Third generation wireless networks and beyond will solicit the cooperation of heterogeneous access networks, so as to provide multimedia traffic to different classes of users, with varying quality requisites over regions and time zones. We address the problem of how to partition the traffic demand efficiently onto the underlying radio access networks. The design objective is a resource allocation strategy, which provides a maximal resource utilization across all access networks. At the same time, the allocation should respect quality levels related to handover dropping performance; these levels can be predefined per service and per region. We propose a solution based on reinforcement learning, which runs independently at each of the cells of every access system, and report results. In the case where network revenue does not depend solely on resource utilization, but on parameters such as the type of service and/or the service duration, the method is readily extensible to include these factors.
[1]
Timothy X. Brown,et al.
Adaptive call admission control under quality of service constraints: a reinforcement learning solution
,
2000,
IEEE Journal on Selected Areas in Communications.
[2]
Krishna M. Sivalingam,et al.
Dynamic resource allocation schemes during handoff for mobile multimedia wireless networks
,
1999,
IEEE J. Sel. Areas Commun..
[3]
Bruce Hajek,et al.
On Load Balancing in Erlang Networks
,
1996
.
[4]
Dimitri P. Bertsekas,et al.
Reinforcement Learning for Dynamic Channel Allocation in Cellular Telephone Systems
,
1996,
NIPS.