1. 2005/12/06OPLAB, Dept. of IM, NTU1 Optimizing the ARQ Performance in Downlink Packet Data Systems With Scheduling Haitao Zheng, Member, IEEE Harish Viswanathan, Senior Member, IEEE IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS VOL. 4, NO. 2, MARCH 2005 Presented by Li-Yi Lin

2. 2005/12/06OPLAB, Dept. of IM, NTU2 Outline Introduction Optimizing the mapping between SINR and MCS Optimizing the scheduler ranking for HARQ Simulation and discussion

3. 2005/12/06OPLAB, Dept. of IM, NTU3 Introduction Two main considerations of designing systems which provide high-speed packet data service on the downlink - the selection of MCS based on the channel quality of the link - the selection of the user to whom a particular slot is assigned Adaptive techniques - Dynamic Link Adaptation of Adaptive Modulation and coding - Automatic Repeat reQuest (ARQ) or Hybrid ARQ (HARQ) - Scheduler

4. 2005/12/06OPLAB, Dept. of IM, NTU4 Dynamic Link Adaptation of Adaptive Modulation and coding Link adaptation continuously adjusts the modulation and coding scheme (MCS) The transmitter selects an appropriate MCS, based on the user’s channel quality feedback The performance of link adaptation largely depends on the accuracy of channel quality measurement.

5. 2005/12/06OPLAB, Dept. of IM, NTU5 Automatic Repeat reQuest (ARQ) or Hybrid ARQ Packet data is delay-tolerant - feasible to use retransmission schemes to recover erroneous packets. HARQ can compensate for link adaptation errors and provide a finer granularity of coding rate HARQ - simple ARQ - chase combining - incremental redundancy

6. 2005/12/06OPLAB, Dept. of IM, NTU6 Simple ARQ - simple ARQ based transmitter retransmits the same packet and repeats the procedure until the packet is received successfully. Chase combining - the base station repeatedly sends the same packet and the receiver aggregates the energy from the (re)transmissions to improve signal to noise ratio (SNR) Incremental redundancy (IR) - transmits additional redundant information in each retransmission and gradually refines coding rate and SNR till a successful decoding is achieved.

7. 2005/12/06OPLAB, Dept. of IM, NTU7 Scheduler Scheduler can take advantage of channel variations by giving certain priority to the users with transitorily better channel conditions. The choice of the scheduling algorithm critically impacts the system performance.

8. 2005/12/06OPLAB, Dept. of IM, NTU8 Introduction (cont ’ ) Aggressive MCS The conventional mapping design fails to take into account the performance improvement by HARQ. The instantaneous rate does not fully represent the true data throughput.

9. 2005/12/06OPLAB, Dept. of IM, NTU9 Outline Introduction Optimizing the mapping between SINR and MCS Optimizing the scheduler ranking for HARQ Simulation and discussion

10. 2005/12/06OPLAB, Dept. of IM, NTU10 Optimizing the mapping between SINR and MCS Assume that the channel quality feedback carries SINR. The mapping is between the SINR and the MCS. For both Chase combining and simple ARQ, the MCS used in retransmissions is the same as that in the original transmission. Assume the channel condition stays constant during retransmissions as that of the initial transmission.

11. 2005/12/06OPLAB, Dept. of IM, NTU11 Notation

12. 2005/12/06OPLAB, Dept. of IM, NTU12 The traditional mapping selection criteria

13. 2005/12/06OPLAB, Dept. of IM, NTU13 A single unified mapping criterion - The average throughput of an AWGN channel using MCS i ： the average number of successfully received information bits ： the average time taken by the packet

14. 2005/12/06OPLAB, Dept. of IM, NTU14 TRPT-Chase ：

15. 2005/12/06OPLAB, Dept. of IM, NTU15 (6) TRPT-SARQ ：

16. 2005/12/06OPLAB, Dept. of IM, NTU16

17. 2005/12/06OPLAB, Dept. of IM, NTU17 Outline Introduction Optimizing the mapping between SINR and MCS Optimizing the scheduler ranking for HARQ Simulation and discussion

18. 2005/12/06OPLAB, Dept. of IM, NTU18 Optimizing the scheduler ranking for HARQ Two important characteristics - Frame Error Rate Information - Retransmission Information The scheduler design and the mapping selection can be conducted jointly to optimize the system performance. But for simplicity, we assume the scheduler design and the mapping selection are performed independently. Replacing the instantaneous rate with effective rate.

