development of an internet-based traffic simulation framework for transportation education and
TRANSCRIPT
Development of an Internet-Based Traffic Simulation Framework for Transportation Education and Training
Chen-Fu Liao*Center for Transportation Studies and theIntelligent Transportation Systems InstituteUniversity of Minnesota200 Transportation and Safety Building511 Washington Ave SEMinneapolis, MN 55455E-mail: [email protected]: 612-626-1697Fax: 612-625-6381
Ted MorrisCenter for Transportation Studies and theIntelligent Transportation Systems InstituteUniversity of Minnesota200 Transportation and Safety Building511 Washington Ave SEMinneapolis, MN 55455E-mail: [email protected]: 612-626-8499Fax: 612-625-6381
Max DonathIntelligent Transportation Systems InstituteUniversity of Minnesota200 Transportation and Safety Building511 Washington Ave SEMinneapolis, MN 55455E-mail: [email protected]: 612-625-2304Fax: 612-625-6381
Word Count: Body text (3,531) + 11 Figures (222) = 3,753
*Corresponding Author
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 1
ABSTRACT
Many traffic simulation software packages are available to help traffic engineers and researchers study and evaluate
the potential impact of proposed traffic management strategies and policies. However, existing tools require a
significant investment in time for learning how to create models, perform calibrations and finally analyze the results.
This substantial learning curve severely restricts their application and makes it difficult for engineering students, the
general public and policy makers to take advantage of these tools. An Internet-based traffic simulation framework
was developed to enhance the learning experience for transportation students and engineers. Pre-generated traffic
scenarios were first implemented as part of a Civil Engineering undergraduate class. Based on feedback, an
interactive simulation tool was developed to allow users to make changes to the model and examine the traffic
impacts. This now allows students to for example, minimize the queue length by changing the cycle length or splits.
This interactive traffic simulation tool was deployed and tested in an undergraduate class of 73 students and
feedback was collected from instructors and students that will facilitate additional enhancements of the lab module.
This web-based traffic simulation framework can also incorporate larger road networks that allow one to consider a
multiplicity of control strategies, thus providing a valuable tool for educating and training transportation
professionals.
KEYWORDS
Traffic Simulation, Distance Learning, Transportation Visualization
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 2
INTRODUCTION
Traffic simulation tools have been widely used by transportation engineers and consultants to assist traffic managers
and operators with the evaluation and analysis of the potential impact of design or control strategy changes. As an
early example, Hourdakis and Michalopoulos [1] created a framework that allowed traffic operations personnel to
try various strategies for managing a 'virtual' freeway incident. This was achieved by real time integration of traffic
microsimulation software with a GUI (Graphical User Interface) front end. Their GUI displayed a 2D traffic map of
the modeled freeway network (traffic volume is mapped to a color), variable message sign (VMS) I/O, and an
interface to remote cameras (for live video feeds). The GUI and the traffic management strategies were formulated
from rules written with a 'natural language' expert system programming language (http://www.gensym.com). Thus,
they were able to evaluate how traffic operations personnel managed incidents while varying the content and UI
(User Interface) presentation of the 'sensed' traffic data. The framework could also be 'switched' to use 'live' traffic
instead of the simulated traffic (i.e., by collecting freeway loop detector data in real-time). These tools are however
rather complex and have been developed for and used by researchers, and trained traffic engineers. Existing tools,
usually only available in dedicated laboratories, require a significant investment in time for learning how to create
models, perform calibrations and finally analyze the results. This substantial learning curve severely restricts their
application and makes it difficult for engineering students, the general public and policy makers to take advantage of
these tools.
The benefit of using simulation to train people who deal with matters of life and death (e.g. pilots) is clear.
However, many do not see the benefit of using simulation to teach material that has been taught using traditional
methods in the classroom [2]. Yet, it is well understood that conveying complex concepts can be achieved by
exploring them through simulation. Accordingly, in order to enhance the learning experience and understanding of
intelligent transportation systems (ITS) by students and traffic engineers, we created a web-based traffic simulation
module for an undergraduate transportation engineering course at the University of Minnesota. This web-based
traffic simulation lab module was developed based on a microscopic traffic simulation package, AIMSUN
(Advanced Interactive Microscopic Simulator for Urban and Non-urban Networks, http://www.aimsun.com) [3].
