Fast Finite Frame Length IRSA Optimization Based on Bayesian Optimization

Abstract

Simplified packet reception models such as collision model are usually adopted for optimizing irregular repetition slotted ALOHA (IRSA), which may lead to performance penalty due to its poor accuracy. In this letter, we consider the problem of online IRSA optimization in the finite frame length regime using a more accurate reception model. In order to allow a fast optimization of IRSA, we present a new method based on Bayesian optimization with Gaussian processes. Our method finds the optimal user degree distribution minimizing packet loss rate (PLR) in an iterative way. At each iteration, a surrogate is built to model the unknown IRSA PLR performance function using Gaussian process (GP) regression, and then an acquisition function defined from the surrogate is utilized to choose the next degree distribution, whose PLR is evaluated by the more accurate model. The proposed method is able to infer the PLR of an untested degree distribution, thus converging quickly within only tens of iterations. Simulation results show that IRSA schemes optimized using our method can achieve lower PLR compared with those optimized based on the collision model.