多阈值随机汇池网络自适应估计性能研究  

Analysis of Estimation Performance of Stochastic Pooling Networks with Multilevel

作　　者：景文腾 耿金花[1] 韩博 段法兵[1] JING Wenteng;GENG Jinhua;HAN Bo;DUAN Fabing(Institute of Complexity Science,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学复杂性科学研究所

出　　处：《复杂系统与复杂性科学》2019年第3期87-92,共6页Complex Systems and Complexity Science

基　　金：国家自然科学基金(61573202)

摘　　要：本文研究了数模转换中多阈值随机汇池网络的自适应信号估计性能,给定网络节点数目,将模数转换的阈值进行均匀划分,分析了随机汇池网络输出的分布函数,理论给出了多阈值随机汇池网络的最优权向量和最小均方误差表达式,以及大规模网络输出的Fisher信息量近似值,实验验证了多阈值随机汇池网络中超阈值随机共振现象,随着阈值数量的增加,噪声的有益性逐渐减弱,而网络估计的最小均方误差不断变小且逐渐接近Fisher信息意义下的误差界。研究结果表明多阈值随机汇池网络的自适应信号估计方法具有重要的应用价值。An optimal weighted stochastic pooling network is used as the basic framework for analog-to-digital converter (ADC) with multilevel quantizers.This paper,for a fixed number of network nodes,divides the threshold uniformly for easily implement and low costs.Based on the output distribution of the networks,the expressions of the optimal weight vector and the minimum mean square error are derived theoretically.For a sufficiently large size of networks,the Fisher information of the network output is also obtained.The results show that,as the network size increases,the minimum mean square error becomes smaller and smaller,and the noise benefit gradually disappears.However,the minimum mean square error at the optimal noise level approaches the bound denoted by the Fisher information.These theoretical and experimental results of multilevel networks are significant for adaptive signal estimation.

关 键 词：随机汇池网络 多阈值划分 超阈值随机共振 均方误差 Fisher信息 

分 类 号：TN911.7[电子电信—通信与信息系统]