AIMSUN is embedded in GETRAM (Generic Environment for TRaffic Analysis and Modeling), a simulation
environment inspired by modern trends in the design of graphical user interfaces adapted to traffic modeling
requirements [3]. GETRAM comprises a traffic network graphical editor (TEDI), a network database, a module for
storing results, and an Application Programming Interface (API) to allow interfacing to other simulation or
assignment models. An additional library of DLL functions (GETRAM Extensions) enables the system to
communicate with external applications [3, 8]. AIMSUN has been used successfully for numerous large-scale traffic
modeling research projects within the lab [4, 5] and provides a well-documented API to access and modify all
elements of the simulation state (signal control, sensing, vehicle characteristics and state) while the simulation is
running. Of course, other simulation packages with similar capabilities can be integrated into our web-based traffic
simulation framework. These include, CORSIM [6] (Corridor Simulation, a comprehensive microscopic traffic
simulation, applicable to surface streets, freeways, and integrated networks,
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 3
http://ops.fhwa.dot.gov/trafficanalysistools/corsim.htm) operated under the TSIS (Traffic Software Integrated
System) environment and VISSIM (a microscopic, behavior-based multi-purpose traffic simulation program,
http://www.ptvamerica.com/vissim.html).
Our goal was to help students and engineers understand issues related to traffic management and operations. Using
this new simulation tool accessed in real time over the Internet, students were able to analyze existing traffic
situations and provide potential solutions to improve traffic operational efficiency.
TRAFFIC SIMULATION ON THE INTERNET
There are many commercial traffic simulation software packages available to help traffic engineers and researchers
study and evaluate the potential impact of proposed traffic management strategies and policies. Creating a traffic
model and calibrating the model is a fairly complicated process which often discourages students or engineers from
learning and understanding the importance and impact of traffic control. The availability and interactive nature of
web-based traffic simulation provides an excellent medium for students to experience the complexity and dynamism
of traffic operation, management and control. From an educational point of view, this interactive web-based tool
allows people to try their own design/control approach in the simulated network without disrupting the actual traffic.
From an operation and management point of view, traffic simulation helps the user test and verify different control
strategies (ramp metering, signal timing control, etc.) for different traffic conditions. Clearly this approach can be
integrated with other distance learning approaches already in place for teaching ITS technologies (Web Research
Modules, http://www.its.umn.edu/education/modules.html#webmodule, for high school students). For example,
Helbing et al [7] developed multilane freeway traffic models to help people better understand on-ramp vehicle
merging, lane-changing, car following, lane-closing, and signal control through online traffic simulation and
visualization (http://www.mtreiber.de/MicroApplet/index.html). Lastly, new traffic control strategies that are
proposed often meet considerable resistance from the general public. We believe that by providing traffic
simulations which incorporate visualization tools available on the Internet along with the ability to investigate and
analyze different cause and effect scenarios, various stakeholders will better understand the impact of changes in
traffic operations or proposed ITS technologies.
OUR APPROACH
An initial study site of Washington Avenue SE on the Minneapolis campus of the University of Minnesota was
chosen. The Public Works Department of the City of Minneapolis provided intersection signal timing and traffic
flow data. Digital Orthorectified Quad aerial images (DOQs) were acquired from the Metropolitan Council (the
MPO for the greater twin cities 7 county region) for network geometry layout. Scaling factors of the aerial images
were calculated to reflect the corresponding scale in the simulation model. Network geometry was created using the
graphical traffic editor (TEDI) that is integrated into the GETRAM environment [8] with the aerial images placed in
the background. In addition to traffic network geometry modeling, a software interface was developed to access
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 4
traffic simulation data through the GETRAM extension API (Application Program Interface) and store data to a
central database (MySQL, http://www.mysql.com) server while the simulation is running. Java client/server
applications were developed to handle data communication between the client computer, traffic simulators, web
server and database server. Java has been widely used in web based application and web simulation. The Java
technology provides an ideal platform for developing reliable, portable and scalable applications for web-based
traffic simulations.
