Wireless Sensor Networks (WSNs) are commonly deployed in dynamic environments where events, such as moving sensor nodes and changing external interference, impact the performance, or Quality of Service (QoS), of the network. QoS is expressed by the values of multiple, possibly conflicting, network quality metrics, such as network lifetime and maximum latency of communicating a packet to the sink. Sufficient QoS should be provided by the WSN to ensure that the end-user can successfully use the WSN to perform its application. We propose a distributed reconfiguration approach that actively maintains a sufficient level of QoS at runtime for a heterogeneous WSN in a dynamic environment. Every node uses a feedback control strategy to resolve any difference between the current and required QoS of the network by adapting controllable parameters of the protocol stack. Example parameters are the transmission power and maximum number of packet retransmissions. Nodes collaborate such that, with the combined adaptations, the required network QoS is achieved. The behavior of the reconfiguration approach and the tradeoffs involved are analyzed in detail. With the use of simulations and experiments with actual deployments, we show that our approach allows a better optimization of QoS objectives while constraints are met; for example, it achieves the same packet loss with a significantly longer lifetime, compared to current (re-)configuration approaches.

A Wireless Sensor Network (WSN) consists of a large collection of spatially dis- tributed autonomous devices with sensors to monitor physical or environmental conditions, such as air-pollution, temperature and traffic flow. By cooperatively processing and communicating information to central locations, appropriate ac- tions can be performed in response. WSNs perform a large variety of applications, such as the monitoring of elderly persons or conditions in a greenhouse. To correctly and efficiently perform a task, the behaviour of the WSN should be such that sufficient Quality-of-Service (QoS) is provided. QoS is defined by constraints and objectives on network quality metrics, such as a maximum end- to-end packet loss or minimum network lifetime. After defining the application we want the WSN to perform, many steps are involved in designing the WSN such that sufficient QoS is provided. First, a (heterogeneous) set of sensor nodes and protocols need to be selected. Furthermore, a suitable deployment has to be found and the network should be configured for its first use. This configuration involves setting all controllable parameters that influence its behaviour, such as selecting the neighbouring node(s) to communicate to and setting the transmission power of its radio, to ensure that the WSN provides the required QoS. Configuring the network is a complex task as the number of parameters and their possible values are large and trade-offs between multiple quality metrics exist. High transmission power may result in a low packet loss to a neighbouring node, but also in a high power consumption and low lifetime. Heterogeneity in the network causes the impact of parameters to be different between nodes, requiring parameters of nodes to be set individually. Moreover, a static configuration is typically not sufficient to make the most efficient trade-off between the quality metrics at all times in a dynamic environment. Run-time mechanisms are needed to maintain the required level of QoS under changing circumstances, such as changing external interference, mobility of nodes or fluctuating traffic load. This thesis deals with run-time reconfiguration of dynamic heterogeneous wire- less sensor networks to maintain a required QoS, given a deployed network with selected communication protocols and their controllable parameters. The main contribution of this thesis is an efficient QoS provisioning strategy. It consists of three parts: a re-active reconfiguration method, a generic distributed service to estimate network metrics and a pro-active reconfiguration method. In the re-active method, nodes collaboratively respond to discrepancies be- tween the current and required QoS. Nodes use feedback control which, at a given speed, adapts parameters of the node to continuously reduce any error between the locally estimated network QoS and QoS requirements. A dynamic predictive model is used and updated at run-time, to predict how different parameter adap- tations influence the QoS. Setting the speed of adaptation allows us to influence the trade-off between responsiveness and overhead of the approach, and to tune it to the characteristics of the application scenario. Simulations and experiments with an actual deployment show the successful integration in practical scenar- ios. Compared to existing configuration strategies, we are able to extend network lifetime significantly, while maintaining required packet delivery ratios. To solve the non-trivial problem of efficiently estimating network quality met- rics, we introduce a generic distributed service to distributively compute various network metrics. This service takes into account the possible presence of links with asymmetric quality that may vary over time, by repeated forwarding of informa- tion over multiple hops combined with explicit information validity management. The generic service is instantiated from the definition of a recursive local update function that converges to a fixed point representing the desired metric. We show the convergence and stability of various instantiations. Parameters can be set in accordance with the characteristics of the deployment and influence the trade-off between accuracy and overhead. Simulations and experiments show a significant increase in estimation accuracy, and efficiency of a protocol using the estimates, compared to today’s current approaches. This service is integrated in various protocol stacks providing different kinds of network metric estimates. The pro-active reconfiguration method reconfigures in response to predefined run-time detectable events that may cause the network QoS to change signifi- cantly. While the re-active method is generally applicable and independent of the application scenario, the, complementary, pro-active method exploits any a-priori knowledge of the application scenario to adapt more efficiently. A simple example is that as soon as a person with a body sensor node starts walking we know that several aspects, including the network topology, will change. To avoid degradation of network QoS, we pro-actively adapt parameters, in this case, for instance, the frequency of updating the set of neighbouring nodes, as soon as we observe that a person starts to walk. At design time, different modes of operation are selected to be distinguished at run-time. Analysis techniques, such as simulations, are used to determine a suitable configuration for each of these modes. At run time, the approach ensures that nodes can detect the mode in which they should operate. We describe the integration of the pro-active method for two practical monitoring applications. Simulations and experiments show the feasibility of an implementa- tion on resource constrained nodes. The pro-active reconfiguration allows for an efficient QoS provisioning in combination with the re-active approach.

