Quantifying Aspects of Cognitive Radio and Dynamic Spectrum Access
Performance; and Interference Tolerance as a Spectrum Principle
Preston MarshallUniversity of Southern California
Viterbi School of EngineeringInformation Sciences Institute
pmarshall @isi.edu
Centre for Telecommunications Value Chain Research, Electrical Engineering Department
Trinity College, Dublin, Irelandpmarshal @tcd.ie
University of Southern CaliforniaViterbi School of EngineeringInformation Sciences Institute
pmarshall @isi.edu
Presentation Topic
• General Trend to View DSA as:– Of Benefit to Unlicensed, Secondary Users of Spectrum– Not Particularly Beneficial to Primary Users Already Provisioned
with Spectrum
• Present Alternative Vision– DSA is Highly Beneficial to Environmentally Stressed Devices– Existing “Mission Critical” Primary Users Could Most Benefit from
DSA, Even if they Have Adequate Spectrum Access– Interference Tolerance Can Be More Effective Than Interference
Avoidance
• Implication:– Instead of Relocating Existing Services, We Could Provide Mutual
Benefit By Transitioning Them to DSA– Applicability:– Emerging Self-Forming Networks, Many Hub-Spoke Systems
Instead of “Interference Avoiding” DSA, Transition to DSA-Enabled Interference-Tolerance
Agenda
• A Model for Spectrum Density and Energy• Front End Overload and Non-Linearity Issues:
– Reliability Issues with Fixed Spectrum Assignments– Improvement in Likely Front End Performance with DSA– Reduction in Required Linearity for Equivalent Performance
• What Density an Be Achieved if a Device Can Assume Other Devices are Interference Tolerant?– The Impact of DSA + Propagation Exponent Awareness– Selection of Optimal Constellation Depth
• How Can Topology Management Enable Density?• Fungibility of Benefits• Implications on Spectrum Management Policy
Objectives of Closed Form Expression of Spectrum
Enable Cognitive Radio and Dynamic Spectrum Access Researchers to:1. Simulate a wider range of spectrum environments than
can be sampled and analyzed; 2. Perform analysis of radio performance, without
researchers having large databases of environments; and
3. Provide provable assertions about cognitive radio performance in a range of potential environments.
Examined from Two Perspectives:1. Low Signal Levels and Fixed Bandwidths for Signaling
Channels2. High Energy, Proportional to Frequency Bandwidths
for Effects on Front End Linearity
Spectrum Analysis Methodology
• Used Six NSF Spectrum Measurements Reported by McHenry (Shared Spectrum and IIT)
– All Had Consistent Methodology, Instrumentation and Reporting
• Developed Closed-Form Cumulative Distributions for the Signaling (Fixed b0) and Pre-Selector (BW) Bandwidths
• Developed Estimators to Synthesize Arbitrary Environments in Terms of Density and Intensity Variables
• Bandwidth Treated as Independent to Recognize Correlation Between Adjacent Frequencies
Sample Location Date(s)
Chicago Illinois Institute of Technology, Chicago, IL November 16 to 18, 2005
Riverbend Riverbend Park, Great Falls, Virginia April 7, 2004
Tysons Tysons Square Center, Vienna, Virginia April 9, 2004
New York
Republican National Convention, New York City, New York (Day 1 and Day 2)
August 30, 2004 - September 2, 2004
NRAO National Radio Astronomy Observatory (NRAO), Green Bank, West Virginia
October 10 -11, 2004
Vienna Shared Spectrum Building Roof, Vienna, Virginia
Dec. 15-16, 2004
A Total of 52,436 MATLAB Files and 1,073 MB of Data
Monotonic Estimator
Intended to Provide a Mechanism to Synthesize Spectrum Distributions for Arbitrary Environments
– Like Chicago, just …
Two Indices:IDensity Mean Signal Level of the
Median EnergyIIntensity Range from Weakness
to Strongest Signal (25kHz)
1 MHz Used for Indices
Importance of Front End Energy Effects
