Computational complexity reduction in Taguchi method based joint optimization of antenna parameters in LTE-A networks : a thesis presented in partial fulfilment of the requirements for the degree of Master of Engineering in Telecommunication and Network Engineering
Long Term Evolution-Advanced (LTE-A) system is operated with cellular technology
based on frequency reuse. Due to the co-channel interference between cells, one cell‟s
performance is decided by not only its own configurations but also other cell‟s settings.
Therefore, joint optimization of antenna parameters in LTE-A cellular networks is the
key to maximizing coverage and capacity. This can be achieved by setting the antenna
parameters such as azimuth orientations and tilts to the optimal values. Nevertheless,
the large number of cell parameters and the interdependencies between these parameters
make it difficult and time-consuming to optimize a cellular network. In practice, the
joint setting of the parameters of all cells with irregular layout and coverage areas
becomes an important and challenging task.
There are several methods to search for the optimal settings of a cellular network. One
commonly used search method is Simulated Annealing (SA). SA can produce good
results in cellular network optimization, but it takes a long time and its performance can
easily be degraded if the input parameters are misconfigured. Other methods include the
trial-and-error approach that requires manual selection of parameter values and has no
guarantee for good results, and the brute-force approach that searches through all
possible combinations of parameter values and is thus computationally prohibitive.
Among the various algorithms proposed for this time-consuming optimization task, the
iterative approach based on the Taguchi method (TM) is a recent development that has
been shown to be promising. This thesis presents some further improvements to the
TM-based approach aiming at enhancing optimization performance and reducing
computational complexity. The proposed improvements include the use of the mixedlevel
Nearly-Orthogonal Array (NOA) to cater for the different optimization ranges of
different types of parameters, an improved mapping function to select testing values that
are more representative of the optimization range, and a hybrid approach using multiple
NOAs with decreasing number of experiments to exchange small degradation in
optimization performance for significant reduction in computational complexity. The
effectiveness of the proposed improvements is demonstrated by numerical examples.