An integrated simulation environment for sensor data fusion applications in wireless mesh networks

For the performance evaluation involving wireless mesh networks the concept of simulation is often used instead of measurements within a real environment. This is partly due to the lack of real-world testbeds but also caused by the unsteady nature of wireless mesh networks making it difficult to reproduce results. Typically, no real data is transmitted in simulations as for the calculation of network-related metrics like average throughput, delay, or loss rates, the knowledge of the packet sizes is sufficient. In order to evaluate the user level performance of a specific application that is subject to certain network conditions it is necessary, though, to use a set of application-related metrics. For the calculation of these metrics, the processing of the transmitted data in the time and order it has been received under the simulated network conditions is required. In the military domain, an important application is the timely transmission and processing of sensor data like GPS positioning information or vital signs data of deployed units in order to maximize situational awareness. In these tactical scenarios, wireless mesh networks are commonly used for the transmission of the sensor data from mobile units to the command point. In this paper, we present an integrated simulation environment based on ns-2 for sensor data fusion applications in wireless mesh networks. The simulation environment allows for the injection of real sensor data into the simulator and the subsequent execution of the data fusion application using the data in the shape it has been received during the simulation. Thus, application- related metrics that quantify the quality of the fusion result can be calculated and be opposed to conventional network-related metrics. This enables the analysis of the direct impact of network behavior, node mobility, or transport and routing protocols on the fusion result.

I. Introduction

The use of simulation for the performance evaluation of networks and their components like protocols or medium access policies is a very common concept. The advantages of simulation over measurements in a real testbed are the easier practicability, the possibility of reproduction, and independence from existing hardware or prototypes. On the other hand, simulations are based on models of the reality and are, therefore, immanently imprecise due to abstractions. For this reason, the convergence to the true-to-life behavior has to be carefully considered when using simulation.