakoff Basic Categories

41

Cognitive Linguistics Metaphors• Things [Lakoff-Jackendoff]

– Body experience of physical things: touch, eat, don’t eat, don’t touch• Mind metaphor: ideas, good ideas, bad ideas

• Places [Sensorimotor interaction with a Thing] Video Game– Body is container: eat/ defecate, drink/ urinate, feel good, get sick– Room is a container in which we are contained

• Activity is a place: goodness, success, failure, consequences– Non-visual space continues to exist: [certain] abstractions are real

• Paths [partially ordered (<) set of Places] [Lakoff-Jackendoff]– Physical movement experience

• Activation is motion metaphor; Activity is journey (plan) metaphor• Actions [Thing moving down a Path] [Mitola]

– Anger as burden, dangerous animal, opponent, fire, heat• Causes [Thing.Action that initiates or modulates Action] [Mitola]

– Some Thing initiates {some [other] thing to move down a path}• Reason via metaphor: bodily experience => internal experience => expertise

=> abstraction => logic, language as shared experience

42

LJM Basic Types for Radio

43

Cognitive Linguistics for Dynamic Spectrum

• Lakoff[1], Jackendoff[8], Ziemke[2], Mitola[6]– Language: pointer to sensorimotor behavior– Reasoning: metaphor (bindings) not logic

• LJM Cognitive Linguistics Orthogonalization– <Thing/> The most common entities in a domain– <Place/> Named vector fields for <Thing/>– <Path/> Associated <Places/> (sequence, hierarchy)– <Action/> Motion of <Thing/> on <Path/>– <Cause/> <Thing/> initiates-constrains <Action/>

44

Cognitive Linguistic Modeling of Radio

Things “Everybody knows”

CLA first identifies the few universally common Things of a domain

Places are the vector fields that characterize interaction possibilities for each Thing

Places “Near Field, BLOS”designate subspaces

Paths are sequences of places through which action may occur

Paths “Multipath, MIMO”Basis for actions

45

Modeling Radio DynamicsAction is the movement of a Thing along a Path

Actions“TX, RX”Induces

What happened

Transmit: Signal-In-Space = Thing.new(Tower)

Cause is the Thing that initiates an Action

Causes“ETSI”

ConstrainsWho to fine

Air Interface * Channel State

46

Genetic Programming [7]

• Specialized form of Genetic Algorithm (GA)• Population of Individuals

– Genome: Set of software components {+,-,if …}– GP Gene: Program tree or script per individual– Random initialization of [working] programs

• Iteration of populations to enhance fitness– Run programs on problems to evaluate fitness– Propagate, procreate, prune, cross-over, [mutate]– 1000 individuals/ 100 populations vs 20/10

47

SWR, SDR, DySPAN, and CR Domains

PHY

MAC

Network

…

Application

User

Location

Health…

Context

HW FW SWTX/RX

SWR Band

Ideal Cognitive Radio (iCR)Cognitive WirelessNetworks (CWN)QoI

Speech VisionRFSensory Perception

“CR0” – Dynamic Spectrum

Location Aware

Voice, Data, Video AdvancedGUI & QoI

© 2007 Federal Communications Commission, Used with Permission

SDR (Handsets)

Context Aware…

…

QoS

48

Regulatory Parallels: FCC-NTIACommercial Domains (FCC)

Military Domains (NTIA)49

Conclusions• Cognitive Radio

– Dynamic Spectrum Plus New Opportunities• Systems Engineering Approach

– Smarter Devices– Secure, Resilient Businesses, Homes, Ports– 3D RF-Video Sensor Integration

• Economics of Spectrum Management– Minimize Capital Expense for Infrastructure– Minimize Operating Expense for Services– Sustains social contracts

50

References1. George Lakoff, Women, Fire, and Dangerous Things ()19872. Ray Jackendoff, Semantics and Cognition, volume 8 of Current Studies in

