1. COLLABORATIVE SPECTRUM MANAGEMENT FOR RELIABILITY AND SCALABILITY Heather Zheng Dept. of Computer Science University of California, Santa Barbara

2. The Critical Need for Dynamic Spectrum Management 2  Explosion of wireless networks and devices  Static spectrum assignments are inefficient  Under-utilization + over-allocation  Artificial spectrum scarcity  Solution: Migrate from long-term static spectrum assignment to dynamic spectrum access

3. Challenges Facing DSA Dynamic, Heterogeneous Spectrum Demand Dynamic, Heterogeneous Spectrum Availability Large number of nodes Manhattan (Courtesy of Wigle.net)

4. Requirements for DSA  Scalability and speed  Support a large number of nodes  Adapt to time-varying demands  Efficiency + Fairness  Maximize spectrum utilization  Avoid conflict  Reliability  Provide QoS  Minimize outages outage

5. A Few Observations

6. Collaborative Spectrum Allocation 6 Action: Iterative Explicit Coordination Self-organize into coordination groups Negotiate to allocate spectrum in each group Iteratively set up groups to improve utility Fast convergence: coordination stops when no local improvement can improve utility Action: Iterative Explicit Coordination Self-organize into coordination groups Negotiate to allocate spectrum in each group Iteratively set up groups to improve utility Fast convergence: coordination stops when no local improvement can improve utility Goal Goal: Allocate spectrum to maximize system utility Assumption Assumption: 100% willingness to collaborate Goal Goal: Allocate spectrum to maximize system utility Assumption Assumption: 100% willingness to collaborate Node Collaboration Cao & Zheng, SECON 2005, Crowncom07, JSAC08, MONET08

7. Analytical Properties 7 Fast Convergence Fast Convergence: The system converges after at most O(N 2 ) local adjustments, N= network size Guaranteed Spectrum Allocation Guaranteed Spectrum Allocation: Each node n’s allocated spectrum A(n) ≥ Poverty Line PL(n) Guaranteed Spectrum Allocation Guaranteed Spectrum Allocation: Each node n’s allocated spectrum A(n) ≥ Poverty Line PL(n) Total usable spectrum Conflict degree Cao & Zheng, SECON 2005 Node Collaboration

8. Tightness of Poverty Line 8 Percentage of Instances A(n)/PL(n)

9. Bandwidth-Aware Poverty Line  Each channel i has a weight of B i (n)  Each node’s spectrum allocation A(n)= ∑ a i (n)B i (n)  Extended poverty line A(n) > PL(n) 9 Cao & Zheng, Crowncom07

10. Traffic-Aware Poverty Line 10  Each infrastructure node n supports t n users  Maximize end-user fairness  Each infrastructure node’s spectrum has a lower bound

11. Making it Work in Practice: Distributed Coordination Protocol 11  Poverty line is an integrated knowledge about spectrum sharing  Use it to initiate coordination  Enable multiple parallel coordination events  Minimize adaptation delay

12. Simulations: Coordination Delay 12 # of Local coordination scales linearly with the # of APs Adaptation delay flattens out because of parallelism. 1Mbps Wireless Backhaul running CSMA/CA among APs

13. Rule Regulated Spectrum Allocation Implicit Coordination Action: Iterative Independent adjustments Nodes observe spectrum usage in proximity Independently adjust self spectrum usage predefined rules Regulated by predefined rules Action: Iterative Independent adjustments Nodes observe spectrum usage in proximity Independently adjust self spectrum usage predefined rules Regulated by predefined rules Goal Goal: Allocate spectrum to maximize system utility Assumption Assumption: comply to rules, no handshaking Goal Goal: Allocate spectrum to maximize system utility Assumption Assumption: comply to rules, no handshaking Zheng & Cao, DySPAN 2005 JSAC 2008 Poverty Line based Rules Poverty Line based Rules: Rely on poverty line to determine whether to adjust and how to adjust. 13 The same analytical Poverty Line Bounds and O(N 2 ) complexity The same analytical Poverty Line Bounds and O(N 2 ) complexity

14. Required Hardware Functionality  Conflict Detection  Explicit coordination  A control path among conflicting peers  Implicit coordination  Sophisticated environmental sensing module  Non-contiguous spectrum usage  Behavior enforcement

15. From Adaptation to Reliability outage See Lili Cao’s Poster Tomorrow

16. Lessons Learned  Much of large-scale distributed wireless systems depend on mutual cooperation  To build robust systems that can be deployed in real life, we need to be flexible in our design to allow for flexible levels of cooperation  Hybrid architecture helps to provide reliability  Controlled regulation at a coarse time-scale  Individual adaptation at a fine time-scale  Interference makes it very challenging  Current: Simplification via conflict graph  Future: Addressing physical interference constraints