Self learning call admission control for multimedia wireless DS-CDMA systems

The call admission problem for wireless CDMA systems is formulated as a semi-Markov decision process with constraints on the blocking probabilities and SIR (signal to interference ratio). We show that the optimal call admission policy can be computed via a stochastic gradient algorithm. Similar to neuro-dynamic programming algorithms, the algorithms proposed are s simulation based and do not require explicit knowledge of the underlying parameters such as transition probabilities (or equivalently invariant distributions). However, unlike Q-learning or temporal difference methods, the algorithms proposed here can straightforwardly handle constraints.