1. 1 An Analytical Model for the Dimensioning of a GPRS/EDGE Network with a Capacity Constraint on a Group of Cells r02922008, r02922133, r02944039 Nogueira, Georges, Bruno Baynat, and Pierre Eisenmann ACM 2005

2. Outline  Background  Single Cell System  Multiple Cell System  Model Validation  Performance Results  Examples  Conclusion 2

3. Background  2G - GSM system (Global System for Mobile Communications)  2.5G system  GPRS(General Packet Radio Service)  EDGE (Enhanced Data rates for Global Evolution) 3

4. Circuit Switched vs. Packet Switched Circuit Switched Packet Switched

5. GPRS – Evolve from GSM  Timeslot is the basic unit for sending packet  Provide fast reservation  Four channel coding schemes (CS1 、 CS2 、 CS3 、 CS4)  Hardware  Add PCU(packet control unit) in BSC  Add SGSN, GGSN for sending packet 5

6. 6

7. EDGE  also known as Enhanced GPRS  EDGE uses higher-order PSK/8 phase shift keying (8PSK) 7

8. Single Cell System  Some Assumption and parameters  All mobiles have the same reception capability. they are “(d+u)”  tB : the system elementary time interval  xB : the number of data bytes transferred during tB over one time-slot 8

9. Single Cell System  For GPRS  For EDGE 9

10. Single Cell System  tbfmax : the maximum number of mobiles that can simultaneously have an active downlink TBF( Temporary Block Flow )  ON periods correspond to the download of an element  Size is characterized by a discrete random variable Xon  an average value of xon bytes  OFF periods correspond to the reading time  modeled as a continuous random variable Toff  average value of toff seconds 10

11. Markovian Analysis  ON/OFF distributions : memoryless (assumed)  Linear discrete-time Markov chain. 11

12. Markovian Analysis  12

13. Markovian Analysis 13

14. Markovian Analysis 14

15. Markovian Analysis  All average performance parameters of a single cell can be expressed as function of  dimensionless parameter x  cell capacity T  mobiles capacity d  numbers of mobiles in the cell N 15

16. Multiple Cell System  16

17. 17 Multiple Cell System

18. 18 Multiple Cell System

19. 19 Multiple Cell System

20. 20 Model Validation Validate the analytical model by comparison with simulation results. OPNET

21. 21 Identical cells P cells are identical in terms of available radio resources and offered traffic. All the mobiles generate the same traffic.

22. 22 @ P

23. 23 Different cells The characteristics of all cells in terms of offered traffic and radio conditions are randomly generated. All the mobiles generate the same traffic. Typically, we could represent a cell with a majority of business users, having a specific call profile.

24. 24 @ P

25. 25 Performance Results Assume that all the cells are identical. Similar studies can be performed on heterogeneous cells systems with no additional complexity.

26. 26 @ P@ M max Pr↓ Q↑Q↑ U↑U↑X↓X↓ ↑

27. 27 Pr↑ ↑ Q↓Q↓ U↓U↓X↑X↑ @ T@ P

28. 28 ↑ Q(-) U(-)X(-) Pr↑ @ T@ N

29. 29 Performance Graphs Can be instantaneously obtained with our analytical solution. They allow to directly derive any performance parameter knowing the traffic load profile (N, x).

30. 30 @ N @ x U

31. 31 @ N @ x X

32. 32 @ N @ x Pr

33. 33 Assume here that the network dimensioning is based on a maximum acceptable blocking rate of 2% for data transfer requests. In GPRS or EDGE, a transfer request rejection results in 5 seconds idle time before a subsequent request is allowed. For this target blocking rate, we want to find the values of: P max : the maximum number of cells N max : the maximum number of GPRS mobiles that can be admitted in each cells Example: Maximum blocking probability

34. 34 @ N @ x P max, with Pr ≦ 2%

35. 35 @ P @ x N max, with Pr ≦ 2%

36. 36 A typical 1 time-slot threshold is chosen, i.e. a mobile that starts downloading a page has the guarantee to obtain at least 1 time-slot per TDMA for the entire transfer duration. For this target blocking rate, we want to find the values of: P min : the minimum number of cells N max : the maximum number of GPRS mobiles that can be admitted in each cells Example: Minimum normalized throughput

37. 37 @ N @ x P min, with X ≧ 1TS

38. 38 @ P @ x N max, with X ≧ 1TS

39. Conclusion & future work Provide computational efficiency and accuracy for performance and dimensioning analyses Intend to extend this work and methodology to UMTS and HSDPA modeling. 39