Projects

	Current Projects
	Old Projects

Internal Use Only

CMC Projects

This page lists current CMC Projects. The dates associated with each project indicate when the project description was written, and where possible, we provide links to people or pages that have more recent information.

See the Old Projects page for a list of previous CMC projects.

Data Assurance in Medical Sensor Applications
2007

We expect that wearable, portable, and even embeddable medical sensors will enable long-term continuous medical monitoring for many purposes, such as patients with chronic medical conditions (such as the recently announced blood-sugar sensors for diabetics), people seeking to change behavior (e.g., losing weight, or quitting smoking), or athletes wishing to monitor their condition and performance. The resulting data may be used directly by the person, or shared with others: with a physician for treatment, with an insurance company for coverage, or by a trainer or coach. Such systems have huge potential benefit to the quality of healthcare and quality of life for many people.

Since the sensor data may be gathered through a patient's mobile device (such as a mobile phone), a wireless network, and the Internet, there are many opportunities for the sensor data to be tampered or otherwise inaccurate. How can we assess confidence in sensor data? How can we present that level of confidence, in context, with the sensor data? This project will develop methods to assess confidence in medical sensor data.

Funded by the Intel University Research Council.

MetroSense: scalable secure sensor systems
September, 2006

Sensor networks will provide a foundation to protect and monitor our national infrastructure, including economically important businesses with global reach (e.g., stock markets), critical transport and industrial facilities, the enterprise, and the border. These tiny, low-cost wireless devices embed on-board sensing, are fully programmable, and can spontaneously form large sensor webs with thousands of distributed sensor devices. In this project, we will study, analyze, propose, deploy, and evaluate MetroSense, a radically different scalable secure sensor architecture and system capable of reliable real-time monitoring and data fusion for large-scale critical infrastructure, resources, and assets. MetroSense opportunistically leverages mobile sensors when available to deal with sparse coverage and communications when sensing. We plan to develop a campus-area sensing architecture based on three integrated components (sensing and communications, sensor security, and sensor fusion) and deploy the system incrementally across campus with the goal of using static and mobile sensors for reliable monitoring and data fusion of campus plant, spaces, and people flow. Results from this project will serve as a foundation for building secure sensor networks capable of monitoring large-scale critical infrastructure.

Real Time Monitoring of a Wireless Mesh Network for Emergency Response Operations
Soumendra Nanda and David Kotz
July, 2005

Wireless mesh networks can be used to provide communication infrastructure for emergency response operations in areas with limited or damaged infrastructure. We imagine the formation of a wireless mesh of heterogeneous devices such as transceivers on ambulances, fire trucks and police cars. This mesh would support a network of Personal Digital Assistants (PDAs) on first responders and an ad hoc network of rapidly deployed micro-sensor devices. Monitoring of such a mesh network will be crucial to the success of first responder operations. Standard techniques for monitoring wired networks or even wireless infrastructure networks are unsuitable for a wireless mesh network with unpredictable links and resource-constrained devices. Our goal is to develop a wireless mesh monitoring system to detect and identify real-time problems and aid system administrators in making proactive as well as reactive management decisions.

We propose to develop a mesh monitoring system that can be used to generate real-time network topology maps, power maps and provide real-time data on network traffic and user locations to aid mission planners. The aim of this project is to present new ways to efficiently implement a real-time wireless monitoring system that assists in fault detection, repair and the automation of network management tasks. It may also be possible to use the monitored information about the state of the network to improve and optimize the performance of the mesh routing protocol. Some other contributions of this work will be in the use of error codes to recover information from corrupt or lost packets and to maximize utility of monitoring information sent over an unreliable channel. Our initial plan is to deploy a 15-node multi-radio mesh network and to monitor it using in-band channels as well as out-of-band channels (such as a wired backhaul or a separate wireless channel) for the traffic being monitored. Thus we can study the effectiveness of the monitoring system and its impact on the behavior of the mesh network. In parallel, we will simulate a mesh network to study scalability and other characteristics of the monitoring system.

Digital Living: Understanding PLACE (Privacy in Location-Aware Computing Environments)
Faculty: Denise Anthony, Andrew Campbell, David Kotz (lead), Tristan Henderson
Staff: Ronald Peterson
July, 2005

Digital technology plays an increasing role in everyday life, and this trend is only accelerating. Consider daily life five years from now, in 2010: we will each be surrounded by far more digital devices, mediating far more activities in our work, home, and play; the boundary between cyberspace and physical space will fade as sensors and actuators allow computers to be aware of, and control, the physical environment; and the devices in our life become increasingly (and often invisibly) interconnected with each other and with the Internet. Today, typical home users struggle to maintain the security of their home computer, and have difficulty managing their privacy online. Tomorrow, these challenges may become unimaginably complex. This 18-month project studies, and begins to address, the security and privacy challenges involved in developing this world of Digital Living in 2010.

