Stochastic Optimization for Variable Rate Applications with Time-Varying Statistics

We present a scheme to ensure Quality of Service for buffered variable rate applications. The scheme does not require knowledge of the channel distributions, indeed will track changing statistics, and is effectively spectrum optimal. The scheme employs stochastic approximation where the optimal solution is learned through updates of the Lagrange multiplier. The convergence speed and nearness to optimality are found, and the buffer stability probabilities are met. Analysis and simulations are provided to validate the scheme´s performance.