• The Last Slides Show that High Energy Signals Are Rare in terms of Frequencies Containing them, but Common in Terms of Frequencies Impacted– A High Power 100 kHz Signal may impact only 4 of 10,000s of possible 25 kHz
Channels, but– It can Dominate the Energy in 20% of the Pre-Selector Settings
• All these Frequencies May be Unusable, Even through they are “White Space” Due to the Effect of Limited Receiver Dynamic Range– AGC No Help, since this is Adjacent Channel
• Looking at Spectrum Occupancy Alone Does Not Paint a Sufficient Picture of the Interaction of a Cognitive Radio and its Environment
• DSA Bands Are More Likely to Stress Linearity than Current Allocations as We Go Beyond the Wi-Fi Bands!– No Longer Segregated with Low Power Sources– Sharing Bands with High Power Sources, Like Broadcast– 10 Times More Density → 10 dB Increase in Energy → 30 dB Increase in 3rd
Order Intermodulation
• Reduced RF Performance of Low-Dynamic Range CMOS RF Circuits and Digital Filters
RF Environment Energy Management Key to Robust Operation and
Affordability• Even Open Frequencies Not Usable in
High-Energy RF Environments, ex. Co-Site– Frequencies can be “Perfectly
Assigned”, but RF Cannot Deal with Energy Density
– Even Ultra-High Quality Front Ends, Experience 20+ dB Increase in Noise due to Inter-Modulation
• “Better” Frequency Management not an Answer– Intractable Problem for Centralized
Management• “Better” Technology not an Answer
– Can Not Throw Linearity at the Problem– Energy Costs of High Linearity
Unacceptable in Battery Devices
More “Nextel-Public Safety” interactions due to Non-Linear Effects More “Nextel-Public Safety” interactions due to Non-Linear Effects (Co-Site) Make Frequency Management Inadequate in Some Dense (Co-Site) Make Frequency Management Inadequate in Some Dense
SpectrumSpectrum
INPUT SIGNALS
Example is input power = IIP3
LNA OUTPUT
Mapping Input Energy to IMD Noise Energy
• Analysis of over 90 Million Spectrum Measurements yield expected relationship of Input and Output Energy
• Order is 3.25, Reflecting Higher Degree of Correlation at Upper Energy Range
• Mean 11 dB Below Pure Two Tone IMD Product
IMD3 = k1 Pin - 2 IIP3 -k2
k1 = 3.25, k2 = 11.8
where IMD3, IIP3 and Pin
are in dBm
Only 1 in 10-4 points shown
Noise Floor Elevation
Probability Distribution of Intermodulation Induced Noise Floor Elevation when using Pick
Quietest Band First Algorithm
Non-Cognitive Radio Noise Floor Elevation for IIP3 = -5 dBm in Chicago Spectrum
Non-Cognitive Radio Has Significant 3rd Order Intermodulation Noise Elevation, Even for High Performance Filters
Cognitive Radio, Even with Poor Filters, Has Very Low Noise Elevation
With Reasonable Filter (<20% bandpass) there is Essentially Zero Chance of Noise Floor Elevation
Comparison of IMD3 Noise for a Range of IIP3 Points (90% Case)
Cognitive Radio (ideally) Enables a 30 dB Reduction in Required IIP3 Performance, and Creates a Lower Noise Floor Simultaneously, even for Moderate Filter Selectivity (20%)
Non-Cognitive
Cognitive
Lower Intermodulation Noise Floor and Major Reduction in Required
Linearity
Noise Floor Reduction at the Same IIP3 Level
Benefits are a Function of Required Reliability
• The Benefits of Front End Loading Adaptation is Driven by the Environment and the Level of Reliability• As Reliability Needs Increase, the Benefits of Adaptation Increase Accordingly• Intensity Can Be Handled, But At Extreme Values of Density, Even Cognitive Adaption Has Constraints
on Performance Enhancement– Not Surprising, a Few Strong Signals are OK, but Many Strong Signals Have a Chance of Hitting all Pre-Selector Candidates
• Note that if DSA Succeeds, Most RF Environments Will Become Denser, and More Like the Urban Environments
90% Environments 99% Environments
Front End Performance Conclusions
• Linearity and Filtering Are Major Cost Drivers in Reasonable or Better Performing Wireless Devices
