CAC for Multimedia Services in Mobile Cellular Networks ： A Markov Decision Approach

1. CAC for Multimedia Services in Mobile Cellular Networks：A Markov Decision Approach Speaker：Xu Jia-Hao Advisor：Ke Kai-Wei Date：2004 / 11 / 18

2. Outline • Introduction • System Model Description • SMDP Approach in Our CAC • Numerical Results • Conclusion

3. Outline • Introduction • System Model Description • SMDP Approach in Our CAC • Numerical Results • Conclusion

4. Introduction • There is a growing interest in deploying multimedia services in mobile cellular networks. • Call Admission Control (CAC) is a key factor in Quality of Service (QoS) provisioning for these services. • We model a one-dimensional cellular network and describe how to find out optimal admission decisions.

5. Problems • For mobile multimedia services, the existing MCN (mobile cellular network) for voice-oriented services, needs to be adapted in numerous aspects. • The connection-level QoS in MCNs is usually expressed in terms of call blocking probability and call dropping probability (handoff). • Multimedia calls belong to multiple and different types of class => multiclass calls

6. Typical CAC policies -- Coordinate-Convex policy • Complete Sharing ( CS )： - Every class share the bandwidth pool. • Complete Partitioning ( CP )： - Bandwidth for each class is exclusively reserved. • Threshold： - A newly arriving call is blocked if the number of calls is >= a predefined threshold.

7. Another Solution • The coordinate-convex policy boasts of easy tractability. But in certain cases, it turns out strictly suboptimal. • CAC using semi-Markov Decision Process (SMDP) can maximize the revenue for multi-class networks. • We can use linear programming (LP) formulation to find out optimal decisions.

8. Outline • Introduction • System Model Description • SMDP Approach in Our CAC • Numerical Results • Conclusion

9. Our System Model • The cellular system under consideration is one-dimensional, which is deployed in streets and highways. • Our system consists of N cells and we consider a general model of multiclass calls with mobility characteristics.

10. Notation • 　：Call requests of class-i in cell-n, a Possion distribution with mean arrival rate. • 　：The call holding time of a class-i call is assumed to follow an exponential distribution with mean. • 　：The number of channels required to accommodate the call of class-i. • 　：The rate of class-i call that handoff to our system from outside. (n = 1 or N) • 　：For each on-going class-i call, revenue rate.

11. Notation ( cont. ) • The cell residence time (CRT), independent of class： - The amount of time that an MT (mobile terminal) stays in a cell before handoff, is assumed to follow an exponential distribution with mean (the parameter represent the handoff rate). • The rate that a call in a given cell will handoff to one of its adjacent cells is . • The total bandwidth in each cell is the same and denoted by C, assuming a fixed channel allocation.

12. Notation ( cont. ) • The current state of our cellular system： denotes the number of class-i calls in cell-n • All possible states： For each state x, a CAC policy should find out an ”accept / reject” decision for all kinds of traffic.

13. Traffic Model in Our Cellular System

14. Outline • Introduction • System Model Description • SMDP Approach in Our CAC • Numerical Results • Conclusion

15. SMDP Introduction • The original SMDP model consider a dynamic system which, at random points in time, is observed and classified into one of several possible states. • After observing the state, a decision has to be made and the corresponding revenue for each state is gained.

16. SMDP in Here • For each state x, a set of actions is available. • This controlled dynamic system is called an SMDP when the following Markovian properties are satisfied： If at a decision epoch the action a is chosen in state x, then the time until, and the state at, the next decision epoch depends only on the present state x.

17. Linear programming ( LP ) • It has an advantage that additional constraints can be easily incorporated. • It can guarantee the upper bound of the handoff dropping probability. • We use it to solve the SMDP-formulated CAC problem in our cellular system, which aims at both maximum revenue and QoS guarantee.

18. LP in MATLAB • ”linprog” function

19. LP Example

20. SMDP Description • The decision epoch：s = ( x , e ) , • The action space B：

21. SMDP Description ( cont. ) • The action space is actually a state dependent subset of B： • The expected time until a new state is entered：

22. SMDP Description ( cont. ) : • Transition probability： • The total revenue rate for the cell：

23. LP Formulation • The LP associated with SMDP： ：the long-run fraction of decision epochs at which the system is in state x and action a is taken

24. Optional Constraint • We also need to consider the QoS requirements: - the upper bound of the handoff dropping probability. • Let denote the maximum tolerable handoff dropping probability of a class-i call. - external handoff from outside and internal handoff between cells in our system.

25. Optional Constraint ( cont. ) • From outside： • Internal：

26. Outline • Introduction • System Model Description • SMDP Approach in Our CAC • Numerical Results • Conclusion

27. Simulation • Simulate one-cell model (N = 1) and two-cell model (N = 2). • Compare our SMDP CAC with the upper limit (UL) CAC policy that has a threshold for a class-i call originating in a cell. ( threshold [2,1] ) • C = 5 ; K = 2 ; (b1,b2) = (1,2) ; (D1,D2) = (0.02,0.04)

28. Utilization vs. Erlang Load (N=1)

29. Utilization vs. Erlang Load (N=2)

30. Handoff Dropping Probability fromthe outside vs. Erlang Load (N = 1)

31. Handoff Dropping Probability fromoutside vs. Erlang Load (N = 2)

32. Handoff Dropping Probability betweenCells vs. Erlang Load (N = 2)

33. Revenue Ratio vs. Erlang Load

34. Outline • Introduction • System Model Description • SMDP Approach in Our CAC • Numerical Results • Conclusion

35. Conclusion • Optimal CAC is essential for the efficient utilization of scarce radio bandwidth. • By using SMDP, we can maximize the revenue while satisfying the QoS requirements.

36. Reference • Call Admission Control for Multimedia Services in Mobile Cellular Networks: A Markov Decision Approach--Jihyuk Choi; Taekyoung Kwon; Yanghee Choi; Naghshineh, M.;Computers and Communications, 2000. Proceedings. ISCC 2000. Fifth IEEE Symposium on , 3-6 July 2000 • Keith W. Ross and Danny H. K. Tsang, “Optimal Circuit Access Policies in an ISDN Environment: A Markov Decision Approach,” IEEE Transactions on Communications, • Subir K. Biswas and Bhaskar Sengupta, “Call Admissibility for Multirate Traffic in Wireless ATM Networks,” INFOCOM '97. Sixteenth Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings IEEE , Volume: 2 , 7-11 April 1997 Pages:649 - 657 vol.2