Specifically, this project focuses on the advent of sensor networks, and their applications in the home and work environment. Although sensor networks have been an active area of academic research, and are becoming commercially available for deployment in industrial settings, sensor networks will soon have many uses in enterprise and residential settings. People will live in spaces, or work with devices, that have embedded sensing capability. For people to accept this new technology into their lives, they must be able to have confidence that the systems work as expected, and do not pose unreasonable threats to personal privacy. This confidence results from a variety of technical and organizational mechanisms. This project delves into the sociological underpinnings of privacy and trust in digital living, into the technological foundations for secure and robust sensor networks, and into mechanisms for users to express control over information about their activity.

CRAWDAD: a Community Resource for Archiving Wireless Data at Dartmouth.
July, 2005
http://crawdad.cs.dartmouth.edu/

As a community resource, the CMC is building an archive with the capacity to store wireless trace data from many contributing locations, with the staff to develop better tools for collecting, anonymizing, and analyzing the data. This Community Resource for Archiving Wireless Data At Dartmouth, CRAWDAD, will work with community leaders to ensure that the archive meets the needs of the research community, work with the other leading centers that develop network tracing tools and metadata, and work with research organizations and corporations to ensure continuing support for the archive after NSF's funding ends.

MAP: Measure, Analyze, Protect: Security through measurement for Wireless LANs.
July, 2005
http://www.cs.dartmouth.edu/~map/

With the rise of Voice over wireless LAN (VoWLAN), any complete WiFi security solution must address denial of service attacks, such as kicking off other clients, consuming excessive bandwidth, or spoofing access points, to the detriment of legitimate clients. Even authorized clients may be able to sufficiently disrupt service quality to make the network ineffective for legitimate clients. Our approach provides a new foundation for wireless network security, able to dynamically measure, analyze and protect a WiFi network against existing and novel threats, including rogue clients and access points, with a focus on VoWLAN use cases. Our goal is to support thousands of APs and clients, quickly recognize most new attacks, and generate few false alarms.

Automated Remote Triage and Emergency Management Information System
May, 2005
Susan McGrath, Bob Gray, George Blike, Stephen Linder, Christopher Carella, Janelle Chang, Michael De Rosa, Aaron Fiske, Curtis McClurkin, Suzanne Wendelken

The Automated Remote Triage and Emergency Management Information System (ARTEMIS) is an ongoing research effort at Dartmouth College's Institute for Security Technology Studies that aims to provide real-time physiological information to first responders and command personnel in emergency/disaster situations. The prototype system is capable of monitoring and assessing physiological parameters of individuals, transmitting pertinent medical data to and from multiple echelons of medical service personnel, and providing filtered data for command and control applications.

The system employs wireless networking, portable computing devices, and reliable messaging technology as a framework for information analysis, information movement, and decision support capabilities. Physiological status assessment is based on a medical model that relies on input from humans and a pulse oximetry device. Our physiological status determination methodology follows NATO defined guidelines for remote triage and is implemented using an approach based on fuzzy logic. The approach described on this website can be used in both military and civilian settings.

The long-term goal of the ARTEMIS project is to integrate advances in communications and analysis technologies into a remote triage system that can expedite and improve care of the wounded in small-to-large scale emergency situations. Our aim is to provide an unprecendented degree of medical situational awareness at all levels of the first-responder command heirarchy.

Mobility modeling
April, 2005
Minkyong Kim, David Kotz

Many people who design, develop, or deploy wireless networks use simulations to evaluate the impact of their design decisions on the performance of the network. For these simulations to be effective, however, one must have a realistic model of device mobility. Currently available models of device mobility do not reflect the movement patterns of real users. Using the traces collected by access points (APs) on our campus, we aim to develop realistic mobility models.

We are interested in developing models of both AP-association patterns and physical user movements. The former presents how mobile users roam from one AP to another, while the latter describes how mobile users move in a physical space. To develop an association model, we first extract the characteristics of association patterns directly from the syslog messages (available on this site). We then derive an association model from these characteristics. To develop a physical-mobility model, however, we first need to estimate the physical location of users from the association patterns; this task is not easy because a mobile device does not necessarily associate with the geographically-closest AP. Our path extractor estimates paths from AP-association patterns and has been validated against GPS track data as shown in the figure. These extracted paths are used for developing a physical-mobility model.

