Share Email Print

Proceedings Paper

Massively parallel optical interconnect for long data stream convolution
Author(s): Yan Zhang; Yao Li; H. John Caulfield
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

Convolution of long sequences of data is often needed for various sensing, signal processing, and pattern recognition applications. In this paper, massively parallel opto-electronic interconnect schemes are proposed to solve the problem of real-time convolution of long (104 - 106) data systems. Based on the Chinese remainder theorem, a 1D data sequence of length N (N equals r1r2, where r1 and r2 are mutually prime to each other) may be permuted into a 2D array of a size r1 by r2. A convolution of the two 1D data sequences each of N points is therefore converted into a convolution of the two corresponding 2D arrays each of a size r1 X r2. A standard 2D optical image convolver sandwiched between an input and an output opto-electronic data permutation devices (interconnect networks) can thus perform the required convolution. Two video rate opto- electronic data permutation schemes which are based on: (1) the use of a modified cathode ray tube (CRT) and (2) a combination of a standard CRT and an optical geometric transformer, respectively, are described. The permuted 2D data are subject to a standard free-space optical convolution before a 2D to 1D inverse permutation to generate the final 1D convolution result. Computer simulation for the entire three-stage algorithm and their results are discussed. Technical problems and fundamental limitations of the proposed schemes are also discussed.

Paper Details

Date Published: 24 August 1992
PDF: 13 pages
Proc. SPIE 1704, Advances in Optical Information Processing V, (24 August 1992); doi: 10.1117/12.139914
Show Author Affiliations
Yan Zhang, CUNY/City College (United States)
Yao Li, NEC Research Institute, Inc. (United States)
H. John Caulfield, Alabama A&M Univ. (United States)

Published in SPIE Proceedings Vol. 1704:
Advances in Optical Information Processing V
Dennis R. Pape, Editor(s)

© SPIE. Terms of Use
Back to Top
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?