1. Rate-Optimal Multiuser Scheduling with Reduced Feedback Vegard Hassel, Mohamed-Slim Alouini, Geir Øien and David Gesbert Gotland Workshop 24. - 26. August 2004

2. Agenda What is scheduling? Scheduling and adaptive modulation MCS, SMUD & ORRF scheduling algorithms ORRF: –Feedback load –MASSE –Impact of scheduling delay –Impact of outdated channel estimates

3. What Is Scheduling? How capacity is shared between users in a network TDM: A set of rules that specify which user is allowed to transmit and which user is allowed to receive at each time-slot

4. Scheduling and Adaptive Modulation To have efficient multiuser scheduling in wireless networks, we have to combine scheduling with adaptive modulation

5. BASE STATION EXTERNAL NETWORK USERS Feedback TRANSCEIVER SCHEDULER CHANNEL ESTIMATOR TRANSCEIVER CHANNEL ESTIMATOR TRANSCEIVER CHANNEL ESTIMATOR Data User 1 Data User 3 Data User 2 Data User 3 Data User 2 Data User 1 System Model

6. MCS Algorithm MCS=Maximum CNR Scheduling The time-slot is given to the user with the best CNR Rate-optimal algorithm (?) All users feed back channel state information for each time-slot This algorithm is only fair if the channels of the users are i. i. d.

7. SMUD Algorithm SMUD=Selective Multi-User Diversity The users feed back channel state information only if their CNRs are above a threshold The user with the highest CNR is chosen If none of the users succeed to exceed the threshold value, a random user is selected

8. ORRF Algorithm ORRF=Optimal Rate, Reduced Feedback The users feed back channel state information only if their CNRs are above a threshold If none of the users succeed to meet the threshold value, the scheduler requests full feedback The user with the highest CNR is always chosen!

9. SMUD CDF Expression for the CDF of the CNR of N users:

10. SMUD PDF Expression for the PDF of the highest CNR between all users using a Rayleigh channel:

11. ORRF PDF Expression for the PDF of the highest CNR between all users using a Rayleigh channel: Same as for the MCS algorithm!

14. Feedback Load The normalized feedback load for the ORRF algorithm can be minimized using an optimal threshold value:

15. Closed-Form Expressions for ORRF Maximum average system spectral efficiency: Shannon: AWGN & Rayleigh Adaptive M-QAM: –Constant-Power, Continuous-Rate –Constant-Power, Discrete-Rate –Continuous-Power, Continuous-Rate –Continuous-Power, Discrete-Rate

16. Closed-Form Expressions for ORRF and SMUD (II) Average BER: Adaptive M-QAM: –Constant-Power, Continuous-Rate –Constant-Power, Discrete-Rate –Continuous-Power, Continuous-Rate –Continuous-Power, Discrete-Rate

28. Scheduling Delay In real-life systems we have to assume a delay because the scheduler: Collects channel estimates Takes a scheduling decision Notifies the selected user Assumes that a perfect channel estimate is available at time t+ 

30. Outdated Channel Estimates A perfect channel estimate is NOT available at time t+  Both the scheduling decision and the decision of the constellation size have to be taken based on an estimate at time t