Investigation On New Blind Detection Algorithm Based On Improved Hopifeld Neural Network

Blind detection algorithms based on Hopfield Neural Networks have been widely used inwireless communication because of its ability to be fit for these cases of “do not rely on statisticalinformation, good performance in complex channel environments and good university”. Theresearch work can be summarized as follows:(1)We study the Continuous Hopfield Neural Networks and blind detection algorithms which isbased on Continuous Hopfield Neural Networks. According to the subspace relationship betweenreceived signals and transmitted signals, the proposed algorithm constructs new optimizationperformance functions to blindly detect signals. Besides, through the efficient construction of theweight matrix blind, the signal detection problem can be made to correspond to solving the problemof Hopfield neural network stable equilibrium point. Simulation analysis shows that blind detectionalgorithm based on Hopfield neural network does not depend on the statistics, and is suitable forchannel with common zeros and certain universality. All aspects of performances are much betterthan the traditional blind detection algorithms based on statistics (TXK algorithm, SSA algorithm,LPA algorithm). But the algorithm has the shortcoming; that is, error rate is too high in the low SNRenvironment.(2)To conquer the shortcoming of blind detection algorithms based on Continuous HopfieldNeural Networks, a new kind of activation function is put forward. The stabilities for networks withasynchronous and synchronous operating mode are also analyzed separately. Simulation analysisshows that the improved algorithm can effectively reduce the algorithm’s sensitivity to noise andgreatly improve its anti-jamming capability, so its performance has been greatly improved. But thealgorithm has the shortcoming; that is, the active function is lack of flexibility.(3) To conquer the shortcoming of blind detection algorithms and its improved algorithms basedon Continuous Hopfield Neural Networks, a new kind of blind detection algorithms based onHysteresis Hopfield Neural Network is put forward. The stabilities for Hysteresis Hopfield NeuralNetwork have been proved. In addition, the adjustment mechanism of the activation function andthe impact of noise on the experiment are analyzed separately. Simulation analysis shows that blinddetection algorithms based on Hysteresis Hopfield Neural Network has a tendency to overcominglocal minima and its performance is better than blind detection algorithms based on ContinuousHopfield Neural Networks. But the algorithm depends on more than one starting point and the complexity is too high.(4) In order to only use a single starting point to obtain the optimal solution to this problem,blind detection algorithm based on transiently chaotic neural network is proposed. The simulatedannealing and the parameters’ design are analyzed. Simulation analysis shows that the algorithmdoes not depend on starting points and has low complexity.