• Integration of Dynamic Spectrum and Cognitive Radio Offers a Unique Opportunity to Address one of the Critical Analog Circuit Limitation in Wireless Systems– Significant Anecdotal Evidence of Severity– Will Become More Significant as Density Increases
• Offers Designers Opportunity to Both Significantly Increase Reliability and Performance and Reduce High Analog Performance Requirements
• New Business Case for DSA: It Can Be Less Expensive than a non-DSA Device of Equivalent Performance
Interference Tolerance Requires We “Break Up” Network into Small,
Interconnected Sub-NetworksToday’s Mesh or MANET Multi-Frequency Network
• Low Reliability Due to Single Link Routes• All Radios Interfere with Each Other,
Even if they can not Communicate• Bandwidth Drops as More Radios Added
to Network• Bandwidth Constrained by Mutual
Interference – More Nodes do Not Create More Capacity
• Large Number of Nodes on Single Frequencies
Color Depicts all radios on the same frequency
Color Depicts sub-net Frequencies
MIMO Mode Not Depicted
• Multiple Links and Routes Provide High Reliability
• No Single sub-Network is Large Enough to Have Scaling Issues
• More Sub-Networks are Created as More Nodes Join the Overall Network
• Bandwidth Increases as More Radios Added to Network
• Diversity in Frequency Avoids Interference
A Fundamentally New Approach to Network Organization Was Needed to Ensure Scalability
A Fundamentally New Approach to Network Organization Was Needed to Ensure Scalability
Dynamic Adaption as Enabler of Dynamic Networks
MIMOMIMO
BeamBeamFormingForming NullingNulling
TopologyTopologyPlanningPlanning
SpectrumSpectrumPlanningPlanning
DeviceDeviceSpurs, …Spurs, …
RelocateAround
Spur
SpectrumToo Tight
Re-planAcross
Network
Re-planTopology
UnavoidableStrongSignal
NeedMore Range
Each Technology Can Throw “Tough” Situations to other More Suitable Technologies without Impact on User QOS
No Good MIMO Paths
Network-Wide
Radio Device
Link
Move to New Preselector
BandStrongStrong
NeighborNeighborSignalSignal
DynamicDynamicSpectrumSpectrum
Interference Avoidance vs.Interference Management
Session
Presentation
Link
Physical
Network
Transport
Application
Avoid Interference
Interference Avoidance(Evolving Dynamic Spectrum Access)
Interference Tolerance Essential to Maximize Network Capacity
Interference Tolerance Essential to Maximize Network Capacity
Sense and Balance Interference
Manage Collision Events
Multiple RoutesAvailable
Delay Tolerant Transfers
Delay Tolerant Applications
Interference Tolerance and Management
MIMO for Nulling
Network Interference Tolerance vs. Node Interference Avoidance
• We Imagine A Mobile Operating Area• When Interfered with, Nodes Respond by Relocating• Closed Form, with Probability Distribution of Propagation Exponent Modeled as in Anderson• Index of DSA Performance (IDSA):
– (Event Time+ Relocation Time)/Event Time– Used Worse Case: Each Sensing Event is Independent– Used Reported XG Performance for an IDSA of 1.75 (100 ms Sensing, 175 ms relocation)
• Optimal Interference Rate is Orders of Magnitude Higher than Typically “Acceptable”• Resulting Aggregate Throughout is Orders of Magnitude More• Increase in Density Is More than Results from Just “Finding” Open Spectrum • Conclusion: Interference is Best Solved as a Network Issue, Not a Link Issue
Probability of Interference vs. Density Aggregate Throughput vs. DensityMaximum Aggregate Throughput Occurs at High Interference Rate
Typical Manual De-confliction
Interference Tolerant
Operating PointThroughput Benefit in
Moving from Manual to Maximal Aggregate
Throughput Operating Points
Density Benefit in Moving from Manual to Maximal Aggregate Throughput Operating Points
α-Aware, Optimal Bits/Hertz
• Optimal Spectrum Usage Does Not Occur With Maximal Bits/Hertz – WHEN SPECTRUM RE-USE IS INCLUDED IN CONSIDERATION!
