An integral component of many NASA missions involves remote sensing of the environment, both terrestrial and celestial. This is a challenging problem, since quantities of interest typically can not be directly measured but instead must be inferred. These inferences are made by solving inverse problems, where complex forward models are inverted to estimate parameters of the model. These parameters correspond to physical properties of the environment. Because of the complexity of many forward models, inversion is usually accomplished by minimizing the difference between observations and model predictions through adjustment of model parameters. This minimization process is computationally demanding, since it requires evaluating the forward model many times and minimizing a function of many variables. In this project, we propose to develop, using low-cost high performance hardware accelerators, a fast general-purpose parameter fitting software tool suite for fitting model parameters to observed data. The tool suite will allow NASA scientists to use state of the art high performance computing resources to speed their work. In the Phase I of this project we have shown that the three key components of model fitting, namely model evaluation, gradient calculation and cost functional minimization, can be accelerated using graphical processing unit (GPU) technology. The Phase I work has laid the foundation for Phase II of the project, where the components investigated and developed will be integrated into a parameter fitting tool suite. During Phase II, we will work closely with NASA scientists from the Stratospheric Aerosol and Gas Experiment (SAGE) III mission, the Solar Dynamics Observatory (SDO) and other missions to develop further capabilities of the tool suite.
More »Because of the widespread use of optimization techniques in parameter estimation, the software developed in this project will be used extensively by NASA scientists working on present and future missions. Moreover, with the increasing amounts of data to be processed, hardware accelerated optimization software will provide a valuable, time-saving tool to NASA scientists. Missions that will directly benefit from the proposed software are SAGE III, SDO and Glory, to name a few.
Parameter fitting is a general capability used across many disciplines. GPULib interfaces to IDL, which is used extensively in astronomy and earth sciences. GPULib also interfaces to MATLAB, which is used extensively across most engineering disciplines, fields of study in science, and in finance. Both IDL and MATLAB are focused on data analysis/reduction, which rely heavily on curve fitting. The parameter fitting software resulting from Phase I will be incorporated into GPULib. These new capabilities will increase revenue for the product, and related consulting services, across all technical computing application areas. Also, the capabilities developed in this project will generate new consulting revenue and additional research contracts from government agencies. Tech-X has an established track record of leveraging knowledge gained and software developed under the SBIR program to obtain additional federal funding from various organizations.
Organizations Performing Work | Role | Type | Location |
---|---|---|---|
Tech-X Corporation | Lead Organization | Industry | Boulder, Colorado |
Langley Research Center (LaRC) | Supporting Organization | NASA Center | Hampton, Virginia |