Mao, Zhi Qiang; (1992) Optoelectronic high order feedback neural network (HOFNET). Doctoral thesis (Ph.D), UCL (University College London).

Text out.pdf Download (5MB)

Abstract

In this thesis, a High Order Feedback neural NET or HOFNET and its optoelectronic implementation are discussed in detail. The HOFNET implements 'high order' by using a feedback loop and normalisation in each iteration. The order of the HOFNET is not fixed but increases by one after each iteration. In the HOFNET system, the input pattern is inserted in the middle of the system and is correlated in every iteration, so the time varying noise in the input pattern will be averaged. This is confirmed by the mathematical analysis and computer simulations. An optoelectronic system of the HOFNET is set up by using a Fourier Transform hologram array for information storage and an electrically addressed spatial light modulator in the feedback loop. The gain for the compensation of the loss in the optical correlation system is obtained by using a computer and a spatial light modulator. We have stored initially 3 patterns and later 14 different patterns in the holograms for pattern recognition and with a partial pattern as an input, the system successfully recognised the input pattern and recovered the obscured parts after 2 (in 3 pattern case) and 3 or 4 (in 14 pattern case) iterations, respectively. In this primary system, the feedback is serial through a computer. We designed two optical parallel feedback systems for the HOFNET by using optoelectronic devices. One is based on an optical fibre amplifier. The other is more attractive and the design uses an optically addressed spatial light modulator in one feedback loop for providing gain and a multiple quantum well based SEED device in the other feedback loop for providing normalisation. Thorough analysis and computer simulation have been done for this net.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item