Optimal control of variable rate coding with incomplete observation in integrated voice/data packet networks

This paper investigates the design of dynamic voice coding rate control with doubly stochastic packet arrival processes to reduce traffic congestion in an integrated voice/data packet network. Incomplete observation is assumed. A control maximizes the average coding rate of voice packets subject to an average voice/data packet waiting time constraint on a finite horizon with incomplete observation. The optimal control has a simple structure and is a randomization of two linear feedbacks with input saturation. The general separation principle for the optimal control with incomplete observation, that is, the sufficiency of the posterior distribution for the optimal control, is derived. Since the complexity of the analytical results prevents us from obtaining an explicit control when observation is incomplete, an adaptive suboptimal control is proposed. This is also a random modification of two linear feedbacks with saturation and is easily implementable. The effectiveness of the suboptimal control is verified by simulations.