Neural Approximation of Resource Allocation Functions in Multiservice Access Multiplexers

摘要: In an access node to a multiservice network (e.g., base station in integrated services cellular wireless the hub double hop satellite system, or Optical Line Terminal (OLT) broadband Passive Network (PON)), output link bandwidth is adaptively assigned different users, and dynamically shared between guaranteed traffic asynchronous types. The allocation effected by admission controller, whose goal minimize refusal rate of reservation requests as well loss probability cells queued finite buffer. Optimal control strategies are approximated means backpropagation feedforward neural networks, acting on embedded Markov chain dynamics; networks operate conjunction with higher level which performs stochastic optimization algorithm. case unknown, slowly varying input rates explicitly considered. Numerical results presented that evaluate approximation ability adapt parameter variations.