Network Optimization and Analysis
Analyzing and understanding increasingly complex networks has been a significant challenge for today’s society. Some of these networks, such as the network of genetic interactions in organisms, have always been there but are only recently a mayor subject of research. On the other hand, communication- and energy transmission networks are man-made. Even though they have been established decades ago, progressively higher performance requirements, density, and heterogeneity necessitate advanced methods for network planning, operation scheduling, and ongoing optimization.
Our research capitalizes on mathematical tools and methodologies to gain deeper insight about the interactions of entities in these networks. Based on statistics, optimization and signal processing, we then aim to develop advanced modeling tools, approaches for efficient network planning and scheduling, and methods that increase the in-operation performance.
Radio Resource Management and Optimization
Spatial reuse and cell splitting in wireless cellular networks are effective mechanisms for increasing the system's capacity. They are foreseen to be essential for accommodating the ever-increasing volume of mobile data in 4G and 5G cellular networks. However, the gains achieved by deploying more and closer base stations (BSs) might be counteracted by the increased interference among them, when operating on the same frequency (or time) resources. As an example, in heterogeneous cellular networks (HetNets), considered of paramount relevance to boost the capacity of cellular networks, the significant differences between the high-power macro base stations (BSs) and low-power BSs (micro, pico, and femto BSs) represent a challenging scenario in terms of load balancing and interference management.One of the popular techniques discussed in the literature to reduce the interference involves the introduction of so-called almost blank subframes, in which some of the frequency (or time) resource blocks (RBs) of the interfering BSs are muted. Nevertheless, the interference is the combined effect of user association, RB allocation, and RB muting. Therefore, only the joint optimization of all these procedures can yield the optimal throughput in the network.
In the communication system group, we consider different techniques and tools such as convex optimization and integer linear programming, to jointly optimize user association, RB allocation, and RB muting to maximize the network's throughput while taking the users' fairness into account. We also develop low-complexity algorithms and investigate the trade-off between their complexities and performances. System Level Simulations are performed based on standardized methodologies which allows us to evaluate the proposed algorithms under more practical conditions such as different network load and realistic Channel State Information (CSI).
Projects and Collaborations:
5G Mobile Communication Networks: Planning, Scheduling and Optimization
With the fourth generation of mobile communication networks, represented by the standard LTE, reaching maturity and widespread deployment, the fifth generation (5G) is a topic of current research. These 5G networks have to address the demands of a connected society. Private consumers demand high data rates and high mobility, for example for video streaming. For other applications, such as virtual reality, very broad application areas ranging from education and training to industrial control devices are envisioned. In order to operate these devices in a connected and coordinated fashion, the underlying mobile communication network has to fulfill high requirements for its reliability, security, latency and flexibility.
In order to meet these requirements, multiple measures to increase the overall network performance are currently under investigation. Promising technology candidates include millimeter-wave communications, Massive MIMO and densely deployed small cell networks. The common aspect of these technologies is an increased heterogeneity of the network, because the amount of cell types, used frequency bands and different services will increase over time. This increased heterogeneity requires sophisticated approaches for the planning, scheduling and optimization of a 5G network.Our research focuses on finding network optimization approaches for 5G networks that improve network performance without causing much additional coordination- and computation overhead. We focus on the deployment of small cells in heterogeneous networks, specifically their effect on overall cell loads and customer quality-of-service. We also investigate the possibility to make 5G networks 'service-centered' by design, with varying transmission and network scheduling methods being dynamically applied depending on varying service requirements.
Smart Microgrid Control: Risk-Limited and Economic Dispatch
The way in which electric power is generated and electrical grids are operated has experienced a significant paradigm shift in recent decades. Renewable energy resources have been integrated into the grid to increase its overall environmental compliance. Due to the volatile and mostly non-controllable nature of these resources, such as wind and solar energy, additional challenges for the reliability and economic operation of the power system arise. For example, additional uncertainties are caused by the need for intermediate storage, consumer participation in the market, and clustered operation in potentially standalone microgrids. Addressing these challenges with advanced technologies in network monitoring, communications and machine learning is an important subject of current research.
Smart microgrids are proposed as a promising way to increase the grid flexibility and reliability by decentralizing the energy generation. Meanwhile, smart microgrids bring about additional advantages, such as reducing carbon emissions and improving energy efficiency by incorporating renewable sources, utilizing waste heat of generators, and decreasing transmission distances. These microgrids may operate in standalone mode or connected to the main grid. The capability of standalone operation is essential for parts of the power grid. The reasons for this are manifold, such as reliability requirements or geographical circumstances on islands and in remote locations. We aim to develop control strategies that enable an economic and risk-free operation of microgrids, while causing a limited amount of coordination- and computation overhead. Additionally, we want to preserve privacy and security of customers by favoring indirect (stochastic) approaches to load scheduling rather than direct control of individual appliances.
Learning Genetic Interactions from Noisy Knockout Experiments
Genetic interaction analysis aims at uncovering the interactions among a set of genes with respect to a specified cellfunction of a biological system, e.g., bacteria. Reliable knowledge about the interactions among a set of genes is of greatinterest since it allows for the development of new drugs against various diseases as well as for the design of customizedbacteria which are genetically modified in order to maximize the production of a desired substance for medicine andindustry. In recent years, biology as a whole has underwent strong improvements based on technological advances inmicro arrays and the development of the synthetic-genetic-array (SGA) technologies. Large scale knockout experimentson bacteria and other simple organisms, where a plenty of single genes or different sets of genes are functionally switchedoff and the phenotype of the cell function under study is measured, have become practicable. The interactions amongthe genes under study are well described by a directed-acyclic-graph (DAG) and can be detected based on data generatedby single-knockout (SK) and double-knockout (DK) experiments.Since genetic interactions can only be observed indirectly via SK and DK experiments learning the DAG best matching thedata is a non-trivial task due to strong noise effects disturbing the experiment setup. At NTS we are working on graphlearning algorithms based on linear integer programming to enhance the reliability of the learned DAGs and to reducethe computational complexity of the DAG-learning process.
Projects and Collaborations:
- European Molecular Biology Laboratory, Heidelberg