Contribución a la planificación sistémica de redes móviles 4G

Resumen

Long Term Evolution (LTE) is the fastest developing mobile technology and one of the strongest driving forces of Mobile Telecommunication market. In turn, it is one of the most relevant and dynamic players in our globalized economy, and helps accelerate innovation in other markets. LTE offers unprecedented very high data rate and very low latency for a number of different applications and services, specially in downlink (DL). This is one of the reasons why LTE subscriptions have grown at a high rate during Q1 2016, with 150 million new subscriptions during this quarter, reaching a total of 1.2 billion worldwide. However, although huge traffic increases are good news for mobile operators, the negative counterpart is that revenues do not rise at the same rate. This leads to an increasing gap between traffic (which evolves exponentiallike in time) and revenues (which grow at much slower rate than traffic because of the competition among operators). In the effort of achieving reasonable profit margins, operators aim at optimizing investments while preserving user’s quality of service (QoS), via a twofold strategy: (1) selecting the best available technology, and (2) optimizing the deployment of network equipment. Specifically, the optimization of the Radio Access Network (RAN) –the set of base stations (BSs) or “evolved Node B” (eNB) in LTE– is crucial since it takes about 60% of total investment, and even a higher part of operational expenditures (OPEX). Within this context, the purpose of this thesis consists in implementing novel algorithms and tools aiming at improving the dimensioning of LTE access networks. One the one hand, this thesis tackles the problem of users assignment to eNBs in LTE mobile networks. We propose a model that aims at assigning NU users to NB eNBs by minimizing a function called Download Time of the complete System (DTS), defined as the minimum time required for all the users in the system to complete their downloads. This strategy helps the network use its resources in a more balanced way, by assigning users from cells –which otherwise would be overloaded when using a simple Channel Quality Indicator (CQI)-based approach– to others with less load. The user-cell association problem is, in general, a combinatorial hard NP problem. Although other approaches indirectly tackle this complexity by transforming the problem into another whose optimal solution is less difficult (but still with very high computational cost), we tackle directly our DTS minimization problem by finding an approximate solution provided by a specific implementation of an Evolutionary Algorithm (EA). Its most interesting aspect is the way in which the candidate solutions (user-cell assignments) are coded. The chromosome is a vector of dimension NU in which each element represents a user. The element in position j contains some information about the user uj . That information is an integer representing which eNB of the available NU has been associated to that user. The mutation, crossover, and selection operators are designed to work with this encoding. The crossover operator, in particular, is a tournament of all against all. The other novel aspect of the implementation of the proposed evolution algorithm is found in the initial population. As we have information about a suboptimal solution of the problem (that provided by the conventional method based on CQI –which assigns a user to the eNB for which it has the best CQI–), this is included in the initial population, and the rest of the individuals are generated, basically, by applying the mutation and crossover operators on that solution. In any case, the solution found (association of each user to an eNB) is better (less DTS) than the assignment made with conventional methods. On the other hand, the complete dimensioning tool, which includes the aforementioned EA for the user-cell association problem, exhibits (1) a simple and efficient parameterization of the multiple input parameters and the initial location of the eNBs, which (2) allows simulating a multi-service and multi-user environment, using (3) different user-eNB association methods, and (4) several scheduling algorithms, so that (5) the Minimum Download Rate required for each service is guaranteed. To fulfill the requirements (1)-(5), the tool computes the average speed of the services offered, taking into account the download times of each of the users. If, when using the number of eNB previously computed, the rate demanded by the different simulated services is fulfilled, then the value is assumed to be valid. Otherwise, a new eNB is added iteratively until the above requirement is fulfilled. None of the available tools is able to meet these requirements. A variety of experiments in different realistic scenarios points out that the proposed method outperforms the conventional CQI-based and Load-Balancing approaches, specially in urban and dense urban environments (where the assignment is more critical in terms of capacity). We have also tested our approach in simple scenarios that mimic Heterogeneous Networks and Ultra-Dense Networks in which the number of users is similar to that of nodes, proving its beneficial properties. Our method is aimed at the design/dimensioning stage and not to the operation stage, and, in this sense, it is an off-line algorithm that could be useful to make better decisions on cost-effective network dimensioning and upgrading.