Description
Date depot: 1 janvier 1900
Titre: Optimisation automatique de la gestion de ressources radio dans la radio cellulaire de 4ième génération
Encadrante :
Berna SAYRAC (Orange Labs)
Directeur de thèse:
Eric MOULINES (CMAP)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
The wireless ecosystem is becoming more and more heterogeneous with co-existing and co-operating
technologies and deployment scenarios (i.e. macro, micro, pico and femto cell structures). In this context,
efficient management of future Beyond 3G (B3G) networks is a major challenge for network operators, both
in terms of cost reduction and performance enhancement, as well as profitability and user satisfaction. Selfmanaging
(or self-x) principles have long been considered as a means to achieve efficient network
management, covering a wide range of applications from internet routing to interference cancellation in
Radio Access Networks (RANs). The main idea is to provide autonomy to interconnected intelligent entities
of the network that manage themselves without direct human intervention. These intelligent entities need to
be environment-aware through individual or collective observations/measurements; using statistical/machine
learning principles and optimization techniques to attain pre-defined global objectives defined by the
operator. At the same time, attention must also be paid to the coherence of the requirements of the used
techniques with the existing state of the technology and market [1].
The combination of simple statistical/machine learning techniques with optimization methods has been
shown to provide effective and tangible solutions to automated self-healing in B3G networks [2]. The main
idea is constructing a statistical model of the functional relationships/dependencies between Key
Performance Indicators (KPIs) and Radio Resource Management (RRM) parameters through regression;
and using this model in a constrained optimization framework for direct derivation of the optimum value of
the RRM parameter in an iterative fashion. At each iteration, two consequential tasks are achieved: 1-the
statistical model is updated with the observation data at hand and its precision is improved, 2-using the
statistical model, a new value of RRM parameter value is proposed that attains the pre-defined objectives of
the constrained optimization. This new RRM parameter is in turn used in conjunction with the existing
observation data for statistical model improvement. Application of the proposed approach to automated
healing provides an off-line implementation: the temporal scale of iterations is long (in the order of hours
and/or days) so that observation data is composed of hourly and/or daily measurements supplied by a
system simulator. The results obtained with simulation data demonstrate substantial performance
enhancements and fast convergence on automated healing of intra-system mobility and Inter-Cell
Interference Coordination (ICIC) mechanisms in Long Term Evolution (LTE) RANs [3]-[7].
The combined statistical learning/optimization approach described above is very generic and can be applied
to the construction of any control loop that governs the network functionalities (interference
cancellation/mitigation, congestion control, admission control, scheduling, mobility, RF parameter
optimization, cell planning/deployment etc.) and on any temporal scale (from milliseconds to days). The
preliminary studies with simulation data and the field trial constitute a proof-of-concept of the combined
statistical learning and optimization approach as a self-x methodology in B3G RANs. The approach being
much more general and applicable in a large range of temporal scales, further applications can be foreseen
on other B3G RAN functionalities operating in a variety of temporal scales, such as:
intra- and inter-system load balancing
admission control,
transmission power optimization (power masks in LTE),
scheduling,
cell planning, etc.
Doctorant.e: Khan Yasir