Wireless sensor networks are typically operating in a dynamic context where events, such as moving sensor nodes and changing external interference, constantly impact the quality-of-service of the network. We present a distributed feedback control mechanism that actively balances multiple conflicting network-wide quality metrics, such as power consumption and end-to-end packet latency, for a heterogeneous wireless sensor network operating in a dynamic context. Nodes constantly decide if and how to adapt controllable parameters of the entire protocol stack, using sufficient information of the current network state. Using experiments with an actual deployment we show that our controller allows to maintain the required network-wide quality-of-service, with up to 30% less power consumed, compared to the most applicable (re-)configuration approaches.

The quality of the communication links in a Wireless Sensor Network often shows significant asymmetry and variation over time, due to, for example, heterogeneous settings of the transmission power, moving nodes or changing external interference. This makes it difficult for nodes to accurately maintain system-level properties, such as the minimum-energy path from the node to a given reference node, as required by many protocols. In this paper, we introduce a distributed service that allows nodes to maintain accurate information related to the minimum-cost path, such as its cost or parent on that path. Using controlled n-hop forwarding, to deal with asymmetric links, every node disseminates minimum-cost path and connectivity information allowing every connected node in the network to iteratively derive minimum-cost path information. This controlled n-hop forwarding is repeated to avoid stale information due to dynamic changes in link qualities. The parameters of the service allow a trade-off between the accuracy and overhead. We study the characteristics of a deployment that impact this trade-off and how the service should be parameterized accordingly. Extensive simulations and experiments for an actual deployment show a significant increase in the accuracy of the maintained minimum-cost path information, compared to the typically used local broadcasting approach.

Network dynamics, such as mobility and increase in network load, can influence the performance of a Wireless Sensor Network (WSN). In this paper, we introduce a method which exploits design-time knowledge of the application scenario dynamics to construct a proactive run-time reconfiguration approach. The approach anticipates for the impact that predefined dynamic events can have on the performance of the WSN by switching between various modes of operation defined at design-time. A mode defines the values for the controllable parameters of the network protocol stack. Our approach explicitly differentiates between parameters that can be adapted locally, per node, and those that should be considered globally for the whole WSN. Design-time definition of modes results in a very low run-time overhead as we only require detection of the mode to use and a low overhead synchronization to change global parameters. The approach is made robust by using a recovery approach for nodes unaware of their global mode after, for example, (re-)joining the network. Experiments with an office monitoring deployment and extensive simulations of a cow-health monitoring scenario show that our approach can easily be adopted by practical WSN deployments and results in a significant reduction in resource usage, e.g., power consumption in our examples, at a very low run-time overhead cost.

Within the Wirelessly Accessible Sensor Populations (WASP) project a research technology prototype has been build of an advanced Wireless Sensor Network (WSN) enabling Smart Dairy Farming. The current integrated WASP solution demonstrated in herd health control test bed can be well used as a research tool for animal and human behaviour scientists to enable data acquisition for periods up to several weeks to study behaviour and, in the end, come up with better models and treatments. Within the WASP test bed for Smart Dairy Farming, the focus has been on remote monitoring of activity-related problems like claw health and locomotion. Encouraged by the large-scale herd control deployment whose operation is limited to 10 to 12 days, we performed an exercise to assess whether it seems feasible to monitor a large number of cows over a period of a year without replacing the 8Ah batteries. Through this exercise, we wanted to get a feeling on application and deployment trade-offs like 1) acceptable time between packets generated at application level, 2) acceptable number of battery-powered mobile sensor nodes, 3) required number and positioning of battery- and/or mains-powered forwarder nodes and 4) required number and positioning of sink nodes, while considering application Quality of Service constraints in terms of size (and hence energy capacity) of batteries, packet delivery ratio and packet delivery latency. Keywords: wireless sensor network, test bed, herd control, locomotion, application

The inherently unreliable communication infrastructure compel WSN protocols to employ error control mechanisms. Traditionally, error control is achieved by a retransmission scheme using acknowledgment mechanisms. WSN architectures are severely resource constrained and the additional energy expense of transmitting error control messages can seriously degrade network lifetime. In this paper, we analyze performance of error control schemes for the case of point-to-multipoint communication. An explicit acknowledgment mechanism may provide for reliable communication, but has two major drawbacks: 1) the over head is significant for small data messages, and 2) in case of asymmetrical communication links, multi-hop dissemination of acknowledgments is required. As an alternative to such explicit acknowledgment schemes we propose the use of probabilistic acknowledgments. In this probabilistic scheme, a sender estimates the probability that a message has been successfully delivered, based on information about the quality of the radio channel. A message is then retransmitted until the probability of successful delivery reaches a defined threshold value. Network capacity available for error control can be distributed prudently among all information items to be disseminated, possibly taking into account different application requirements. We formulate are transmission control strategy which results in minimal latency and maximal message delivery ratio.

Currently, the guaranteed throughput of a stream processing application, mapped on a multi-processor system, can be computed with a conservative dataflow model, if only time division multiplex (TDM) schedulers are applied. A TDM scheduler is a budget scheduler. Budget schedulers can be characterized by two parameters: budget and replenishment interval. This paper introduces a priority-based budget scheduler (PBS), which is a budget scheduler that additionally associates a priority with every task. PBS improves the guaranteed minimum throughput of a stream processing application compared to TDM, given the same amount of resources. We construct a conservative dataflow model for a task scheduled by PBS. This dataflow model generalizes previous work, because it is valid for a sequence of execution times instead of one execution time per task which results in an improved accuracy of the model. Given this dataflow model, we can compute the guaranteed minimum throughput of the task graph that implements the stream processing application. Experiments confirm that a significantly higher guaranteed minimum throughput of the task graph can be obtained with PBS instead of TDM schedulers and that a conservative bound on the guaranteed throughput of the task graph can be computed with a dataflow model. Furthermore, our bound on the guaranteed throughput of the task graph is accurate, if the buffer capacities in the task graph do not affect the guaranteed throughput.