1 / 12

Rigo Dicochea University of California at Santa Cruz Research Advisor: Dr. Donald Gavel

Rigo Dicochea University of California at Santa Cruz Research Advisor: Dr. Donald Gavel Research Supervisor: Marc Reinig. A Matrix Multiplication Implementation for Pre-Conditioning Back Propagated Errors on a Multi-Conjugate Adaptive Optics System. Mission Statement.

yan
Download Presentation

Rigo Dicochea University of California at Santa Cruz Research Advisor: Dr. Donald Gavel

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Rigo Dicochea University of California at Santa Cruz Research Advisor: Dr. Donald Gavel Research Supervisor: Marc Reinig A Matrix Multiplication Implementation for Pre-Conditioning Back Propagated Errors on a Multi-Conjugate Adaptive Optics System

  2. Mission Statement • The goal is to implement a matrix multiplication on a Field Programmable Gate Array (FPGA) to reduce the total number of iterations necessary to solve a system of equations with unknown variables. Without AO With AO

  3. However, since each of these rays passes through different voxels, the total effect of the atmosphere on each of them is different. Background/Iterative Approach Light Rays from Excited Sodium Ions We propagate our initial estimate of phase delay through each voxel. A B Possibly take 100’s of iteration to converge! C D = E Photo credit: Marc Reinig

  4. Evolution of Project A B State Machine C D Hardware = E Resources

  5. State Machine for Matrix Multiplication mult1by11 alu_input = ^b0001; store1x11 alu_input = ^b0000; mult1by21 alu_input = ^b0001;

  6. Pre-Conditioning Implementation

  7. Resources Utilized • 208 FPGA Slices • 18 Registers • 1 ALU • Control Logic/Gates • Multiplexor • MANY Bus Lines(wires interconnecting different hardware) 6 Virtex 4 FPGA’s

  8. Timing • Initial Simulations yield a timing constraint of 200MHz. • Must be able to converge in less than 1 milli-second.

  9. RESULTS!! • Previous iterative solutions took in excess of 90 iterations to converge. • With Matrix Multiplication/Pre-Conditioning we NOW converge in approximately 25 to 50 iterations! • A reduction of 50 to 75 iterations!

  10. What's Next? • Implement fast Fourier transform which will allow for more accurate convergent values. • Multiplex existing hardware to reduce resource consumption and cost. • Determine the total number of FPGA’s necessary to implement system on TMT size telescope.

  11. Illustration of Matrix Benefit

  12. Acknowledgments • Lab for Adaptive Optics • Dr. Don Gavel • Marc Reinig • ATMAOS Project Leader Carlos Andres Cabrera • Center for Adaptive Optics • Xilinx This project is supported by the National Science Foundation Science and Technology Center for Adaptive Optics, managed by the University of California at Santa Cruz under cooperative agreement No. AST - 9876783.

More Related