Software Optimization for Array Processors. An AP-120B Kalman Filter.
Abstract
Aircraft navigation imposes critical speed requirements on the embedded avionics processor, requirements which will become even more rigorous in the future as additional, increasingly sophisticated navigation data becomes available in the cockpit. The fusion of navigation sensor data to arrive at an accurate position determination is done using a programmed algorithm called a Kalman filter. Achieving the necessary processing speed for next-generation navigation filters will require the use of innovative machine architectures. One of the most promising configurations, given the computational nature of the Kalman filter, is the array processor. To explore the software issues and demonstrate the potential speedup made possible by an array architecture, an in-house research project was undertaken to install a Kalman filter on a vector machine and maximize its execution speed. The actual hardware consisted of a DEC PDP 11/70 host for initialization with a slave FPS AP-120B array processoar to exeute the filter algorithm. In order to assess software optimization techniques, processing times for a simulated scenario were measured initially with the AP-120B programmed to function as if it were a strictly serial processor, and then again after the software had been optimized to exploit the machine's parallel architecture. Algorithm restructuring, expression-tree height reduction, maximization of loop parallelism, and other techniques were applied to the Kalman filter algorithm chosen for this project.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1982
- Accession Number
- ADA119221
Entities
People
- Edward C. Dudzinski
Organizations
- Wright Laboratory