Optimization of data transmission over a stochastic channel with partially observed state

Abstract

The optimization problem for data transmission over a stochastic channel is considered. The data transmission model is described by the triple “queue–channel–observations”. The queueing system is fed by a non-stationary Poisson stream of data packets for further transmission over a communication channel governed by a non-homogeneous Markov chain. The system is considered with a single server, finite bu er, and service rate proportional to the transmission rate with channel-dependent factor. The observations are formed by round-trip times of sent packets and described by the Markov counting process with channel-dependent intensity. The transmission rate is to be optimized within the class of feedback controls given two performance characteristics: the average number of lost packets and the mean level of energy consumption. The approach proposed for control optimization is based on the optimal filter equations, the complete-information control algorithm, and Monte Carlo techniques. The paper presents both theoretical and simulation results.