• Optimal Modulation Depth is a Function of the Propagation Exponent – Situational, rather than Specifiable
• Cognitive Radio Can Increase Density of Usage by Factor of 5, or More, if it Adjusts Modulation Based on Actual Propagation (But Uses More Hz)
Bits/Area vs. Bits/HertzShowing Bits * Area with More than 3 dB
Interference vs. the Bits/Hertz for a Range of Propagation Constants (α)
Optimal Bits/Hertz is a Function of Propagation α
The root of the derivative of SIE ((bits/Hertz)/ Area) ratio yields the optimal operating point
Consistent
Reference Point is 1 Bit/Hertz
Published Papers
“Extending the Reach of Cognitive Radio,” Proceedings of the IEEE, Vol. 97, No. 4, pp. 612-625, Apr. 2009.
“Cognitive Radio as a Mechanism to Manage Front-End Linearity and Dynamic Range,” IEEE Communications Magazine, Mar. 2009.
“Spectrum Awareness and Access Considerations,” in Cognitive Radio Technology, 2nd Edition, B. Fette, Ed. Academic Press, 2009.
“Recent Progress in Moving Cognitive Radio and Services to Deployment,” in 9th IEEE International Symposium on a World of Wireless, Mobile and Multimedia Networks, June 2008.
“From Self-Forming Mobile Networks to Self-Forming Content Networks,” in Association of Computing Machinery Mobile Communications Conference, Sept. 2008.
“Closed-Form Analysis of Spectrum Characteristics for Cognitive Radio Performance Analysis,” in 3rd IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks, 2008.
“Progress towards Affordable, Dense, and Content Focused Tactical Edge Networks, in 2008 IEEE Military Communications Conference, 2008.
“Dynamic Spectrum Management of Front End Linearity and Dynamic Range,” in 3rd IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks, 2008.
Summary of Generalized Cognitive Radio Functionality
Cognitive Radio Environments
Front-end Linearity Management
Minimization of Interference Effects through
Interference Tolerant DSA Mechanisms
Spectral Footprint Management
Extension of Principles to Network Level Decision Making
Overall
Spectrum Policy Implications
• DSA is Highly Advantageous, Even if You “Own” Spectrum
• Current “Relocation” Approach Fails to Recognize Advantages of DSA to Incumbent Users
• New Concepts Possible• Instead of Relocation” Trust; Have “Interference
Tolerance Trust”– Fund Transition to Interference Tolerant Systems by Current
Primary User– Enable Secondary Use of DSA, Subject to Aggregate Loading
which Impacts Primary’s Performance– Primary and Secondary Benefit!– No Need to Change “Ownership”– There is a Win-Win Available (for Primary Users that Can Create
Interference Tolerant Modes)
P. Marshall, “A Potential Alliance for World-Wide Dynamic Spectrum Access: DSA as an Enabler of National Dynamic Spectrum Management”, New America Foundation Issue Paper #25, June 2009.
Questions?
Preston MarshallUniversity of Southern California
Viterbi School of EngineeringInformation Sciences Institute
pmarshall @isi.edu
Centre for Telecommunications Value Chain Research, Electrical Engineering Department
Trinity College, Dublin, Irelandpmarshal @tcd.ie
University of Southern CaliforniaViterbi School of EngineeringInformation Sciences Institute
pmarshall @isi.edu