Dynamic User Demand Driven Online Network Selection

• Published in: IEEE communications letters Vol. 18; no. 3; pp. 419 - 422
• Main Authors: Du, Zhiyong, Wu, Qihui, Yang, Panlong
• Format: Journal Article
• Language: English
• Published: New York, NY IEEE 01.03.2014; Institute of Electrical and Electronics Engineers; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Applied sciences; Convergence; Delays; Equipments and installations; Exact sciences and technology; Heuristic algorithms; Mobile radiocommunication systems; network handoff cost; Network selection; online learning; Operation, maintenance, reliability; Quality of service; Radiocommunications; Systems, networks and services of telecommunications; Telecommunications; Telecommunications and information theory; Throughput; user demand; Wireless communications; Wireless networks; Performance evaluation; Mobile radiocommunication; E-learning; Network selection; Wireless telecommunication; network handoff cost; Algorithm; Time variation; Remote teaching; Convergence speed; Selection problem; Simulation; On line processing; user demand; User requirement; Convergence rate; Handover; Wireless network; Service quality; online learning
• ISSN: 1089-7798; 1558-2558
• DOI: 10.1109/LCOMM.2014.011214.132617

Network selection plays a key role in reaping the potential benefit of heterogeneous wireless networks. Aiming at improving the user's quality of experience, we study the network selection problem with time-varying user demand and non-uniform network handoff costs in a dynamic environment. One appealing solution in converging the time-vary user demand and the diverse network performance is dynamic network selection, which, however, poses the dilemma between satisfying user demand and controlling the network handoff cost. To get around this problem, we propose an online network selection algorithm to learn the optimal network selection policy with network handoff cost consideration. In addition, we exploit the inherent dependency in the problem and derive another two algorithms with much faster convergence speed. Simulations reveal that the proposed algorithms can achieve 10%~ performance gain over existing methods.