NSF Program: CAREER/NCR/CISE	NSF Award Number: NCR-9624791
PI Name: Timothy X Brown	Period Covered: 9/1/96-9/1/97
PI Organization: University     Date: 9/19/97
   of Colorado	
PI Address: Campus Box 530, 
   Boulder, CO 80303-0530

Summary:

In the spirit of the CAREER, progress was made on the educational and
research front. In education, a $600,000 software and workstation
donation was secured from Mobile Systems International to equip a lab
for the Telecommunications course I teach in wireless communications.
Over the summer, labs have been developed using this equipment, and
they are being used this Fall. I have also spearheaded an effort to
develop a Capstone Program as an alternative to the current thesis
masters option with a potential for greater industrial involvement in
student education.  These address the projects educational goals of
Section 2.3 in the proposal.

On the research front, I have been looking at the adaptive admission
control problem. The research scope has been broadened from admission
control to the more general statistical multiplexing problem which
applies to admission control, routing, and bandwidth dimensioning [3].
Theoretically, we have developed proofs that in general non-adaptive
approaches to this problem will be sub-optimal, and that in the limit
of infinite monitoring data the adaptive methods will be optimal
[1][4]. Various versions of this were successfully applied to ethernet
traffic traces, and simulated data with heavy-tail distributed packet
sizes, inter-source correlations, and measurement errors [2][4].

Whereas the approach so far has formulated the adaptive method as a
statistical classification problem that classifies what types of
traffic will meet or not meet quality of service levels, a doctoral
student, Hui Tong, has been working on this as a regression problem
that estimates the quality of service as a function of traffic types
[6]. This latter method may lead to more flexibility application.

Current work in progress includes evaluating which statistical
classification methods are suitable for the access control problem and
the question of on-line vs. off-line adaptation (Section 3.6.3 in
proposal). This may tie to related work in packet arbitration [5].

Approximately 19% of the four year project's expenses were spent in
this first year. A web version of this report is available at New
support for this research comes from a recently funded Special
Projects/NCR/CISE/NSF grant High-Performance, Low-Power Wireless
Communication.

Project Sponsored Publications and Presentations:

[1] Brown, T.X, "Adaptive ATM Control," Invited Talk, SNNS Annual
Meeting, Nov. 20, 1996, Stockholm Sweden.

[2] Brown, T.X, "Adaptive Access Control Applied to Ethernet Data,"
Advances in Neural Information Processing Systems, 9, MIT Press, 1997.
pp. 932-938.

[3] Brown, T.X, "Bandwidth Dimensioning for Data Networks," Proceedings
of the International Workshop on Applications of Neural Networks to
Telecommunications, Erlbaum Pub., Hillsdale, NJ, 1997. pp. 88-96.

[4] Brown, T.X, "Adaptive Statistical Multiplexing for Broadband
Communications," Invited Tutorial: Fifth IFIP Workshop on Performance
Modelling & Evaluation of ATM Networks, 21 July, 1997 Ilkley, U.K. Also
revised and submitted for review as a book chapter in upcoming book
organized by the conference.

[5] Brown, T.X, "A High-Performance Two-Stage ATM Switch," Fifth IFIP
Workshop on Performance Modelling & Evaluation of ATM Networks, 21
July, 1997 Ilkley, U.K. Also revised and submitted for review as a book
chapter in upcoming book organized by the conference.

[6] Tong, H., Brown, T.X, "Neural Network Loss Rate Estimation,"
submitted to Neural Information Processing Systems, 10, Denver, Dec.
1997.