Novel reconfigurable computing architectures for embedded high performance signal processing and numerical applications

Field-Programmable Gate Arrays (FPGAs) have a unique feature set that enables them to be adapted to virtually any computational problem. Among their advantages are flexibility, I/O bandwidth, and computational power. Using this technology, I have developed FPGA-based architectures for a variety of computationally intensive problems. This research has enabled portable, low power, small form factor, and high throughput coprocessors for a variety of computationally intense tasks. Applications discussed include electromagnetic simulation, synthetic aperture radar imaging, and real-time compensation for atmospheric effects in long-range imaging. Some of these applications were previously unfeasible in portable systems, due to power and size constraints, while others had software implementations too slow to be of practical use. Combining reconfigurable technology and the techniques described in this document, these problems can now be solved. I begin with a description of the basic building blocks for these architectures, including floating-point arithmetic, trigonometric, and exponential units, as well as memory and communication controllers. Next, each of the architectural designs is presented in detail, including a discussion on the speedup associated with them. Finally, predictions will be given for how the field of reconfigurable computing will evolve in the coming years.