Numeric and Scientific Computing
Much of his software and systems development at GTRI, Lockheed, and Unisys had a high level of numeric and scientific computing content, and so directly benefited from his numerical analysis expertise. Some methods he proposed and solutions he developed were fundamentally (mathematically provable) orders-of-magnitude more efficient than those that previously had been developed. Examples:
- At GTRI, provided algorithmic (numerical analysis) expertise to turn around several contracted projects from the verge of over-running their time-share computing budgets by developing significantly (orders of magnitude) more efficient implementations of their signal processing programs and libraries. Thus, several previously ad hoc applications were transformed into a set of standard tools and libraries for use by other projects of GTRI.
- As the resident numerical analyst in the department, advised other staff on the use of other standard mathematical libraries, LINPACK, EISPACK, et. al.—so that many projects benefited from his expertise.
- While many performance improvements were realized through the application of better programming, the most significant improvements resulted from his rethinking the problems, deriving better formulations of problems to be solved, together with better numerical methods and implementation techniques, etc.
- In addition to the above standard signal processing efforts, other particularly diverse examples include tailoring radar algorithms for a MARS—a massively parallel (long instruction word) array processor with extremely fast multiplies but no hardware divides, and for a PDP RT-11 system used in the field for real-time data acquisition and processing. In each case, the algorithms were developed and tailored to match the capabilities of the platform.
- Derived from Richard Huynen’s dissertation, “Phenomenological Theory of Radar Targets,” he developed analytical solutions (in terms of radar observables, hence termed phenomenological) of the equations derived there, and applied them to the analysis of in-house radar data. Developed the mathematical modeling, software to validate the results, the contract proposal, etc. —leading to DARPA contracts for GTRI, and to further additions to the GTRI standard signal processing programming libraries.
- At Lockheed, rescued another numerically intense project by providing a novel solution to a graphics display problem—tailored to particular capabilities (and limitations) of the targeted hardware—that was orders of magnitude more efficient (storage and processing) than the text-book solutions previously considered by the project’s staff and the department.
- At Unisys, enhanced an existing GIS (geographic information system) by re-implementing its geographic display routines and database organization, thereby reducing by orders of magnitude both the storage requirements—to less than one-hundredth original storage requirement, and the processing map display generation time from one-half hour to only a few seconds time—yet with no loss of display fidelity.
- Ported the GIS (the system and its databases) to a military-capable version—that is, one targeted for military-capable platforms, in support of DoD contracts.