The Study Of DHNN And The Implementation Based On FPGA

Since J.J.Hopfield put forward DHNN,scholars did profound researches on it.Assoc-iative memory is its important research direction,and the ability is implemented through weight matrix,so designing appropriate weight matrix is the key to building DHNN.Hebb Rule is the common method of designing the weights,but the method has its drawbacks that the network is easy to fall into the pseudo states and has a smaller storage capacity. On the base of Hebb Rule,this paper puts forward four novel methods of designing the weights.Using these methods,the networks could recall more non-orthogonal memory samples accurately and have larger storage capacities,which improves the recognition ability of DHNN.Through theoretical proof and Matlab simulation,the four novel methods are proved correct and practicable.Furthermore,this paper proves their convergence through Energy Function.Because the FPGA implementation of Artificial Neural Network overcomes the drawbacks of software implementation in a great extent,this paper describes the FPGA design of DHNN.The study in this paper offers better theory and technology support for the weight design and hardware realization of DHNN.