Open-Source Educational Resources
Following the completion of my Ph.D., I have been spending some time working as a Post-doctoral Investigator
for Connexions (cnx.org), an open educational resources project, continuing on a full-time basis a part-time
involvement with the project I have maintained since its inception in 1999. Motivated by the idea that
basic human knowledge and access to educational materials should be free to everyone in the world, Connexions
creates an environment wherein authors can collaboratively create instructional materials which are made
available for free use under a Creative Commons Attribution license. Using Connexions, authors can fashion their
own instructional modules and modify those of other authors, instructors can collect their modules and other
authors' modules into a course, and students can access these items at no cost and even contribute back into the
repository, becoming authors and instructors themselves. Content is encoded in CNXML, an XML-based markup language
which allows modules from a variety of different authors to be mutually compatible, enabling easy re-use of materials
and collection into courses for display both online and in print. Content in Connexions spans disciplines from
elementary music education to advanced digital signal processing, and the volume of the repository is growing daily.
My duties with Connexions are varied and run the gamut from technical management to project marketing,
and they are described in more detail in my résumé.
In addition, I still maintain an active technical research involvement, as described below.
Distributed Data Processing for Sensor Networks
My primary technical research interests lie in the realm of distributed digital signal processing in wireless sensor
networks. Such networks consist of large collections of wirelessly-linked sensor nodes; they can collect,
process, and share data with each other and a central data sink using multi-hop routing. Transmission typically
consumes the lion's share of power in a sensor node, so intelligent, in-network data processing to reduce the amount
of traffic can produce substantial increases in network lifetime. I am currently involved in two main lines of inquiry:
distributed multi-scale data processing algorithms and network programming support for all manner of distributed
algorithms (multi-scale or otherwise). Both of these activities are carried out in collaboration with
the inter-disciplinary COMPASS group at Rice.
Distributed Multi-Scale Algorithms for Sensor Networks
My Ph.D.-thesis research centered around irregular-grid wavelet processing for sensor networks. Wavelet transforms have long been utilized in signal processing to compact signal features of interest for efficient data compression, restoration, and analysis. Operating in a distributed fashion within a sensor network, they can enable substantial reduction of raw data prior to its transmission out of the network.
Applying wavelet techniques to sensor networks poses a twofold challenge. First, wavelet transforms typically operate on data in a centralized setting, with access to the entire dataset. Such omniscience is not feasible for transforms within a sensor network, where sensors can only share data efficiently in regions defined by the networking protocol at work in the sensornet. Second, the bulk of wavelet theory applies to data sampled on a regular grid, a restriction to which any realistic placement of sensors will not typically conform.
To answer these challenges, my collaborators and I have developed efficient, implementable multi-scale wavelet analysis tools for sensor networks. Our state-of-the-art distributed wavelet transform is described in detail in my Ph.D. Thesis. Based on the wavelet theory of lifting, the transform only requires sensor nodes to perform simple calculations using neighboring transform coefficients at various scales and is capable of adapting to changing network conditions. We have demonstrated the utility of the transform to such tasks as distributed compression of measured data and distributed restoration of sensor measurements corrupted by additive noise. The transform and processing are applicable to both purely spatial data, where each node collects a scalar measurement, and spatio-temporal data, where each node collects a time series of measurements. Work continues with the development of TinyOS code artifacts implementing the transform protocol using the network application programming interface discussed next.
Sensor Network Programming Tools for Distributed Data Processing Algorithms
While a great deal of distributed algorithms have been proposed in the sensor networking literature, few have been developed to the point of yielding working networking code for real sensor network hardware platforms. As the goal of many of these designs is reducing the overall node energy consumption for a particular task, such development is crucial to ascertain their true practicality in terms of the complicated power economics induced by sensor network radio environments and routing protocols. These designs often go untested, however, since many algorithm designers lack the networking background required for easily implementing their designs in real hardware. To encourage them to extend their evaluations to hardware, they need an easy-to-use networking application programming interface (API) to enable rapid algorithm prototyping in real sensor networks.
As a first step to developing such an API for sensor networks, we have surveyed all distributed data processing algorithms proposed in the proceedings of one of the premier sensor networking conferences, Information Processing in Sensor Networks (IPSN), from 2003-2006. This study has allowed us to categorize the variety of communication patterns required by the proposed designs. These patterns fall into three main families of sending --- address-based, geographic, and device hierarchy-based --- as well as one of receiving. The details of the API designed to support these communication requirements, along with the results of the survey itself, can be found in Rice University Technical Report TREE0705. We are currently developing a TinyOS implementation of these API calls in a network resource-efficient manner, with an eye toward avoiding competing realizations of different classes of API calls running in parallel.
Collaborators
All the work described here results from collaborations with many talented researchers, including:
- Prof. Albert Cohen (Université Pierre et Marie Curie, Paris France)
- Prof. David B. Johnson (Rice University, Houston, TX)
- Prof Richard Barniuk (Rice University) - my advisor
- Prof. T.S. Eugene Ng (Rice University, Houston, TX)
- Dr. Véronique Delouille (Royal Observatory of Belgium, Brussels, Belgium)
- J. Ryan Stinnett (Rice University)
- Shu Du (Rice University)
- Marco Duarte (Rice University)