1. Capacity Enhancement with Relay Station Placement in Wireless Cooperative Networks Bin Lin1, Mehri Mehrjoo, Pin-Han Ho, Liang-Liang Xie and Xuemin (Sherman) Shen Department of Electrical and Computer Engineering,University of Waterloo,Canada IEEE WCNC 2009

2. Outline Introduction Goal Problem Formulation A GA-Based Heuristic Algorithm Simulation Conclusions

3. Introduction IEEE 802.16j Transparent Relay Station Capacity Enhancement MR-BS Transparent Relay MS Low rate RS High rate

4. Introduction Transparent Relay Station Placement Exploiting the utmost performance benefits base on OFDMA using a different code Capacity Maximization relay(RS) destination(SS)source(BS) Cooperative relaying on 3-node relay model

5. Introduction The achievable rates for the destination node [10] [10] A. Host-Madsen and J. Zhang, “Capacity bounds and power allocation for wireless relay channels,” IEEE Trans. on Inf. Theory, vol. 51, no. 6, pp. 2020-2040, Jun. 2005. α is the path loss exponent Θ is the transmit power allocation ratio of the source node between the “source-relay” path and “source-destination” path SNR of “source-relay” path SNR of “source-destination” and “relay-destination”path r = Case1: Waiting for “source-relay” r = Case2: Buffer in “source-relay” Shannon function P s is the transmit power of BS P r is the transmit power of RS destination(SS) source(BS) relay(RS) d sr d rd C( SNR ) = B log 2 ( 1+SNR )

6. Introduction Capacity Maximization RS Placement (CMRP) problem The CMRP problem is to maximize the system capacity Given the locations and the minimum traffic demands of N SSs The finite locations of M CPs for deploying RSs Total bandwidth allocated to the cell Transmit power of BS and RS

7. Goal Based on 802.16j Relay Station Cooperative relaying on 3-node relay model using a different code on OFDMA Develop an optimization framework for the Capacity Maximization RS Placement (CMRP) problem Using Genetic Algorithm based heuristic to solve CMRP problem

8. Problem Formulation Network Model One BS, multiple RS s, and fixed SS s CP s is candidate positions to deploy RS s

9. Problem Formulation Capacity Maximization RS Placement (CMRP) formulation N : The number of SSs M : The number of CPs : indicates the index of the chosen CP serving a specific SS : Bandwidth-allocation of SS n CP m SS n =1 CP m SS n =0 r mn is the achievable rate for the destination node using cooperative relay CP m BS SS n

10. Problem Formulation Constraint for CMRP problem (1) States that each entry in the location-allocation matrix is binary (2) Ensures exclusively allocation of each SS to an RS(CP) CP 1 SS n CP 2 CP m … : indicates the index of the chosen CP serving a specific SS : The set of SSs ， | | = N. : The set of CPs, | | = M. CP m SS n =1 CP m SS n =0

11. Problem Formulation Constraint for CMRP problem (3) Ensures the throughput of each SS is larger than its minimum traffic demand : The minimum traffic demand of SS n : The set of SSs ， | | = N.

12. Problem Formulation Constraint for CMRP problem (4) Satisfies locating K RSs among the M CPs : indicates the index of the chosen CP serving a specific SS :The number of RSs to be deployed within the cell M=1 => max(0,1-2(2)) = 0 CP 1 CP 2 CP 3 SS 1 SS 2 SS 3 M=2 => max(0,1-2(1)) = 0 M=3 => max(0,1-0) = 1 1 = 3-2 RS

13. Problem Formulation Constraint for CMRP problem (5) T he bandwidth constraint of the cell : Bandwidth-allocation of SS n B : The total radio bandwidth allocated to the cell.

