5pUW17. Computing the environment of an artificial neuronal network for underwater signal classification.

Session: Friday Afternoon, December 6

Time: 5:32


Author: Jinwen Zhang
Location: P.O. Box 19, College of Marine Eng., Northwestern Polytechnical Univ., Xi'an 710072, China

Abstract:

An environment for ANN signal processing has been designed and developed, which is written with Turbo Vision's CLASS in Borland C++ 3. Algorithms for mapping a BP network on a machine and task assign tactics are devised to implement ANN parallel algorithms. To use the PC as controller of the whole parallel system, some problems have to be solved. So the floating-point conversion between TMS320C30 and IEEE formats, executable program transforms, as well as data-loading techniques, are developed and presented in this paper. This application system has been used to quickly classify three kinds of underwater signals with a three-layer BP network in the ANN. Results show that the above classification can be finished in 28 (mu)s. The FFT algorithm with 1024 points is also simulated on the system and it takes 2.3 ms. The other advantage of this software is that it is easily expanded to include new functions.


ASA 132nd meeting - Hawaii, December 1996