Translator Disclaimer
1 July 1990 Bounds on achievable accuracy in analog optical linear-algebra processors
Author Affiliations +
Proceedings Volume 1319, Optics in Complex Systems; (1990)
Event: 15th International Optics in Complex Systems, 1990, Garmisch, Germany
Upper arid lower bounds on the number of bits of accuracy achievable are determined by applying a seconth-ortler statistical model to the linear algebra processor. The use of bounds was found necessary due to the strong signal-dependence of the noise at the output of the optical linear algebra processor (OLAP). 1 1. ACCURACY BOUNDS One of the limiting factors in applying OLAPs to real world problems has been the poor achievable accuracy of these processors. Little previous research has been done on determining noise sources from a systems perspective which would include noise generated in the multiplication ard addition operations spatial variations across arrays and crosstalk. We have previously examined these noise sources and determined a general model for the output noise mean and variance. The model demonstrates a strony signaldependency in the noise at the output of the processor which has been confirmed by our experiments. 1 We define accuracy similar to its definition for an analog signal input to an analog-to-digital (ND) converter. The number of bits of accuracy achievable is related to the log (base 2) of the number of separable levels at the P/D converter output. The number of separable levels is fouri by dividing the dynamic range by m times the standard deviation of the signal a. 2 Here m determines the error rate in the P/D conversion. The dynamic range can be expressed as the
© (1990) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Stephen G. Batsell, John F. Walkup, and Thomas F. Krile "Bounds on achievable accuracy in analog optical linear-algebra processors", Proc. SPIE 1319, Optics in Complex Systems, (1 July 1990);

Back to Top