Linguistics Series. MIT Press, Cambridge, Massachusetts 1983. 3. John R. Bender. 2001. Connecting language and vision using a conceptual

semantics, Master’s thesis, Massachusetts Institute of Technology.4. Keith Bonawitz, Anthony Kim, and Seth Tardiff, An Architecture for Word

Learning using Bidirectional Multimodal Structural Alignment Report for NSF Award IIS-0218861 W03-0605.pdf (Cambridge, MA: MIT) 2003

5. Peter Gärdenfors, “Representing actions and functional properties in conceptual spaces,” Body, Language, Mind: Embodiment T. Ziemke et al, Editors (Berlin: Mouton de Gruyter) 2007

6. M. Johnson and T. Rohrer “We are live creatures: Embodiment, American Pragmatism” in T. Ziemke, Ibid

7. John Koza, www.genetic-programming.org; Genetic Programming (Volumes 1 [92, MIT Press] through IV [Kluwer, 2003])

8. Jing Liu et al, “Moving Block Sequence and Organizational Evolutionary Algorithm for General Floorplanning with Arbitrarily Shaped Rectilinear Blocks” IEEE Trans Evolutionary Computation (NY: IEEE Press) Oct 08

51

Supplementary Material

53

Space-time Context

Spatial Context Plane CharacteristicsLevel Plane Members Space Time Contexts1 Global Regions 10,000 km year Itinerary2 Regional Cities 1000 km week Plan3 Metro Districts 100 km day Commuting4 Local Buildings 1-10 km hour Lunch5 Immediate Rooms meters minutes Dead reckoning6 Fine Scale Furniture wavelengths µsec Habit7 Internal HW, SW microns ns Architecture

Resource Description Framework (RDF)/Schema DARPA Agent Markup Language (DAML) Ontology Web Language (OWL)“Semantic Web” – Slow to Catch On

© 1997-2007 Dr. Joseph Mitola III

Knowledge Engineering Bottleneck

54

CLM Radio Image Schema

AM-FM-OOK

FDM FSK-PSK TDM

1G

FDMACoding

2GMSK

TDMA

CDMA

3G

Femto

MIMO

Welcome to the Cognitive Radio “[Physical Layer] Signals” VillageA Cognitive Linguistics Center-Periphery Image Schema

55

Machine Learning and Policy

56

Function Conventional Self-Aware Remarks

Iterate Programmers learn through practice

Loop detect M&S Engines

Acquire Architect defines blocks Learn blocks CBR, RL

Encapsulate Define block boundaries GP per platform Over self-descriptions

Apply Compile, ignore internals Semantic model Behavioral properties

Explain Diagnose via IDE, logs, instrumentation

Re-iterate Causal modeling (vs rule unwrapping)

Signal Detection GPSignal genes: s*.s, PSD, max(*), …Detection GP chromosomes (Places anchored to CL Things):

Prepare Signal : get, normalize, s.*s, …Estimate Spectrum : FFT, PSD.walsh, …Extract Parameters : max(Sp), peaks(Sp), …Assert Estimate : Index = Ri; Value = Rj;Prepare result: (Ri -> fc)

Script is a sequence of function calls defining a PathScript = Path: Signal.unknown … Spectrum.known)

Fitness FunctionsConsistency with Training Things-Places and Signal Models

Progress: spectrumGA writes Matlab code; learns new spectra

57

Cognition Includes Self Awareness

Capabilities

Resources

Self

Dialogs

Scenes

Obs

erve

Pha

se H

iera

rchy

of R

einf

orce

d Se

quen

ces

PHY

Protocols

Applications

RF-Location

Dialogs

Scenes

Self-Perception RF Environment User’s Perception

tn

Scene n

Place

Time

Sensors

Primitives

Objects, Flows

User

Self Scan

58

Space-time-RF Etiquette

1 – Don’t jam other users2 – Overcome interference3 – Create affordable,

effective wireless connection

1

Radiated Power

Interference2

3

BeamformingOr MIMO

© 1997-2007 Dr. Joseph Mitola III Permission for Government Use Only 59