WLAN User Mobility Prediction
April, 2005
Libo Song, Udayan Deshpande, David Kotz, Ravi Jain, Ulas Kozat, and Xiaoning He

In wireless networks users can move from one location to another location without losing their network connection. This flexibility of mobility introduces new challenges in quaranteeing quality of service (QoS) and in locating users and transfering data between them and the access points (APs). By predicting a user's next AP we can reduce the overhead of mobility management and make bandwidth reservations to guarantee the QoS.

Many prediction algorithms have been proposed, but most of them are evaluated by simulations using synthetic data. We have collected the association messages at our campus-wide wireless network. From the association messages, we extracted the mobility traces, and evaluated prediction methods using our real wireless mobility data. We found that low-order Markov predictors performed as well or better than the more complex and more space-consuming compression-based predictors.

Besides predicting the next AP, anticipating a user's handoff time is also important for applications such as bandwidth reservation, which needs to know when to reserve bandwidth. It is easier to estimate the handoff probability with a period of time than to predict the exact time. We developed such a time predictor and combined it with a location predictor to compute the probability that a user handoffs to a certain AP within a given period of time. We simulated several bandwidth reservation schemes using this location-and-time-integrated predictor with our real mobility data. The results show that both call-drop rate and call-block rate are reduced significantly.

Since our simulation indicates that with accurate location-and-time prediction the QoS of calls is improved, we would like to improve the performance of predictors. In the future, we will continue to collect wireless association data and investigate the characteristics of users mobility patterns. We believe these mobility characteristics will help us develop better predictors.

Guiding People and Robots with Sensor Networks
March, 2005
Daniela Rus, Ronald Peterson, Peter Corke, Gaurav Sukhatme, Srikanth Saripalli, Stefan Hrabar

A wireless sensor network can extend the sensory perception of people and robots far beyond their normal range. Wireless sensors are also small computers. When the sensors are used to detect danger they can perform distributed computations to compute the safest path along which a person or robot can be guided. Sensors that detect their own network connectivity can be used to guide a robot to repair holes in that connectivity. Sensors that detect a fault in an industrial process can guide a robot or person to the location of the fault for further inspection. Robots and people can also store information in a sensor network which can later be used for guidance, or by the sensor network itself (for example by telling the sensors their GPS coordinates.)
We have been exploring all these concepts in a large variety of experiments. In the picture on the right, USC's AVATAR autonomous flying robot is repairing the gaps in connectivity in a sensor network. The sensor network computed the locations of missing sensors, the robot queried the network for the gap location, and then flew over the gap, dropping new sensors to repair the network.
In the picture on the left, a crane robot at CMU is interacting with a sensor network. The robot is controlled by precision winches connected to the four cables attached to the robot from the ceiling. This type of robot might be used inside a factory to maintain sensors that monitor industrial processes. The robot first broadcasts location messages while moving in a precise pattern to localize the sensors. A radio message was then broadcast to the sensor network and followed a precise geographic path through the sensors. The robot then queried the sensors to follow the same path as the radio message.
We have also been looking at using maps of sensed data to guide people and robots. The picture on the left shows a temperature map as it varies over time in a room where a large fire has been started. Guidance algorithms can make use of such maps to bring people to safety, or to guide firefighters to the danger.
A device we call a "flashlight", shown in the center of the sensors in the picture below, can be carried by a person or robot to find their way through an area based on the data stored in the sensors or on the readings from the sensors.

Quality-managed Group-aware Stream Filtering
November, 2007
Ming Li, David Kotz

This project considers a distributed system that disseminates high-volume data streams to many simultaneous monitoring applications over a low-bandwidth network. For bandwidth efficiency, we propose a group-aware stream filtering approach, used together with multicasting, that exploits two overlooked, yet important, properties of monitoring applications: 1) many of them can tolerate some degree of ``slack'' in their data quality requirements, and 2)there may exist multiple subsets of the source data satisfying the quality needs of an application. We can thus choose the ``best alternative'' subset for each application to maximize the data overlap within the group to best benefit from multicasting. Here we provide a general framework for the group-aware filtering problem, which we prove is NP-hard. We introduce a suite of heuristics-based algorithms that ensure data quality (specifically, granularity and timeliness) while preserving bandwidth.

