1 / 15

Scheduling and Optimization

Scheduling and Optimization. Criteria and Algorithms for Scheduling of Packet Data over Wireless Channels. Nilo Casimiro Ericsson, Signals & Systems, Uppsala University. Outline. Introduction, background Scheduling for spectral efficiency Latest scheduling insights

Download Presentation

Scheduling and Optimization

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Scheduling and Optimization Criteria and Algorithms for Scheduling of Packet Data over Wireless Channels Nilo Casimiro Ericsson, Signals & Systems, Uppsala University

  2. Outline • Introduction, background • Scheduling for spectral efficiency • Latest scheduling insights • No need for complex optimization • Provide an average throughput • Adaptive criteria – simulation results • Conclusion

  3. Packet data over fading channels Avoid fading dips!

  4. Scheduling of OFDM bins • Perform scheduling based on predicted average SNR in time-frequency bins •  • For each bin let the “best” user transmit; use adaptive modulation and ARQ 1 4 3 5 2 user freq time

  5. What the scheduler does:

  6. Scheduling algorithms • Simple “linear” maximization • Best First • Maximum Allocation • Robin Hood • “Exact” buffer-matching • Controlled Steepest Descent • Exhaustive search

  7. Complexity (25 bins) two-step one-step + swap one-step

  8. But, is the criterion right at all? • Buffer content minimimization at each scheduling instant seems short-sighted • Search algorithms allocate resources to match buffer content as exactly as possible • Sum-of-squares criteria • Uncertain predictions… • ”Academic” interest, off course • Instead: Maintain a (constant?) average (over time) throughput for each active stream • Based on maximized “linear” criteria • If necessary: re-allocations from over-provisioned streams

  9. Traffic adaptive criteria • Previously in Robin Hood (Coarse adaptivity) • Three features compared in some order: • Modulation, Priority, SNR • If two have equal Modulation => compare Priority, etc… • Can change order to (adaptation to traffic situation)Priority, Modulation, SNR • New: Quantize features into (e.g.) Modulation 3 bits m1,m2,m3 Priority 2 bits p1,p2 SNR 2 bits s1,s2 (explain!) • The new feature: m1,m2,m3,p1,p2,s1,s2 • But also: m1,p1,p2,m2,m3,s1,s2

  10. Adaptive criteria example • 3 bits for Modulation (0-7) • 2 bits for Priority (0-3) • 0 bits for SNR (omitted) • mmmpp • mppmm • User 1: • M = 6 (64QAM), mmm = 1102 • P = 1 (medium low), pp = 012 • A) mmmpp = 110012= 2510 • B) mppmm = 101012 = 1910 • User 2: • M = 5 (32QAM), mmm = 1012 • P = 2 (medium high),pp = 102 • A) mmmpp = 101102= 2210 • B) mppmm = 110012 = 2510 > <

  11. Simulation of scheduler • 25 OFDM bins per schedule • 5 MHz carrier @ 1900 MHz • Time-frequency bin size: 0.667 ms x 200 kHz • 108 payload symbols per bin • 12 users • 8 modulation levels (3 bits) • 0-7 (“quiet”-128QAM) • SNR thresholds: [ 6.5 10 14 18 22.5 26 30 ] dB • (why not 1-8?) • 4 priority levels (2 bits) • 0-3 • Random SNR for each user and bin • 100 schedule simulations per criteria setup

  12. 12 users, 4 priorities: 3 users of each priority Same SNR distribution for all: N(10,10) Maximum modulation: 7 (128QAM) Simulation 1: Throughput per user Criteria:mmmpp Criteria:mmppm Criteria:mppmm Criteria:ppmmm Criteria:mmpmp Criteria:mpmpm Criteria:pmpmm Criteria:mpmmp Criteria:pmmpm Criteria:pmmmp (A) (B) Total throughput

  13. 12 users, 4 priorities: 3 users of each priority 4 different SNR distributions: N({15,12,9,6},5) Highest priority for worst SNR Simulation 2: Throughput per user Criteria:mmmpp Criteria:mmppm Criteria:mppmm Criteria:ppmmm Criteria:mmpmp Criteria:mpmpm Criteria:pmpmm Criteria:mpmmp Criteria:pmmpm Criteria:pmmmp (A) (B) Total throughput

  14. Conclusion • For practical scheduler: abandon complex search algorithms • Too many uncertainties (channel prediction, buffer usage) • Scheduling can handle also distant users with worse conditions than near users • Work with “priorities” • Upgrade the importance of “priority” • Probably, average throughput target will also help distant users • Over-provisioned near users will give resources to under-provisioned distant users

  15. 12 users, 4 priorities: 3 users of each priority 3 different SNR distributions: N({5,10,15},5) Maximum modulation: 7 (128QAM) Simulation 3: Throughput per user Criteria:mmmpp Criteria:mmppm Criteria:mppmm Criteria:ppmmm Criteria:mmpmp Criteria:mpmpm Criteria:pmpmm Criteria:mpmmp Criteria:pmmpm Criteria:pmmmp (A) (B) Total throughput

More Related