## WD62 (HPC-based Sensor Analytics) Task 5: Modeling and Simulation of Wireless Sensor Networks

**Agency:** United States Army Tank Automotive and Armaments Command

**Researcher:** Medal, H.R. (PI)

**Amount:** $254,384

**Abstract: **

**Agency:** United States Army Tank Automotive and Armaments Command

**Researcher:** Medal, H.R. (PI)

**Amount:** $254,384

**Abstract: **

We propose to develop new tools for modeling the performance of sensor networks under a variety of environmental conditions (e.g., dust, sand, etc.). Our tools will model the following performance metrics for a given network: vulnerability, reliability, and power consumption / lifetime. Modeling these metrics will provide engineers with a tool for virtual prototyping that can quickly evaluate the performance of candidate network designs and new technologies. In addition, having models for all of these performance metrics will allow engineers to trade off these metrics when evaluating alternative designs. Our modeling approach will model the sensor network as a directed graph, allowing us to use state-of-the-art graph algorithms to model the routing and scheduling of packets through the network. We will also use geospatial data to represent environmental conditions, which will provide needed fidelity to our model. Our model with include functions for intelligent routing and scheduling, in which packets are routed and scheduled through the network in order to both avoid adverse conditions and maximize throughput.

I have also used network interdiction modeling to study the vulnerability of wireless networks. Although the field of network interdiction has produced a mature set of modeling and algorithmic tools, these models and algorithms are tailored for “wired” networks, such as supply chains and road networks, and not for wireless ones. Thus, there is a need to extend the concept of interdiction modeling to the wireless domain.

The ubiquitous nature of wireless networks makes them increasingly prone to jamming attacks as such attacks become more sophisticated. In this paper, we seek to gain understanding about a particular type of jamming attack: the flow-jamming attack. Toward this end, we provide a mixed-integer programming model for optimizing the location of jamming devices for flow-jamming attacks. An accelerated Benders decomposition approach was used to solve the model. We solved the problem for two realistic networks and 12 randomly generated networks and found that the Benders approach was computationally faster than CPLEX for nearly all the problem instances, particularly for larger problems with 1000 binary variables. The experimental results show that optimally locating jamming devices can increase the impact of flow-jamming attacks. Specifically, as the number of possible locations increases the jammers’ efficacy increases as well, but there is a clear point of diminishing returns. Also, adding lower-powered jammers to work in conjunction with higher powered jammers significantly increases overall efficacy in spite of the power difference.

We study a new facility protection problem in which one must allocate scarce protection resources to a set of facilities given that allocating resources to a facility only has a *probabilistic *effect on the facility’s post-disruption capacity. This study seeks to test three common assumptions made in the literature on modeling infrastructure systems subject to disruptions: 1) *perfect protection*, e.g., protecting an element makes it fail-proof, 2) *binary protection*, i.e., an element is either fully protected or unprotected, and 3) *binary state*, i.e., disrupted elements are fully operational or non-operational. We model this facility protection problem as a two-stage stochastic program with endogenous uncertainty. Because this stochastic program is non-convex we present a greedy algorithm and show that it has a worst-case performance of 0.63. However, empirical results indicate that the average performance is much better. In addition, experimental results indicate that the mean-value version of this model, in which parameters are set to their mean values, performs close to optimal. Results also indicate that the perfect and binary protection assumptions together significantly affect the performance of a model. On the other hand, the binary state assumption was found to have a smaller effect.

We study the following questions related to wireless network security: Which jammer placement configuration during a jamming attack results in the largest degradation of network throughput? and Which network design strategies are most effective in mitigating a jamming attack? Although others have studied similar jammer placement problems, this article is the first to optimize network throughput subject to radio wave interference. We formulate this problem as a bi-level mixed-integer program, and solve it using a cutting plane approach that is able to solve networks with up to 81 transmitters, which is a typical size for studies in wireless network optimization. Experiments with the algorithm also yielded the following insights into wireless network jamming: (1) increasing the number of channels is the best strategy for designing a network that is robust against jamming attacks, and (2) increasing the range of the jammer is the best strategy for the attacker.