19. 2005/12/06OPLAB, Dept. of IM, NTU19 Ranking A: Instantaneous Rate Ranking B: ARQ Success Probability Weighted Instantaneous rate,

20. 2005/12/06OPLAB, Dept. of IM, NTU20 Ranking C: Average Packet Throughput Based Effective Throughput

21. 2005/12/06OPLAB, Dept. of IM, NTU21 Ranking C:

22. 2005/12/06OPLAB, Dept. of IM, NTU22 Ranking C:

23. 2005/12/06OPLAB, Dept. of IM, NTU23 Ranking D: Success Probability Weighted Instantaneous Rate Ranking E: Approximated Average Packet Throughput Based Effective Throughput

24. 2005/12/06OPLAB, Dept. of IM, NTU24 Outline Introduction Optimizing the mapping between SINR and MCS Optimizing the scheduler ranking for HARQ Simulation and discussion

25. 2005/12/06OPLAB, Dept. of IM, NTU25 The simulated radio network - Opnet network simulation tool - a radio network controller (RNC) - a base station (Node B) - mobile terminals - total networking delay = 50 ms - time multiplexing - each scheduling interval of frame lasts 2 ms - apply built-in module to IP, TCP, UDP, HTTP, FTP Simulation and discussion

26. 2005/12/06OPLAB, Dept. of IM, NTU26 - Radio Link Protocol (RLP) * performs data block segmentation and reassembly * The RLP PDU size is chosen to be 40 bytes - * performs scheduling, MCS selection and HARQ functionality * HARQ operates in terms of three Stop And Wait process * the Maximum number of retransmissions is 4 * MAC scheduler makes scheduling decision about 1 ms prior to the actual transmission - PHY * assume the uplink channel operates at 64 kbps and 0% FER * For the downlink channel, the frame error is generated by relating the SINR at each mobile terminal to a link level performance curve * Physical mobility of the user has not been considered * 5-MHz spectrum and 2-ms frame

27. 2005/12/06OPLAB, Dept. of IM, NTU27

28. 2005/12/06OPLAB, Dept. of IM, NTU28 4MCS set - 640kbps 、 1.28Mbps 、 1.92Mbps 、 2.56Mbps 6MCS set - 320kbps 、 480kbps 、 640kbps 、 1.28Mbps 、 1.92Mbps 、 2.56Mbps

29. 2005/12/06OPLAB, Dept. of IM, NTU29

30. 2005/12/06OPLAB, Dept. of IM, NTU30 AGG criterion ： 3 dB more aggressive than TRPT-chase

31. 2005/12/06OPLAB, Dept. of IM, NTU31

32. 2005/12/06OPLAB, Dept. of IM, NTU32 Three scenarios Scenario A: Single User With TCP Scenario B: Multiple Users With TCP and Modified Scheduler Rankings Scenario C: Multiple Users With TCP Using Ranking C for Various Mapping Criterion

33. 2005/12/06OPLAB, Dept. of IM, NTU33 Scenario A: Single User With TCP Performance measurement ：

34. 2005/12/06OPLAB, Dept. of IM, NTU34 Scenario A: Single User With TCP

35. 2005/12/06OPLAB, Dept. of IM, NTU35 Scenario B: Multiple Users With TCP and Modified Scheduler Rankings Fairness measurement

36. 2005/12/06OPLAB, Dept. of IM, NTU36 Fig. 7. Multiple-user HTTP performance of various scheduler rankings using Chase combining and TRPT-chase mapping criterion, 6MCS set. (a) 3 km/h and (b) 30 km/h. Scenario B: Multiple Users With TCP and Modified Scheduler Rankings

37. 2005/12/06OPLAB, Dept. of IM, NTU37 Fig. 8. Multiple-user HTTP performance of various scheduler rankings using Chase combining and AGG mapping criterion, 6MCS set. (a) 3 km/h and (b) 30 km/h. Scenario B: Multiple Users With TCP and Modified Scheduler Rankings

38. 2005/12/06OPLAB, Dept. of IM, NTU38 Scenario B: Multiple Users With TCP and Modified Scheduler Rankings

39. 2005/12/06OPLAB, Dept. of IM, NTU39 Fig. 10. User HTTP performance of various mapping criterion using Chase combining, proportional fair scheduler with ranking C, 4MCS set, 3 km/h. (a) System metric and (b) user throughput. Scenario C: Multiple Users With TCP Using Ranking C for Various Mapping Criterion

40. 2005/12/06OPLAB, Dept. of IM, NTU40 Summary & future work The proposed mapping design achieves 5%-50% throughput improvement. The modified proportional fair scheduler achieves 10%-30% performance improvement. The sensitivity of the performance to the mapping depends on the granularity of the MCS set with decreasing sensitivity for larger MCS sets. The channel estimation and prediction error can also be included in the mapping design by modifying the frame error rate accordingly

41. 2005/12/06OPLAB, Dept. of IM, NTU41 Thank You!