Traffic simulation
The vehicle data on a user’s PC is updated and graphically displayed during every simulation step by sending
queries to the web server. The web server processes and forwards the request to the database server and returns the
requested data. Intersection signal timing information and time-space diagrams can also be displayed, as shown in
Figure 1, with the selected vehicle(s) traced in real time. Vehicle information (such as vehicle ID, type, length,
width, location and heading data) can be logged and saved from the graphical user interface.
Interactive simulation model
In the initial phase, our web-based 2D traffic simulation module limited users to choosing only from a fixed set of
pre-generated scenarios for their analysis. The student could not design new signal control strategies or traffic
demand models for the network. We have since developed an “interactive” traffic simulation lab module that will
allow on campus or distance-learning students to configure their own signal control strategies and traffic demand
patterns—and see the results either in ‘real-time’ or in ‘accelerated time’ using the client web interface. As before,
users can select, for example, a bus and watch it run its route through the network while monitoring the distance
traveled on time-space plots overlaid with the signal phases along its route (Figure 1). Simulation results are saved
in a traffic simulation database. Measures of Effectiveness (MOEs), including traffic flow, travel time, delay time,
speed, density, stop time, number of stops, total travel distance and total travel time, for the simulation can be
accessed for each simulation run for further analysis. The ability to run the simulator in ‘real-time’ can be used to
test the efficacy of various operational management strategies. For example, students can create ‘incidents’ or lane
blockages and monitor how well their incident management strategies work.
The new module design consists of an interactive web-based application, a database server, web server, and several
custom developed JAVA back-end services (i.e., Java servlets) as shown in Figure 2. The web application displays
the vehicle movements and signal states either ‘as they happen’ in real-time or by scrolling through post-processed
simulation data. The servers act as a communication ‘bridge’ between the web application and the microsimulation
software running on a PC within the ITS Laboratory network. From here on, we define the PC which runs the traffic
microsimulation software a node, and each execution run of the software a simulation instance. The backend
services also manage simulation node selection and subsequent instance execution by remotely monitoring the CPU
loads and process executions using freely available software (http://www.beyondlogic.org). Finally, the back-end
services interface with the lab’s OpenLDAP authentication system as a means of user access and authentication
control.
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 5
Before the simulation starts, the web interface first communicates with the web server and scans through the
available simulation instances (our lab can execute up to 4 instances simultaneously) and determines the least loaded
node available to execute the simulation in batch mode. User inputs (such as traffic demands, signal timing,
simulation period) are formatted into corresponding data files and sent to the remote simulation engine prior to the
simulation. Resulting statistics (MOEs, such as traffic flow, travel time, delay time, speed, density, stop time,
number of stops, total travel distance and total travel time) are stored in a centralized database during the simulation
at minute intervals. The graphical interface on the user’s computer will start rendering the real-time simulation as
soon as data are available. When the simulator finishes the batch mode execution (a batch execution could take
about one minute for an hour of simulation for the single intersection example), a simulation time slider on the
screen becomes enabled which allows users to jump forward or backward in time within the simulation period.
Users can use this feature to examine traffic conditions such as queue length or delay, more efficiently. Finally,
simulation statistics, as shown in Figure 6, can also be saved to the user’s PC and transferred into a Microsoft Excel
for later analysis.
Previously, students relied mainly on formula suggested in the Highway Capacity Manual [9] to calculate estimated
cycle time, green splits, lost time, delays and make the necessary adjustments for left-turn movement. The design of
a feasible signal timing plan is a complex and iterative process that is generally carried out with the assistance of
computer software (for example, Highway Capacity Software, http://mctrans.ce.ufl.edu/hcs/). However, there was
little or no verification or feedback on how well the design of the signal timing performs. The traffic simulation tool
helped students visualize traffic parameters, such as lost time, gap acceptance, queue length and delay, and make
adjustments to the signal timing plan as needed. We shall now present an example of a prototype module recently
used for an undergraduate transportation engineering class of 73 students which addressed some of these issues.
