1. *P. C. Weeraddana **M. Codreanu, **M. Latva-Aho, ***A. Ephremides * KTH, Royal institute of Technology, Stockholm, Sweden ** CWC, University of Oulu, Finland ***University of Maryland, USA 2013.02.19 Multicell MISO Downlink Weighted Sum-Rate Maximization: A Distributed Approach

2. Motivation  Centralized RA requires gathering problem data at a central location -> huge overhead  Large-scale communication networks -> large-scale problems  Distributed solution methods are indeed desirable  Many local subproblems -> small problems  Coordination between subproblems -> light protocol 12.10.2015CWC | Centre For Wireless Communications2

3. Motivation  WSRMax: a central component of many NW control and optimization methods, e.g.,  Cross-layer control policies  NUM for wireless networks  MaxWeight link scheduling for wireless networks  power and rate control policies for wireless networks  achievable rate regions in wireless networks 12.10.2015CWC | Centre For Wireless Communications3

4. Challenges  WSRMax problem is nonconvex, in fact NP-hard  At least a suboptimal solution is desirable  Considering the most general wireless network (MANET) is indeed difficult  A particular case is infrastructure based wireless networks  Cellular networks  Coordinating entities  MS-BS, MS-MS, BS-BS  Coordination between subproblems -> light protocol 12.10.2015CWC | Centre For Wireless Communications4

5. Our contribution 10/12/2015CWC | Centre For Wireless Communications5  Distributed algorithm for WSRMax for MISO interfering BC channel; BS-BS coordination required  Algorithm is based on primal decomposition methods and subgradient methods  Split the problem into subproblems and a master problem  local variables: Tx beamforming directions and power  global variables: out-of-cell interference power  Subproblems asynchronous (one for each BS)  variables: Tx beamforming directions and power  Master problem resolves out-of-cell interference (coupling) P. C. Weeraddana, M. Codreanu, M. Latva-aho, and A. Ephremides, IEEE Transactions on Signal Processing, February, 2013

6. Key Idea 10/12/2015CWC | Centre For Wireless Communications6

7. Key Idea 10/12/2015CWC | Centre For Wireless Communications7

8. Key Idea 10/12/2015CWC | Centre For Wireless Communications8

9. Key Idea 10/12/2015CWC | Centre For Wireless Communications9

10. Key Idea 10/12/2015CWC | Centre For Wireless Communications10

11. Key Idea 10/12/2015CWC | Centre For Wireless Communications11

12. Key Idea 10/12/2015CWC | Centre For Wireless Communications12

13. System model 10/12/2015CWC | Centre For Wireless Communications13 : number of BSs : set of BSs Tx regioninterference region : number of BS antennas : number of data streams : set of data streams : set of data streams of BS : transmitter node of d.s. : receiver node of d.s. 1 2 3 1 2 3 4 5 6 7 8 9 1011 12

14. System model 10/12/2015CWC | Centre For Wireless Communications14 : information symbol;, : power signal vector transmitted by BS : beamforming vector;

15. System model 10/12/2015CWC | Centre For Wireless Communications15 signal received at : channel; to intra-cell interference out-of-cell interference : cir. symm. complex Gaussian noise; variance

16. System model 10/12/2015CWC | Centre For Wireless Communications16 received SINR of : out-of-cell interference power; th BS to intra-cell interference out-of-cell interference

17. System model 10/12/2015CWC | Centre For Wireless Communications17 1 2 3 5 6 7 8 out-of-cell interference power, e.g.,

18. System model 10/12/2015CWC | Centre For Wireless Communications18 1 2 3 5 6 7 8 out-of-cell interference power, e.g.,

19. System model 10/12/2015CWC | Centre For Wireless Communications19 1 2 3 5 6 7 8 out-of-cell interference power, e.g.,

20. System model 10/12/2015CWC | Centre For Wireless Communications20 1 2 3 5 6 7 8 out-of-cell interference power, e.g.,

21. System model 10/12/2015CWC | Centre For Wireless Communications21 1 2 3 5 6 7 8 out-of-cell interference power, e.g.,

22. System model 10/12/2015CWC | Centre For Wireless Communications22 received SINR of : out-of-cell interference power; th BS to intra-cell interference out-of-cell interference

23. System model 10/12/2015CWC | Centre For Wireless Communications23 received SINR of intra-cell interference out-of-cell interference : set of out-of-cell interfering BSs that interferes : out-of-cell interference power (complicating variables)

24. System model 10/12/2015CWC | Centre For Wireless Communications24 1 2 3 5 6 7 8 e.g.,

25. System model 10/12/2015CWC | Centre For Wireless Communications25 1 2 3 1 2 3 4 5 6 7 8 9 1011 12 : set of d.s. that are subject to out-of-cell interference e.g.,

26. System model 10/12/2015CWC | Centre For Wireless Communications26 1 2 3 1 2 3 4 5 6 7 8 9 1011 12 : set of d.s. that are subject to out-of-cell interference e.g.,

27. System model 10/12/2015CWC | Centre For Wireless Communications27 1 2 3 66 9 12 : set of d.s. that are subject to out-of-cell interference e.g.,

28. Problem formulation 10/12/2015CWC | Centre For Wireless Communications28 variables: and

29. Primal decomposition 10/12/2015CWC | Centre For Wireless Communications29 variables: subproblems (for all ) : master problem: variables:

30. Subproblem (BS optimization) 10/12/2015CWC | Centre For Wireless Communications30

31. Subproblem 10/12/2015CWC | Centre For Wireless Communications31 variables:  The problem above is NP-hard  Suboptimal methods, approximations

32. Subproblem: key idea 10/12/2015CWC | Centre For Wireless Communications32  The method is inspired from alternating convex optimization techniques  Fix beamforming directions  Approximate objective by an UB function  resultant problem is a GP; variables  Fix the resultant SINR values  Find beamforming directions, that can preserve the SINR values with a power margin  this can be cast as a SOCP  Iterate until a stopping criterion is satisfied

33. Subproblem: key idea 10/12/2015CWC | Centre For Wireless Communications33

34. Subproblem: key idea 10/12/2015CWC | Centre For Wireless Communications34

35. Subproblem: key idea 10/12/2015CWC | Centre For Wireless Communications35

36. Subproblem: key idea 10/12/2015CWC | Centre For Wireless Communications36

37. Subproblem: key idea 10/12/2015CWC | Centre For Wireless Communications37

38. Subproblem: key idea 10/12/2015CWC | Centre For Wireless Communications38

39. Subproblem: key idea 10/12/2015CWC | Centre For Wireless Communications39

40. Master problem 10/12/2015CWC | Centre For Wireless Communications40

41. Master problem 10/12/2015CWC | Centre For Wireless Communications41 variables:  Recall: subproblems are NP-hard -> we cannot even compute the master objective value  Suboptimal methods, approximations  Problem is nonconvex -> subgradient method alone fails

42. Master problem: key idea 10/12/2015CWC | Centre For Wireless Communications42  The method is inspired from sequential convex approximation (upper bound) techniques  subgradient method is adopted to solve the resulting convex problems

43. Integrate master problem & subproblem 10/12/2015CWC | Centre For Wireless Communications43  Increasingly important:  convex approximations mentioned above are such that we can always rely on the results of BS optimizations to compute a subgradient for the subgradient method.  thus, coordination of the BS optimizations

44. 10/12/2015CWC | Centre For Wireless Communications44 Integrate master problem & subproblem

45. 10/12/2015CWC | Centre For Wireless Communications45  out-of-cell interference:  objective value computed by BS at ‘ F ’ :  we can show that  optimal sensitivity values of GP -> construct subgradient F subproblem convex

46. 10/12/2015CWC | Centre For Wireless Communications46 Integrate master problem & subproblem subgradient method:

47. Integrate master problem & subproblem 10/12/2015CWC | Centre For Wireless Communications47 Algorithm 1 (subproblem) Algorithm 2 (overall problem)

48. An example signaling frame structure 10/12/2015CWC | Centre For Wireless Communications48 Algorithm 2 (overall problem) note:in Alg.1 or subgradient method, ‘BS optimization GP’ is always carried out in our simulations: - fixed Alg.1 iterations ( ) - fixed subgrad iterations ( ) per switch

49. Numerical Examples 12.10.2015CWC | Centre For Wireless Communications49 channel gains: : distance from to : small scale fading coefficients SNR operating point:

50. Numerical Examples 12.10.2015CWC | Centre For Wireless Communications50  -> the degree of BS coordination  note -> subgradient is not an ascent method  out-of-cell interference is resolved -> objective value is increased  smaller performs better compared to large  light backhaul signaling

51. 10/12/2015CWC | Centre For Wireless Communications51 Numerical Examples B  accuracy of the solution of an approximated master problem is irrelevant in the case of overall algorithm  refining the approximation more often is more beneficial

52. Numerical Examples 12.10.2015CWC | Centre For Wireless Communications52  same behavior  smaller performs better compared to large

53. Numerical Examples 12.10.2015CWC | Centre For Wireless Communications53  algorithm performs better than the centralized algorithm  not surprising since both algorithms are suboptimal algorithms

54. Numerical Examples 12.10.2015CWC | Centre For Wireless Communications54  algorithm performs better than the centralized algorithm  not surprising since both algorithms are suboptimal algorithms

55. Numerical Examples 12.10.2015CWC | Centre For Wireless Communications55  ; one subgradient iteration during BS coordination window  12% improvement within 5 BS coordination  99% of the centralized value within 5 BS coordination  WMMSE performs better with no errors in user estimates  WMMSE with even 1% error in signal covariance estimations at user perform poorly

56. Numerical Examples 12.10.2015CWC | Centre For Wireless Communications56  ; one subgradient iteration during BS coordination window  24% improvement within 5 BS coordination  94% of the centralized value within 5 BS coordination  WMMSE performs better with no errors in user estimates  WMMSE with even 1% error in signal covariance estimations at user perform poorly

57. Conclusions  Problem: WSRMax for MISO interfering BC channel (NP-hard)  Techniques: Primal decomposition  Result: many subproblems (one for each BS) coordinating to find a suboptimal solution of the original problem  Subproblem: alternating convex approximation techniques, GP, and SOCP  Master problem: sequential convex approximation techniques and subgradient method  Coordination: BS-BS (backhaul) signaling; favorable for practical implementation  Substantial improvements with a small number of BS coordination -> favorable for practical implementation  Numerical examples -> algorithm performance is significantly close to (suboptimal) centralized solution methods 12.10.2015CWC | Centre For Wireless Communications57