Knowledge of global network state is crucial for several innovative network optimization techniques. Essentially, incorporating knowledge about the overall network state into locally made decisions at decentralized nodes might improve the overall network performance. A node might for instance perform transitions between network mechanisms that are optimized for certain network conditions. However, an individual node's scope of the network is limited in practice since it is able to overhear the wireless channel only locally, and explicit notification about global network state would result in large overhead. Therefore, we seek to extend a node's view into the network by means of machine learning techniques.


The goal of this thesis is to estimate global metrics of a mobile ad-hoc network (MANET) by means of locally overheard information in a network simulation environment.

  • Literature review: Identify network optimization techniques that rely on global network knowledge and extract their requirements.
  • Define metrics: Make a list of global network properties that should be classified or estimated.
  • Identification of features: Identify potential features that can be obtained by traffic monitoring. Features that comprise relevant information about distant nodes might for instance be obtained by inspecting packet headers of the higher layers (e.g., network layer and transport layer).
  • Feature engineering and machine learning: Select and engineer features that can be obtained by overhearing the wireless channel.
  • Implementation: Run experiments with the ns-3 network simulator and evaluate the estimator's performance.