14. Problem Formulation The achievable rate for the destination node : The distance between node i and node j. : The transmit power of BS : The transmit power of RS : Path loss exponent : Source power allocation ratio between the relay path and direct transmission path, CMRP Shannon function C( SNR ) = B log 2 ( 1+SNR )

15. A G-A Based Heuristic Algorithm CMRP problem is NP-hard SelectionReplacementCrossoverMutation [12] D. E. Goldberg, Genetic algorithms in search, optimization, and machine learning. Reading, MA, Addison-Wesley, 1989. natural selection better individuals (for the environment) population individual natural selection cycle :

16. A G-A Based Heuristic Algorithm Overview SS 1 SS 2 SS 3 BS CP 5 CP 4 CP 3 CP 1 CP 2 If number of RSs is 3 SS 1 SS 2 SS 3 BS CP 5 CP 4 CP 3 CP 1 CP 2 fitness value : randomly select K among M CPs ﹛ CP 1, CP 3,CP 4 ﹜ b n represent the associated CP to SS n r b n,n represents the achievable rate of SS n relayed with the RS located at CP b n b n = m a mn = 1

17. A G-A Based Heuristic Algorithm Overview SelectionReplacementCrossoverMutation SS 1 SS 2 SS 3 BS CP 5 CP 4 CP 3 CP 1 CP 2 If number of RSs is 3 SS 1 SS 2 SS 3 BS CP 5 CP 4 CP 3 CP 1 CP 2 fitness value : Larger ﹛ CP 1, CP 3,CP 4 ﹜

18. A G-A Based Heuristic Algorithm Overview SelectionReplacementCrossoverMutation SS 1 SS 2 SS 3 Solve CMRP problem BS CP 5 CP 4 CP 3 CP 1 CP 2 If number of RSs is 3 SS 1 SS 2 SS 3 BS CP 5 CP 4 CP 3 CP 1 CP 2 ﹛ CP 1, CP 3,CP 4 ﹜

19. Overview SelectionReplacementCrossoverMutation SS 1 SS 2 SS 3 BS CP 5 CP 4 CP 3 CP 1 CP 2 If number of RSs is 3 SS 1 SS 2 SS 3 BS CP 5 CP 4 CP 3 CP 1 CP 2 Crossover SS 1 SS 2 SS 3 BS CP 1 CP 4 CP 3 CP 1 CP 2 Check constraints : ﹛ CP 1, CP 3,CP 4 ﹜

20. Overview SelectionReplacementCrossoverMutation If number of RSs is 3 Mutation SS 1 SS 2 SS 3 BS CP 5 CP 4 CP 3 CP 1 CP 2 SS 1 SS 2 SS 3 BS CP 5 CP 4 CP 3 CP 1 CP 2 Check constraints :

21. A G-A Based Heuristic Algorithm Select Replacement Crossover Mutation (a) a predefined number of iterations N G are completed (b) the resultant capacity of the whole cell is close enough to the upper bound C UB (c) The improvement of capacity is negligible after a certain number of iterations N T

22. Improved GA for the CMRP Problem Search space reduction Fast convergence of bandwidth allocation

23. Improved GA for the CMRP Problem Search space reduction To exclude the CPs that can not provide a significant performance gain in terms of achievable rate. G = r relay − r 0 = 0.8bit/sec,When P BS = 1W, P RS = 0.5W, α = 2

24. Improved GA for the CMRP Problem Fast convergence of bandwidth allocation Random allocation of both a mn ’s and ω n ’s in each iteration of GA Linear Programming (LP) problem can be solved to obtain the optimal values of ω n to maximize C

25. Simulation Simulation parameters GA PARAMETER SETTING P BS 1w P RS 0.5w α2 SSs24 CPs35 Bandwidth allocated to the cell20MHz IEEE 802.16j parameter crossover mutation Error rate Iteration times Number of individual

26. Simulation The cell layout, geographical traffic demand distribution and best CPs (RSs) for SSs.

27. Simulation Convergence comparison of GA and improved GA

28. Simulation Capacity vs. number of RSs given the same GA termination

29. Simulation Achievable rate comparison for each SS in case of with best CPs, 6 RSs and direct transmission. (6RSs)

30. Conclusions In this paper, we have explored the joint RS placement and bandwidth allocation problem in wireless cooperative relay networks. Develop an optimization framework for the Capacity Maximization RS Placement (CMRP) problem And using Genetic Algorithm based heuristic to solve CMRP problem