Single intersection example
In a single intersection example, as shown in Figure 3, the web interface allows users to specify traffic demands
from each approach (Figure 4) and design desired signal cycle time and green splits (Figure 5).
In the lab module designed by the course teaching assistants, existing intersection traffic demand and turning
movements were provided. Students were asked to develop a new signal timing plan. They needed to evaluate three
alternative plans, each with different objectives:
1. Minimize the intersection delay
2. Minimize the number of stops per vehicle
3. Based on a combination of the above two, consider alternatives for the optimization using the performance
index (PI),
PI = 1 * delay (sec)/vehicle + 10 * number of stops/vehicle.
Students needed to find the timing plan with the lowest performance index.
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 6
Students were also asked to develop a signal timing plan based on the traffic demand forecast (e.g. 5% increase for
NB and SB directions, and 2% increase for the EB (East Bound) and WB (West Bound) directions over the next 5
years while the turning proportions remain the same) using the same performance index as described above.
Students were to investigate timing plan changes if the cycle length remained the same.
We collected feedback from students on the interactive web-based traffic simulation model. One of the comments
was to display the statistics graphically on the web interface without the need for post analysis of data using a
spreadsheet. We realized the need for a graphical interpretation and an explanation of the simulation results. We
therefore added several graphs from the resulting statistics including traffic flow, number of stops per vehicle per
Km (Figure 7), travel time, delay (Figure 8), speed, queue length (Figure 9), stop time and a performance index
within the simulation period. Providing these simulation statistical results helps the users visualize the impact of the
applied control strategy and allow him or her to make adjustments more efficiently. For example, the user can
readily identify the vehicle queue length (17 vehicles) at 9:18AM (or 18 minutes from the beginning of the
simulation) as shown in Figure 9. The user can then adjust the simulation time slider (Figure 3) to a time stamp a
few minutes before 9:18AM and inspect the forming queue and the traffic conditions around the interested
simulation timeframe. A portion of queue length from 9:15AM to 9:23PM (or simulation time from 900 sec to 1380
sec) where the maximum queue occurred in northbound traffic was plotted and overlaid with signal phases (see
Figure 10) as an example to better understand the queue formation associated with the signal timing.
FUTURE WORK
We will continue to enhance the design to include larger road networks that allow one to consider a multiplicity of
control strategies, for example, adding a bus signal priority module or a ramp-metering module. We currently
continue to integrate the web-based traffic simulation with a 3D visualization interface. The visualization interface
can render the simulation results in a virtual reality environment and provide real-time interactivity. Future
enhancements for the 3D software will allow the user to program signal timing plans and select vehicle probes
similar to the 2D applications.
In order to further enhance the user’s ability to understand the traffic effects within the context of the infrastructure,
we are currently exploring a 3D component to the simulation. Users can interactively view the simulation from any
perspective—including from within or above a moving vehicle. The user can gain additional insights into the nature
of the simulated traffic flow, such as shockwave propagation behavior [10], car-following model [11], driver-
roadside traffic information infrastructure, and context design and urban planning issues [TRB visualization in
transportation, http://www.trbvis.org], [12,13]. We first developed a 3D web-based traffic simulation for the
Washington Ave corridor in Minneapolis, as shown in Figure 11(a & b), using VRML (Virtual Reality Modeling
Language) EAI (External Authoring Interface). The VRML EAI interface allowed users to control the content of a
VRML browser window embedded in a web page from a Java applet interface. It enabled embedded objects on a
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 7
web page to communicate with each other through a browser plug-in. Unfortunately, we found that the VRML EAI
interface scales poorly (memory and 3D rendering performance) as more vehicles were added to the network. We
are therefore developing a standalone 3D web application based on OSG (OpenSceneGraph,
http://www.OpenSceneGraph.org) and CommonC++ (http://www.cplusplus.org) that can interface with the web-
based traffic simulation framework previously discussed.
