用光阀和光栅实现神经网络的并行相减(英文)

PARALLEL SUBTRACTION USING LCLV AND GRATING IN AN OPTICAL NEURAL NETWORK

本文在以波长编码实现双极光神经网络的方案基础上，对表示加权求和正、负值的两组光．点，通过在光寻址液晶光阀的相应位置加上不同．的偏置，使其分别工作在正、反态，从而实现对代表正、负值的光强的调制．然后利用光栅的移位作用，把各组的两个光点叠加到一起，其合光强即是所期望的加权求和正、负值的相减结果．通过调整液晶光阀的工作参数，得到了理想输入输出响应曲线；并用光栅实现了移位，从而在实验上验证了该方案的可行性．

Based on the method of wavelength and beam direction encoding for implementing a bipolar optical neural network,a liquid crystal light value (LCLV)is used to encode the positive and negative light spots by adding relevant light biases so as to obtain the normal and inversed states, respectively. Then,the two modulated light spats in the same group are moved together by using a grating. The overlapped light spot whose power is just the expected subtractive results of positive and negative weighted summation can be obtained. Taking advantage of said method, parallel operation between positive and negative light spots can be fulfilled. Due to the high discrimination and quick responsibility of LCLV,system operation speed can be greatly enhanced in optical neural network on a large scale. The feasibility is demonstrated by the preliminary experiment.