Design and analysis of opportunistic fair downlink schedulers
Knightly, Edward W.
Doctor of Philosophy thesis
Scarce spectrum has been the impeding force for satisfying users' exploding appetite for anywhere, anytime connectivity. Efficient utilization of the scarce resource while providing high quality of service thus is the main challenge for protocol design over wireless networks. Unfortunately, traditional wireline designs, by assuming constant quality channels, often fail in satisfying these objectives when facing lossy and time varying wireless channels. This thesis develops a class of intelligent media access control algorithms that exploits the adaptation capability of underlying physical layer. Particularly, we design opportunistic schedulers to enable the potential gain on channel diversity among multiple users while satisfying provable weighted fairness. Our key technique in designing these schedulers is to jointly exploit the temporal variations in the resource consumption of multiple users to opportunistically select those with greater throughput potential, while also ensuring that fairness constraints are satisfied. First, we design and analyze Wireless Credit-based Fair Queueing (WCFQ), an opportunistic wireless scheduler targeting at multiple access schemes that are based on single-user-at-a-time. We show that WCFQ satisfies provable long- and short-term probabilistic fairness guarantee while significantly improving system throughput. Second, we extend our system model to exploit concurrent scheduling where multiple users can access the system simultaneously. We design MFS-D and MFS-P, a Multichannel Fair Scheduler with Deterministic and Probabilistic fairness constraints respectively. Finally, we perform a comparison between single and multiple channel scheduling. Particularly, we investigate the fundamental capacity of the system given fairness constraints and limitations from underlying physical layer. Through theoretical and numerical investigations we show that multiple channel scheduling may have significant gain over single channel scheduling in certain scenarios.
Electronics; Electrical engineering; Computer science