We believe that the framework represents a novel, flexible approach to further engaging students and the general
public in the transportation decision making and planning process. Our primary intention was to use our framework
to allow individuals to internalize important concepts in traffic engineering and ITS issues in their mind’s eye in an
efficient and scalable manner. Note that similar approaches are now available for other more generalized numerical
simulation software (www.mathworks.com) [14]. However, one interesting issue arises from a software licensing
standpoint. Essentially, we have created an interface for our site that automates the process of running the software
for a desired set of conditions specified by another person and to then send back the results. The software is run
under our authority at our site on our behalf for a person whom we have knowledge of the request. Certainly, a less
automated interface would be for that same person to use e-mail communication in order to request the input and
outputs to the model for a given scenario. Further discussion is needed on how traffic simulation software, can best
be integrated to emerging pedagogical distance e-learning approaches. For example, we wish to integrate actual
state-of-the-art sensing and signal control hardware into our framework. Other facilities and laboratories have done
this [15, 16], although our framework will allow individuals to learn and ‘experience’ these technologies from any
location having Internet access.
ACKNOWLEDGEMENT
We would like to acknowledge the Intelligent Transportation Systems Institute, University of Minnesota, for
supporting this work and the Institute for New Media Studies, University of Minnesota, for providing some
resources for this effort. We would also like to thank Scott Tacheny, traffic engineer in the City of Minneapolis, for
providing the traffic and signal timing data, Professor David Levinson for giving access to the students and his
teaching assistants, Nebiyou Tilahun and Lei Zhang, and other transportation engineering faculty and students of the
Civil Engineering Department for providing us with their invaluable feedback.
REFERENCES
[1] Hourdakis J., Michalopoulos P.G., “Development And Implementation of A Virtual Traffic Management Center”, in Proceedings of the World Congress on Intelligent Transportation Systems, Turin, Italy November 2000.
[2] Aldrich C., “Simulations and the Future of Learning: An Innovative (and Perhaps Revolutionary) Approach to E-Learning”, Pfeiffer, September, 2003
[3] “AIMSUN Version 4.1 User’s Manual”, Transport Simulation Systems (TSS), Barcelona, Spain, Mar. 2002.
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 8
[4] Hourdakis, J., Michalopoulos, P. and J. Kottommannil. “Practical procedure for calibrating microscopic traffic simulation models” Transportation Research Record 1852: pp. 130-139, 2003.
[5] Xin, W., Michalopoulos, P., Hourdakis, J., and D. Lau, “Minnesota's new ramp-control strategy: Design overview and preliminary assessment” Transportation Research Record (in press).
[6] Owen, L., Zhang, Y., Rao, L., and McHale, G. “Traffic Flow Simulation Using CORSIM,” Proceedings ofthe 2000 Winter Simulation Conference, 2000: pp. 1143-1147.
[7] D. Helbing, A. Hennecke, V. Shvetsov, and M. Treiber “Micro- and macro-simulation of freeway traffic”, Mathematical and Computer Modeling 35(5/6), pp. 517-547, 2002.
[8] “GETRAM Extensions Version 4.1 User’s Manual”, TSS, Barcelona, Spain, Mar. 2002.
[9] Highway Capacity Manual – “Chapter 16: Signalized Intersections”, Transportation Research Board, National Research Council, Washington DC, 2000.
[10] Franklin, R.E. (1961), The Structure of a Traffic Shock Wave. Civil Engineering Pulb. Wks. Rev. 56, 1186-1188.
[11] Del Castillo, J.M. (1996). A Car-Following Model based on the Lighthill-Whitham Theory. In: Lesort, J.B. (ed), Proceedings of the 13th International Symposium of Transportation and Traffic Theory, Lyon, 517-538.
