K. Skadron, M. Humphrey, B. Huang, E. Hilton, J. Luo, and P. Allaire.
In Proc. of the 2001 International Symposium on Magnetic Suspension Technology, Oct. 2001.

Abstract
One approach for implementing a control system for a magnetic bearing suspension system for high-speed rotating machinery is to use embedded DSP boards.Yet control systems based on DSP boards often require specialized programming and development tools, lack interoperability with standardized architectures and tools, and lack flexibility when computational requirements change. For reasons of cost and upgrade capabilities, it is instead desirable to implement these control systems with general-purpose, commodity PCs. Achieving adequate computational throughput is a major challenge however, even with the most advanced computer systems available today. This paper describes several improvements we make on a previous, uniprocessor version of our real-time controls platform in order to support more computationally-intensive, higher-order magnetic bearing controllers. First, the controls platform is made multiprocessor-capable and gain-scheduled controllers are computed asynchronously on the second processor in the dual-CPU system. Second, new floating-point computation instructions supported by the Pentium III and Pentium 4 are used to speed up the matrix calculation. Finally, other performance-tuning techniques are used in combination with increases in commodity processor computing speeds to optimize the controller. The results show a tremendous improvement in the overall throughput of this real time control platform, without sacrificing predictability. This improvement makes it feasible to implement these high-order magnetic bearing controllers which in turn push the performance of rotating machinery to a higher level.