Our work exploits applications' semantics to better managing precious network resources. For evaluation, we integrate group-aware filtering with a general-purpose sensor data dissemination middleware system, Solar, developed at Dartmouth College. Our evaluation shows that quality-managed group-aware filtering is effective in trading CPU time for bandwidth savings, compared with self-interested stream filtering.

Measuring Wireless Networks
March, 2005
Tristan Henderson, David Kotz, Denise Anthony

IEEE 802.11 Wireless Local Area Networks (WLANs) are now commonplace on many academic and corporate campuses. As “Wi-Fi” technology becomes ubiquitous, understanding trends in the usage of these networks becomes increasingly important for network deployment, management, and the development of new wireless and location-aware applications. We have been measuring various aspects of Dartmouth's campus-wide WLAN since the installation of the network in 2001. The extensive coverage of Dartmouth's WLAN allows us to study how the network is used by students, faculty and staff.

We employ a variety of methods to measure wireless network usage. We have deployed “sniffer” boxes (Linux PCs with multiple network interfaces) around the campus to observe the data packets that are transferred over the network; this enables us to measure wireless application usage. By using SNMP and syslog to monitor the access points we can measure user mobility patterns. We have also deployed wireless sniffers to measure the 2.4GHz and 5GHz frequency bands that are used by IEEE 802.11a/b/g networks; this allows us to measure wireless traffic that does not traverse the wired side of the Dartmouth network, and also lets us observe other wireless networks, such as ad-hoc networks or “rogue” access points. Finally, we are also investigating the use of psychological methodologies, such as the Experience Sampling Method, to ask the network users themselves about their experiences with the wireless network.

We have discovered that since the deployment of the network, usage has moved away from non-realtime applications such as the World Wide Web, with an increasing amount of streaming audio/video and peer-to-peer file transfers being conducted over the WLAN. Although Dartmouth has migrated to a Voice over IP telephone system, we have seen little wireless VoIP usage. We encourage other researchers to make use of the data that we have collected, and anonymised datasets are available on this site.

Mobile Sensors for First Responders
March, 2005
Ron Peterson, Daniela Rus

In many emergency calls the presense of deadly, invisible chemicals is first noticed when people start coughing or falling ill. Even after the presence of a toxin has been verified, unless visible it is difficult to avoid exposure due to air motion. Networks of mobile chemical sensors (sensors on robots) can provide a first warning of nearby toxins, and tell us where they are, where they are moving towards, and how to avoid them.
As part of ongoing work in medical and environmental sensors for first responders, we devised a simulated air crash scenario that involves a chemical leak. The crash throws some debris into a nearby farmers field where a tank of anhydrous ammonia used as fertilizer is present on a trailer attached to a tractor. Anhydrous ammonia, when released into the atmosphere, is a clear colorless gas, which remains near the ground and drifts with the wind. It attacks the lungs and breathing passages and is highly corrosive, causing damage even in relatively small concentrations. It can be detected with an appropriate sensor such as the Figaro TGS 826 Ammonia sensor. Our experiments map the presence of an ammonia cloud and guide a first responder to safety along the path of least chemical concentration. The image on the right shows such a danger map with the safest path computed by a network of 38 Mica Mote sensors.
We are currently exploring the utility of mobile sensor networks in warning, guidance, and sensing for search and rescue missions in the difficult environments created by disaster situations, such as the rubble pile from a destroyed building shown to the left.

Mobile Computers for Herding Cattle with Virtual Fences
March, 2005
Daniela Rus, Ronald Peterson, Peter Corke, Zack Butler

Fences on open ranges cost the cattle industry a lot of time and money to install and maintain. Herding cattle also involves much time and effort. A collaboration between Daniela Rus, the CSIRO Robotics Team in Australia and a USDA Ranch Management Research Animal Scientist was initiated at Dartmouth to consider the problem of monitoring and controlling the position of herd animals.

The goal is to apply the vast body of theory in robotics and motion planning to virtual fences for controlling animals and to integrate new technologies, such as wireless adhoc networking, into a field where technology has yet had little penetration. Similar to the "invisible fence" products sold for fencing pets in the yard at home, a virtual fence is a collar or tag worn by an animal which tracks its location via GPS and applies a stimulus to the animal to control its motion. Animals are not robots and their unpredictable reactions mean that existing robotics motion control solutions must be modified to take into account the imprecise control before they can be useful.

The picture on the upper right shows a cow wearing an early prototype of a Smart Collar during an experiment. The picture on the lower left shows an automatic path planner for herding animals around obstacles to a goal. The picture on the lower right shows another early Smart Collar prototype with a PDA, adhoc WiFi multihop networking, GPS, and sound system for producing stimuli.