Presentation on theme: "Multiuser Diversity Gain Enhancement by Guard Time Reduction Hend Koubaa, Vegard Hassel, Geir E. Øien Norwegian University of Science and Technology (NTNU)"— Presentation transcript:
Multiuser Diversity Gain Enhancement by Guard Time Reduction Hend Koubaa, Vegard Hassel, Geir E. Øien Norwegian University of Science and Technology (NTNU) Dept. of Electronics and Telecommunications Trondheim, Norway
Multiuser diversity scheduling Multiple users with different channel fluctuation characteristics Exploiting this diversity among users by selecting the best user to transmit with the best rate Increasing the Maximum Average System Spectral Efficiency (MASSE) Basic mechanism The sender requests the CNRs of users The users send back their CNRs: feedback mechanism The sender chooses the best CNR user and transmits with the rate corresponding to this CNR
Performance criteria Maximising network throughput Maximising the potential rate to be used Reducing the cost of the feedback mechanism Minimising the number of users feedbacks (saving users powers) Reducing the guard time (overhead time between BS queries) Trade off between 1 and 2! Realising fairness among users Fairness: Deeper study in a future work
Using multiple feedback thresholds Opportunistic scheduling algorithm to be presented at VTC Spring 2005 (Hassel, Alouini, Gesbert, and Øien). Employing multiple (nested) optimized feedback CNR thresholds to minimize average feedback load. Denote the thresholds by th,L>th,L1> · · ·>th,0 (For convenience th,L = and th,0 = 0) The base station initially requests feedback from those users whose CNRs are above th,L1. If there are none, the threshold is successively lowered, to th,L2, th,L3, · · ·, th,0. The best user is thus always selected, but the average feedback load is significantly reduced compared to the conventional rate-optimal MCS (Max Channel SNR) scheduling algorithm.
Contention-less feedback algorithm (1) Assume N users. 1. Check if any CNRs are above th,L1. All users wait for a mini-slot T MS. If there are none, the threshold is successively lowered to th,L2 - and users again wait for their mini-slot T MS. And so on... 2. In the interval [th,l,th,l+1], all users having a CNR above th,l will send their feedback - all others stay silent. If only one user is above the threshold, there is no collision problem. If multiple users have a CNR above th,l: collision problem. Solution: ranking the users The base station sends a ranked user list The time is initially slotted to N ranked mini-slots T MS The user of rank j will start sending his feedback at the start of the j th minislot, if his CNR is above th,l. Those users having a CNR lower than th,l will simply be silent during their assigned mini-slots.
Contention-less feedback algorithm (2) Assume T FB is the total time needed to send one users feedback (and assume T FBT MS ). [th,l, th,l+1] denotes the best user interval. Assume that all users can detect feedback initiated from others, and stay silent if feedback from another user is detected. After each feedback transmission from one user (of rank j), the time is again slotted, now to N-j mini-slots. This mechanism is repeated until the user of rank N sends his feedback - or is silent - during the N th (last) mini-slot. The base station finally schedules the best user out of those who gave feedback within the current interval.
Contention-less feedback algorithm (3) Example 4 thresholds (th,3>th,2> th,1>th,0 ) 3 users The CNRs of the best users, u 1 and u 3, are assumed to reside in the interval [th,1,th,2] First step [th,2,th,3] T ms u 1 sends his feedback u1u1 The remaining time is slotted to 2 mini-slots u3u3 u2u2 u 2 is silent and u 3 sends his feedback u3u3 u2u2 th,0=0th,1th,2 th,3= Second step [th,1,th,2]: collision T ms u 1 & u 3 Third step slotting u1u1 u3u3 u2u2 T FB T ms
Contention feedback algorithm N users If only one user exists in the best user interval, then this user will send his feedback just after the base station's query, and data transmission can start. However if multiple users have a CNR above th,l: all send simultaneously, and a collision takes place. Solution: exponential backoff scheme Each user having a CNR above th,l and detecting a collision (by not receiving data) will retransmit his feedback with probability q < 1. After a collision have occurred i times, each user will send his feedback with probability q i This mechanism will last until one successful feedback transmission takes place.
Comparison between both proposed algorithms Contention-less The best user is selected (highest rate but also highest guard time) Contention A random user in the best user interval is selected (suboptimal rate but also lower guard time)
Results: (general and Rayleigh fading model) Expressions for the guard time (GT) are derived for both approaches (see the TD). The MASSE in the contentionless approach (Rayleigh): The MASSE in the contention approach (Rayleigh): NB: The above MASSEs both have to be multiplied by the fraction where T TS is total time between base station queries.
MASSE for both approaches (5 users) Masse with 5 users Masse (bits/sec/Hz) Number of thresholds
MASSE for both approaches (10 users) Masse with 10 users Masse (bits/sec/Hz) Number of thresholds
MASSE for both approaches (long data packet) Masse with 10 users Masse (bits/sec/Hz) Number of thresholds
MASSE for both approaches (short data packet) Masse with 10 users Masse (bits/sec/Hz) Number of thresholds
Guard time in both approaches Contention- less feedback Contention channel for feedback Number of thresholds Guard time (T MS ) Guard time with 5 users
MASSE for both approaches (short data packet case: T TS = 20T MS ) Contention- less feedback Contention channel for feedback Number of thresholds Masse (bits/sec/Hz) Masse with 5 users
Conclusion (1) Study of guard time and MASSE in a certain multi-user diversity scheduling algorithm with two different feedback protocols. Use of multiple thresholds is better than the full feedback mechanism (=no thresholds) w.r.t. overall MASSE. The impact of the feedback mechanism has to be studied in detail. Preliminary conclusion: To maximize overall MASSE, choose the number of feedback thresholds such that the overall guard time is: not too long if the data packet size is short not too short if the date packet size is long
Conclusion (2) Two proposed feedback mechanisms Contention-less: ranked user list Contention: exponential backoff scheme The contention-less feedback protocol is more suitable for long data packet applications (optimal rate counts for more than short guard time) The contention mechanism is more suitable for short data packet applications (shorter guard time compensates for suboptimal rate)