A Correlation-Aware Splitting Algorithm for Opportunistic Selection

Opportunistic selection is a key technique to improve the performance of wireless systems. In it, one among the available users is selected on the basis of their channel gains or local parameters, such as battery energy state. Formally, each user possesses a real-valued metric that only it knows, and the goal is to select the best user, which has the highest metric. The splitting algorithm is a popular, fast, and scalable algorithm to implement opportunistic selection; it is distributed and guarantees selection of the best user. We show that this algorithm, which has thus far been designed assuming that the metrics are independent and identically distributed, is no longer scalable when the metrics are correlated. We then propose a novel correlation-aware splitting algorithm (CASA) and show how it can be applied to practically motivated probability distributions and correlation models. We present computationally feasible techniques for pre-computing the thresholds that CASA specifies, thereby ensuring that CASA can be implemented in practice. We benchmark the performance of CASA with the conventional algorithm, and show that it reduces the average selection time significantly as the number of users or the correlation among them increases.