Joint throughput and coverage optimization under sparse system knowledge in LTE-A networks

Self-organizing networks (SONs) are widely considered as a major facilitator in improving the end users' quality of experience. Developing well-performing and reliable algorithms is still an ongoing research topic. A majority of contributions propose SON algorithms that optimize a desired metric by the use of detailed input data, such as maps of received signal strength for every tilt and cell, or a map of traffic demand. However, such advanced knowledge about the network might not be available for technical or financial reasons. In this work, we focus on the creation of SON algorithms under sparse system knowledge, meaning that advanced input data (mentioned above) is not available. By introducing cost functions for every key performance indicator to be balanced, we enable direct search approaches from literature to be applicable to the problem at hand. Using the Nelder-Mead, a simultaneous perturbation stochastic approximation, and a taxi cab method, we can jointly optimize user throughput and coverage in a heterogeneous traffic demand scenario. The results obtained suggest that sparse knowledge SON algorithms can reliably increase network performance while having very low requirements for operation.