一个改进的指数双向联想存储器及性能分析

AN IMPROVED EXPONENTIAL BIDIRECTIONAL ASSOCIATIVE MEMORY AND ITS PERFORMANCE ANALYSIS

通过分析Wang的修正指数双向联想存储器（MeBAM），本文提出了一个新的指数式双向联想存储器．该存储器不仅保持了MeBAM的优点，如放宽了Kosko对BAM的连续性假定的限制，排除了BAM的补码问题，同时还大大改善了现有BAM的存储性能和纠错能力．通过定义一个随状态变化而减少的能量函数严格证明了改进的eBAM（IeBAM）在同步与异步方式下的稳定，从而保证了所有要存的模式对成为其稳定点．此外，借助信噪比分析方法给出了IeBAM和MeBAM的信噪比估计．理论分析和计算机模拟结果证实了IeBAM的性能确实优于MeBAM和eBAM．

On the basis of analyzing Wang modified exponential bidirectional associative memory(MeBAM), a novel eBAM is proposed in this paper. The model not only retains the advantages ofthe MeBAM, for example, Kosko's continuity assumption is relaxed and its complement encodingproblem is also eliminated, but also greatly enhances its storage performance and error correctioncapability. The stability of the proposed model, in synchronous and asynchronous update modes ofneuron states, is proven by defining an energy function which decreases as the states of the modelchange so that it ensures all the training pattern pairs to become its stable points. In addition, withthe help of the signal-to-noise-ratio (SNR) approach, SNR estimations for the improved eBAM(IeBAM) and MeBAM are given. The theoretical analysis and computer simulation results verifythe advantages of the IeBAM over those of the MeBAM.