离散Hopfield实现多值信号盲检测

Blind Multi-value Signal Detection with Discrete Hopfield Network

传统神经网络算法局限于二值神经网络,无法解决多值信号盲检测问题。提出了基于离散Hopfield网络直接盲检测多值信号的两种算法,构造了4电平的离散激活函数。本算法不依赖统计量,建立直接盲检测发送信号的优化性能函数,利用离散Hopfield网络直接盲检测多值信号。给出新的适用于多值离散Hopfield神经网络的能量函数和网络权矩阵,理论证明该网络具有很好的稳定性。仿真结果表明本算法能快速收敛到真平衡点,盲检测效果好,适用于随机信道。

The traditional neural network algorithms are limited to two-state neurons and unable to solve the problem of blind multi-value signal detection. Two new algorithms based on discrete Hopfield network （DHNN） are proposed to blindly detect multi-value signals. A discrete 4-level signum-type activation function is constructed. Optimization performance function is constructed to blindly detect signals directly, which does not rely on the second or higher order statistics of the received signals. New weight matrixes and energy function of multi-value DHNN are given. Theoretical proof is given of the stability of multi-value DHNN. Simulation results have shown that the proposed algorithms can converge to the real equilibrium points in a few iterations so that they are applicable to blindly detecting multi-value signals in stochastic channels.