Autonomic Joint Session Scheduling Strategies for Heterogeneous Wireless Networks
Yuan Xue, Yuewei Lin, Zhiyong Feng, Huying Cai, Cheng Chi · 2008
In order to optimize usage of radio resource for heterogeneous radio access technologies (RATs) and jointly designed from the user perspective, the joint session scheduling (JOSCH) mechanism has been introduced to split traffic over tightly coupled radio network. This paper presents distributed reinforcement learning (RL) as an autonomic approach for the JOSCH. Through the "trial-and-error" interaction with its radio environment, the JOSCH agent learns to split the traffic in a best way and allocate sub-streams in the proper RATs. A backpropagation neural network is adopted to generalize the large input state space of the RL algorithm to reduce memory requirement. Extensive simulations show that the proposed algorithm not only realizes the autonomy of JOSCH through the online learning process, but also improves the service quality at user side and the spectrum utility at operator side base on the suitable strategies.