Editor’s Note: Welcome to our weekly Reality Check column. We’ve gathered a group of visionaries and veterans in the mobile industry to give their insights into the marketplace.
Self-optimizing network applications for heterogeneous networks address coverage and capacity optimization, energy savings and interference coordination, as well as handset management enabling the best customer experience as possible. SON also aids operators with performance management for the network. As networks evolve into hetnets, the multi-layered functionalities managed by SON software provide over-sight of the network performance automatically. This benefit maintains the best network performance for an operator’s customers, which depends on multiple and interrelated network and environmental parameters that are difficult to manage manually.
Implemented SON techniques today are very basic, but interesting approaches are currently being widely studied by the telecommunication scientific community. SON solutions are based on machine learning techniques, such as: genetic programming, evolutionary methodology inspired by natural selection found in biological evolution to automatically generate computer programs to perform a user-defined task; reinforcement learning, how an agent ought to take actions in an environment so as to maximize some notion of cumulative reward; stochastic approximation, attempts to find the solution or to minimize functions through sequential iterations in noisy processes.
The key idea behind these techniques is the ability to learn and develop responses from processed data collected. We first need to determine the complexity and dynamics in the wireless network, so that the proper data is collected and made available for the learning approach implemented. The learning process can be online solutions found through direct interactions with the environment, or offline solutions developed for a given situation through statistical models. Once an action is formulated, it is applied to the wireless network.
Given the dynamic nature of wireless networks, online approaches appear more suitable for the majority of SON applications. Online learning approaches allow the SON software to stay continually engaged in monitoring network performance data and modify parameters in a reactive manner, after a learning period. Online learning also allows the network to adapt to any adverse environmental situations as they occur. In most cases, an online learning approach is an adaptive system that changes its structure during the learning phase. As an example, Q-learning, which is the ability to compare the expected utility of the available actions without requiring a model of the environment, and stochastic approximation methods have been studied to solve multiple problems in hetnet systems.
However, these techniques present some problems in terms of scalability. Therefore, the most plausible solution seems to be hybrid methods, which are more flexible, adaptable and robust learning approaches. The combination of evolutionary algorithms or fuzzy logic with reinforcement learning techniques are two examples of hybrid solutions. Evolutionary algorithms often perform well approximating solutions to all types of problems because they ideally do not make any assumption about the underlying wireless environment. Fuzzy logic also performs well because it is based on degrees of truth of the considered network parameters.
Self-organization applications are limited only by our imagination and of course by the laws of physics. There are plenty of opportunities for research in this interesting field. We should search for those scalable, agile and stable approaches, which will fit the decentralized and variable characteristics of the hetnets.
Dr. Ana Galindo-Serrano is a Researcher at Orange Labs, Paris, the research and development center of France Telecom-Orange, worldwide operator. Galindo-Serrano can be reached at [email protected].
Eric Moore is the COO/CTO of Axis Teknologies, a wireless infrastructure managed and professional services firm based in Sandy Springs, Ga. Moore can be reached at [email protected].
