Further Research On Blind Detection Algorithm Based On Hopfield Neural Network

The algorithm of blind detection of Hopfield Neural Network (HNN) is not affected by the limitof the channel with common zeroes. Compared with the algorithm of the second order statistics(SOS) and high order statistics (HOS), the algorithm of blind detection of Hopfield Neural Networkrequires shorter data. So the algorithm of blind detection of Hopfield Neural Network can meet thedata’s high-speed transmission in modern communication system. However, the algorithm of blinddetection of Hopfield Neural Network still exists deficiencies.(1) This paper proposes a new network structure (Double Sigmoid Hopfield Neural Network,DSHNN), builds the new activation function and a new energy function, and analyzes the stabilityof the new network. Simulation results show that whether in the random channel or in the fixedchannel, the convergence speed of energy function of DSHNN is faster than the classical HopfieldNeural Network (Hopfield Neural Network, HNN). Besides that, the DSHNN has a betterperformance than HNN. And the algorithm of DS-HNN is also proved to be appropriate for channelwith common zeroes.(2) Though the DSHNN has improved the speed, its ability of overcoming the noise is limited.So this paper builds a new network structure (Positive Feedback Hopfield Neural Network, PFHNN)to strengthen the anti-noise performance, constructs a new energy function of the network. At thesame time, it proves the stability of the network in asynchronous mode and synchronous updatemode updates. Simulation results show that: If the sending signal is BPSK and the length of data isfixed, whether in either fixed or random channel, PFHNN has significantly better error performancethan the traditional network. In addition, PFHNN can also solve channel common zeroes problemwell and has a strong robustness.(3) In spite of the DSHNN and the PFHNN have improved the performance of the algorithm,the algorithm is still difficult to avoid local minimum. So, this paper builds the Double SigmoidTransiently Chaotic Hopfield Neural Network (DS-TCHNN) based on Transiently Chaotic HopfieldNeural Network (TCHNN). And then, according to the characteristics of DS-TCHNN and TCHNN,this paper gives the energy functions of these two networks. The stabilities for DS-TCHNN withasynchronous and synchronous operating mode are also analyzed separately. Simulation resultsshow that: the algorithm of DS-TCHNN has inherited the characteristics of TCHNN, so it can avoid the local optima. Besides that, the convergence speed of DS-TCHNN is faster than TCHNN, and iteffectively overcomes the disadvantages of slow convergence speed of TCHNN. In addition,DS-TCHNN requires shorter data to reach good performance, so that computational complexity isdecreased and the speed is improved.