[12] Hearne, L. P., and Matthews, D., “Improving the Geospatial Data Extraction and Analysis Process Using Stereo Imagery Datasets,” American Congress on Surveying and Mapping (ACSM), Nashville TN, April 16-21, 2004. http://www.acsm.net/HearnePhotogrammetry42004.pdf
[13] Jones, Ted, Hamilton-Smith, G., and Matthews, N.D., “Integrating Remotely Sensed Imagery And Information For Transportation Infrastructure Management,”, American Society for Photogrammetry & Remote Sensing, ISPRS-Pecora, Denver Colorado, November 10-15, 2002.
[14] Sysel, M. and Pomykacz, I., “Extension of MATLAB Web Server”, In Proceedings: Advances in Computer Science and Technology, 2004.
[15] Engelbrecht, R., C. Poe, and K. Balke, “Development of a Distributed Hardware-in-the-Loop Simulation System for Transportation Networks,” in Proceedings of the 78th Annual Conference of the Transportation Research Board, 1999, Washington, D.C.
[16] Bullock, D. and A. Catarella, “Real-Time Simulation Environment for Evaluating Traffic Signal Systems,” Transportation Research Record 1634, 1998: pp. 130-135.
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 9
List of Figures
Figure 1. Java-Based User Interface
Figure 2. Web-Based Traffic Simulation Framework
Figure 3. Interactive Traffic Simulation Interface
Figure 4. Intersection Traffic Demand and Turning Proportion Interface
Figure 5. Intersection Traffic Signal Timing Interface
Figure 6. Traffic Simulation Statistics
Figure 7. Traffic Simulation Result – Number of Stops per Vehicle per Km
Figure 8. Traffic Simulation Result – Average Vehicle Delay
Figure 9. East Bound Queue Length of Timing Plan A
Figure 10. North Bound Queue Length Overlaid with Signal Phases
Figure 11(a). Snapshot of VRML Model of Intersection at Union St. and Washington Ave. SE in Minneapolis
Figure 11(b). Snapshot of VRML Model of Intersection at Harvard St. and Washington Ave. SE in Minneapolis
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 11
Figure 2. Web-Based Traffic Simulation Framework
Figure 3. Interactive Traffic Simulation Interface
TrafficSimulationDatabase
Server
Web Server
HTTP Applications
Traffic demand and signal control settings, authentication
and trigger execution
LDAPAuthenticate
LDAPAuthorize
Application check validation of session key Return session key
validation and user application ID
OpenLDAP Database
Server
PC 2GETRAM
EXTuid.dll
PC 3GETRAM
EXTuid.dll
Traffic simulation nodes
Generate & return
session key
Simulation resulting statistics,vehicle data and trajectories
TrafficSimulationDatabase
Server
Web Server
HTTP Applications
Traffic demand and signal control settings, authentication
and trigger execution
LDAPAuthenticate
LDAPAuthorize
Application check validation of session key Return session key
validation and user application ID
OpenLDAP Database
Server
PC 2GETRAM
EXTuid.dll
PC 3GETRAM
EXTuid.dll
Traffic simulation nodes
Generate & return
session key
Simulation resulting statistics,vehicle data and trajectories
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 12
Figure 4. Intersection Traffic Demand and Turning Proportion Interface
Figure 5. Intersection Traffic Signal Timing Interface
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 13
Figure 6. Traffic Simulation Statistics
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 14
Figure 7. Traffic Simulation Result – Number of Stops per Vehicle per Km
Figure 8. Traffic Simulation Result – Average Vehicle Delay
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 15
Figure 9. East Bound Queue Length of Timing Plan A
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 16
NB Queue Length vs. Signal Phases
0
1
2
3
4
5
6
7
900 1000 1100 1200 1300 1400
Time (sec)
Qu
eue
(veh
/lan
e)
Red Green Amber Queue
Figure 10. North Bound Queue Length Overlaid with Signal Phases
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.
Liao, Morris, and Donath 17
Figure 11(a). Snapshot of VRML Model of Intersection at Union St. and Washington Ave. SE in Minneapolis
Figure 11(b). Snapshot of VRML Model of Intersection at Harvard St. and Washington Ave. SE in Minneapolis
TRB 2006 Annual Meeting CD-ROM Paper revised from original submittal.