The work at NIST on wireless ad hoc networks is multidisciplinary in that it includes not only the traditional aspects of networking technology per se but also the many aspects of the theory and technology of signal processing at the application layer and statistical communication theory and technology at the physical layer.  In seeking to advance the performance of wireless ad hoc networks, we seek to use analysis, simulation, and hardware testbeds to characterize and to assess the performance of new designs and protocols.  In concept, each of these tools provides a reference by which to measure the network's performance, and each can be used as a means for validating the other.

Specifically, we have pursued the following goals:

• Investigate the performance of routing protocols for MANETs that have been submitted for possible standardization by the IETF.
• Investigate schemes for implementing user and traffic-type priority in a wireless ad hoc network.
• Carry out a comparative study of wireless network simulation tools with an emphasis on scalability of these tools to large networks, increased traffic loads, increased mobility, etc.
• Develop efficient distributed detection and estimation algorithms for wireless ad hoc sensor networks.
• Investigate methods of network self-organization and clustering in wireless ad hoc sensor networks.
• Develop sensor networks capable of transmitting video information at very low bit rates in a multi-hop, ad-hoc manner.
• Develop kinetic spanning tree concepts for wireless network routing and collaboration.
• Develop a Linux kernel AODV implementation and testbed.

In our work on performance evaluation of MANET routing protocols, we have used simulations using the OPNET and QualNet simulation software packages.  We have developed analytical methods for predicting the outcomes of particular simulations as a means for validating the simulations.  Our work has resulted in our making available for downloading OPNET models for the Dynamic Source Routing (DSR) protocol and the Ad Hoc On-demand Distance Vector (AODV) protocol. In the process of developing these models, we have improved the OPNET model for the IEEE 802.11b MAC/physical layer protocol and technology, and this model is included in the download package.

Using our DSR OPNET model as a platform, we investigated methods for giving priority access to users in an IEEE 802.11b network by making slight changes to the backoff mechanism for the protocol.  We have also studied the statistical behavior of the backoff algorithm and some alternatives to it that seem to provide better performance and fairness.

Our analytical work has resulted in methods for predicting the distribution of link distances in a network of randomly deployed radio terminals, and measures of the probability of n-hop routing paths.  These methods provide intermediate results for assessing network performance based on the underlying connectivity at a given time.

In our work on distributed detection algorithms, we proposed an extension to the nth root parley distributed detection algorithm of Swaszek and Willet (P F. Swaszek and P. Willett, "Parley as an approach to distributed detection," IEEE Trans. on Aerospace and Electronic Systems, Jan. 1995.). Instead of making a single "hard" decision at each sensor node, a two bit quantizer is used to choose the hypothesis and also to provide a confidence measure of this decision.  These "soft" decisions are broadcast to all nodes, and they are used to create a stopping rule that reduces the number of parleys.  For the Bayesian criterion, the probability of error is unchanged, and it is equal to that of a central processor; for the Neyman-Pearson criterion, the receiver operating curve is essentially the same as that of a central processor.  The performance is also compared to that obtained using one-bit decision makers and the majority fusion rule.  Simulation results are provided for the Gaussian shift in mean problem assuming an ideal channel and the binary symmetric channel.

As a first step in studying self-organization algorithms for wireless sensor networks, we created a C++ implementation of the Linked Cluster Algorithm (LCA) of Baker and Ephremides (IEEE Trans. on Communications, Nov. 1981).  Given a set of sensor nodes, the LCA uses a fixed TDMA frame structure to form clusters so that all the nodes of the cluster are within one hop of a distinguished node called the cluster head.  If the cluster heads of two adjacent clusters are not within transmission range of each other, gateway nodes are designated that connect them.  The set of cluster heads and gateways forms the backbone network. We then proceed to study the interaction of the LCA and the parley distributed detection algorithm.  The primary goal is to determine the parameters that are responsible for the performance, and then to draw conclusions about how to optimize them.  The reason for this approach is our contention that one of the primary performance metrics of a wireless ad hoc sensor network is its ability to detect events of interest.

The self-configuration architecture proposed by Subramanian and Katz (Proc. MobiHoc 2000) leads to a hierarchical network with address auto-configuration and a number of other useful properties.  In this component of our work on wireless ad hoc sensor networks, we complete their architecture by proposing a set of distributed algorithms and message formats that allow an actual implementation.  Specifically, we develop protocols that organize the sensor nodes into clusters and then merge the clusters to form groups.  Groups merge to form larger groups, in a hierarchical process that dynamically assigns a unique address to each sensor node. Additionally, a broadcast tree is constructed in a manner to reduce the maximum number of hops along the tree.  We identify a number of important parameters, and we study their effects on the overall system performance.

To support emergency responses to natural disasters, surveillance and information gathering in hostile territories, and robotic search and rescue operations where existing communication infrastructures are not available, rapid deployment of an unstructured mobile network, where each unit is capable of transmitting video information and sensor data, would be essential.  The requirements may include some or all of the following:  a higher upstream bandwidth (for transmitting video data), mobility, sufficient area coverage, non-line-of-sight (NLOS) communications, and low energy consumption.