Reducing feedback for opportunistic scheduling in wireless systems

We study reducing feedback overhead of users' channel state information required for opportunistic scheduling at a base station while minimizing the throughput penalty incurred due to reduced feedback. We first propose a simple contention based scheme known as 'static splitting' for best effort traffic. The idea is to divide users into static groups, with users that belong to a group, and have their currently supported rate above a threshold, contending to send their current feedback to the base station. This is combined with maximum quantile scheduling -scheduling a user whose current rate is high relative to its distribution, to obtain thresholds that are independent of users' rate distributions even when they are heterogeneous, allowing off-line optimization of thresholds. Next we develop the insight that for a traffic mixture of best effort and real-time traffic one has to combine contention and polling to reduce feedback while providing quality of service. We propose a scheme based on this insight. Under this scheme we prove a lower bound on the service seen by a real-time when users' channel capacities are fast fading. Furthermore, we propose a heuristic modification that is able to exploit a larger fraction of opportunism. Simulation results illustrate the performance advantage of the proposed schemes.