Soft Real-Time Communication in Challenging Environments - Wireless Ad Hoc Communications for Post-Disaster Networks with Low Traffic Load and No Mobility

Communication networks are considered play a central role in many aspects of life. However, the infrastructure is fragile and can be significantly damaged by natural disasters, as events have shown in the past. In times of crisis, a functioning communication infrastructure is essential. Immediate post-disaster measures, such as coordinating rescue operations, assessing the situation, and placing emergency calls, all rely on communication. Moreover, communication networks remain essential throughout the recovery from a disaster. Post-disaster networks must support a wide range of applications quickly and efficiently. They typically need to be deployed very quickly in crisis areas, relying on a variety of wireless technologies connecting end-devices in an ad hoc manner. Several applications depend on real-time communication, such as emergency calls or the control of rescue and monitoring robots. However, existing approaches for real-time communication often place high demands on networks and are therefore unsuitable for the challenging conditions of crisis scenarios. In this thesis, a decentralized approach is presented to provide soft real-time guarantees for data transmission in wireless ad hoc networks. First, an analysis of individual links is conducted to understand how the distributions of delays for individual links are formed. An extensive measurement study of a common network protocol designed for microcontrollers is conducted. Analyzing the data collected in these experiments enables the development of a model that accurately reconstructs observed distributions of link delay with a limited parameter set. These findings about link delays form the basis of a probabilistic network model that allows to predict the distribution of delays of a communication path. This enables performance predictions about the of real-time data across the entire network. Building on this model, a distributed algorithm is introduced to calculate real-time traffic performance across the network. Using this performance data, a routing algorithm is developed that achieves provably optimal routing for real-time data. To evaluate these algorithms, a network simulator for wireless sensor networks was adapted, allowing the modeling of a wide range of realistic scenarios, including sudden traffic bursts and diverse channel conditions. A routing protocol integrating the developed algorithms is then designed and implemented. This protocol can be understood as a probabilistic variant of a distance-vector-based protocol. In all simulated scenarios, the approach outperforms classical, non-probabilistic distance-vector protocols. The performance differences are especially pronounced in challenging scenarios such as high network load, uncertain channel conditions, or very tight packet deadlines. The contributions of this dissertation thus demonstrate the viability of probabilistic approaches for providing soft real-time guarantees in challenging network environments.