Contribution to dynamic resource management in future wireless networks by means of reinforcement learning
2010
Resource optimization in future radio access networks is a topic that constantly opens new challenges due to the technological improvements, development of new services, growth of competition from manufacturers, operators and service provider side, etc. In this sense the reconfiguration and cognition in future networks has become a hot research topic in recent years. The intelligence in network management offers the capability to dynamically adjust resource usage to actual network needs and preferences at any time. This is present in fixed mobile backbones that, on the one side, tend to unification (common IP backbone) to enable easier growth and management of heterogeneous networks, and, on the other side, tend to efficient resource usage that is economically practical. In the radio spectrum, similar trends are followed to achieve efficient spectrum usage. However, the development of new technologies that offer higher throughput in radio access also needs to consider previous market penetration of existing technologies, which leads to heterogeneous access networks where diverse technologies coexist. Finally, with the introduction of the opportunistic radio access, certain recycling of the spectral resources is also envisaged for the optimum spectral resource exploitation. This means that the spectrum that is not being used at a given time by primary users can be released to enable its use by a secondary market, which should additionally improve the spectral efficiency.
This thesis focuses on several aspects in resource management of the future networks. Several cognitive elements are developed in the thesis to provide dynamic tracking and control of networks under varying environment conditions. For each specific case a reinforcement learning algorithm is developed to achieve the desired resource control. In particular a first focus is on the optimization in all-IP backbone network elements with heterogeneous traffic. On the one side, the thesis presents a dynamic active queue management algorithm, on the other side, the thesis presents the possibility to dynamically balance heterogeneous traffic over multiple routes in such scenarios. In both cases autonomous solutions will work only based on the domain edge experience while providing overall adaptation. In the radio access part of the network, similar approaches are considered for the joint radio resource management in heterogeneous LTE and UMTS scenarios. The coexistence of the radio access technologies that can offer the same services introduces flexible radio access technology assignation for various scenarios. Additionally, frequency bands that are not necessary can be released for the secondary markets to increase overall spectral efficiency. Thus, the thesis presents one example for dynamic spectrum management in a UMTS network that can reduce spectrum occupancy for eventual secondary markets. Finally, from the cognitive radio network perspective, a solution for reliable cooperation between nodes is presented. Cooperation can be the key factor to achieve performance improvements in opportunistic networks. However, the selfishness and malicious behaviour by certain cooperating nodes in such networks can present a serious obstacle in these scenarios. In this context, this thesis presents a solution based on reinforcement learning that offers autonomous cooperator selection to increase reliability under highly unreliable conditions.
All the previous aspects are studied in dynamic scenarios under varying environmental conditions. The algorithm solutions developed for the given tasks rely on reinforcement learning mechanisms that are easily implemented. The good properties of the proposed solutions are discussed along the thesis in different scenarios and compared with some classical baseline solutions in each corresponding case.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
0
References
0
Citations
NaN
KQI