Pseudo-Bayesian Broadcasting Algorithm for Opportunistic Splitting Scheduling Systems

Abstract

The opportunistic splitting algorithm (OSA) is known as an efficient algorithm that can exploit multiuser diversity in a distributed manner. However, the conventional OSA (C-OSA) is designed for the systems where users are always backlogged, i.e., always have packets to transmit. In this paper, to accommodate the time-varying nature of the backlog size of the practical systems, we first investigate an ideal OSA (I-OSA), which assumes that the backlog size is exactly known so that it can optimally control the request transmissions from the users to select the best user. I-OSA can give us a guideline on the upper-bound performance of OSA. Second, to realize the concept of I-OSA in practice, we further propose a pseudo-Bayesian broadcast algorithm for OSA (PBB-OSA), which recursively estimates the mean backlog size based on the outcomes of the channel contention among users, such as idle, collision, and success events, and controls the request transmissions. Numerical results show that compared to C-OSA, PBB-OSA needs fewer minislots to find the best user and shows a smaller scheduling outage probability. In addition, it is also shown that the performance of PBB-OSA is close to that of I-OSA.