1. anandps@cs.sunysb.edu Fast Spectrum Allocation in Coordinated Dynamic Spectrum Access Based Cellular Networks Anand Prabhu Subramanian*, Himanshu Gupta*, Samir R. Das* and Milind M. Buddhikot *Stony Brook University, NY, USA Bell Labs, Alcatel-Lucent, NJ, USA

2. anandps@cs.sunysb.edu Current state-of-the-art in Spectrum Allocation Static Allocation Multi-year license agreements Spectrum is access limited rather than throughput limited Rigid specification of usage parameters eg: technology, power,etc Goal:  Break the Spectrum Access Barrier  Enable networks and end user devices to dynamically access variable amount of spectrum on a spatio-temporal scale

3. anandps@cs.sunysb.edu Coordinated Dynamic Spectrum Access (CDSA) Model Regional Spectrum Broker Spectrum Demand and Allocation Spectrum Pricing, Allocation Algorithms And Policies Mesh Networks Cellular Networks Fixed Wireless Access MN Region R 1 MN Region R 2 802.16 CPE 802.16a CPE Region R 4 CPE Region R 3 Internet

4. anandps@cs.sunysb.edu Contributions  Formulate the Spectrum Allocation problem in the CDSA model as two optimization problems  Max-Demand DSA  Min-Interference DSA  Design fast and efficient algorithms with provable performance guarantees

5. anandps@cs.sunysb.edu Spectrum Allocation – Reference Architecture Spectrum Broker A region R controlled by the Spectrum Broker Base stations of different RIPs Coordinated Access Band Demands: (d min, d max ) Batched Demand Processing Model

6. anandps@cs.sunysb.edu Interference Constraints 2 1 354 810976171918 141615 1112 132325242726202122 Different RIPs Co-located Cross Provider Constraint Remote Cross Provider Constraint

7. anandps@cs.sunysb.edu Interference Constraints 2 1 354 810976171918 141615 1112 132325242726202122 Different RIPs Soft Hand-off Constraint

8. anandps@cs.sunysb.edu Interference Graph 1 2 3 4 5 6 7 8 9 12 11 10 1 Base stations of different RIPs 2 3 7 6 4 5 8 9 11 10 12 Spectrum Allocation Variation of Graph Coloring  Cannot always find a feasible solution  Formulate as optimization problems  Max-Demand DSA  Min-Interference DSA  NP Hard

9. anandps@cs.sunysb.edu Max-Demand DSA Maximize the overall demands served among all base stations with the available number of channels such that no constraint is violated Input to the problem:  Interference Graph  Minimum and maximum demands of each node  Available number of channels  Check if the minimum demands of all base stations can be served  If yes, serve as many demands as possible using available channels

10. anandps@cs.sunysb.edu Max-Demand DSA Algorithm 12 43 G(V,E) d min =2 11 22 3 3 44 G min (V min,E min )  Pick K independent sets (IS) in G min  If all nodes in G min are picked proceed to Phase II Phase II:  Add d max (i)-d min (i) copies for each node i to construct G max  Pick as many independent sets as possible in G max Phase I:

11. anandps@cs.sunysb.edu Max-Demand DSA Algorithm: Performance Guarantee  Interference Graph is modeled as a δ -degree bounded graph  When picking independent sets, pick the nodes in the order of maximum degree.  We can prove that |IS| |OPT| δ  Phase II of the Max-Demand DSA achieves an approximation ratio of 1- 1 e 1 δ

12. anandps@cs.sunysb.edu Min-Interference DSA Input to the problem:  Interference Graph  Maximum demands of each node  Available number of channels Minimize overall Interference when all demand ( d max ) of the base stations are serviced 1 2 3 4 5 6 7 8 9 12 11 10 Max K Cut: Assign nodes to different colors so as maximize the number of edges between nodes with different colors

13. anandps@cs.sunysb.edu Algorithm R k for Multi-Color Max-K-Cut: For each node i, randomly pick d max (i) different colors from the available K colors 1 2 3 4 5 6 7 8 9 12 11 10 d max =2 K=5 2 1 1 2 3 3 4 4 5 5 6 6 7 7 88 9 9 10 11 12 11 By a simple probability argument, we can prove that the weight of the cut (edges crossing partitions) produced by R K is 1-1/K of the optimal

14. anandps@cs.sunysb.edu Min-Interference DSA: TABU Search Algorithm  Start from the random solution  In each iteration, generate certain number of neighboring solutions  Pick the solution with least interference  Repeat until no improvement for certain number of iterations 2 1 1 2 3 3 4 4 5 5 66 7 7 88 9 9 10 11 12 11

15. anandps@cs.sunysb.edu Performance  Graph Based simulations with 1000 nodes  40 - 240 channels  Demands 10 - 80  Max-Demand DSA performs very well  Min-Interference DSA: Random  1/K  Min-Interference DSA: Tabu performs extremely well compared to Random

16. anandps@cs.sunysb.edu Future Work  Test our algorithm performance on realistic network topologies from existing service providers  Build an experimental spectrum broker simulator that accounts for advanced features of the CDSA model such as demand scope, stickiness, fairness etc.