Efficient Distributed Downlink Scheduling for Wireless Cellular Networks: A POMDP Framework

Abstract

We propose a decentralized downlink scheduling protocol for wireless cellular networks that allows each base station to control its transmission power. On the beginning of each scheduling period, neighbor base stations exchange their local system states, i.e., channel, traffic and position statistics of their local mobile users. But on each slot of the current scheduling period, the base stations may not be able to receive full information from neighbor cells. Therefore, to enable the base stations to make optimal decisions, for user scheduling and transmission power control, we develop an analytical optimization framework, based on the theory of Partially Observable Markov Decision Process. To help a base station, this decision-theoretic approach uses the history of observations about neighbor base stations system statistics and their scheduling policies. Simulations show that our approach is more efficient than existing centralized scheduling approaches which due to the high complexity they do not allow base stations to communicate with each other.We have conducted our experiments, using a two-tier 19-cell network. We have compared the results of our contribution with the techniques proposed in [1,2], referred to as Baseline 1, 2 respectively. These approaches define a set of patterns for on-off BSs. These on/off patterns are designed such that a BS scheduled to be on can expect the interference mitigation from one of its neighbors. In order to mitigate high overhead of information exchange and complexity, Baselines 1, 2 drop the