前馈神经网络复值盲均衡算法的研究

The Research of the Complex-valued Blind Equalization Algorithm Beased on the Feedforward Neural Network

提出了一种新的前馈神经网络(N-FNN)复值盲均衡算法。新算法改变了传统均衡技术大量发送训练序列而降低系统传输的有效信息率,有效地消除码间干扰,提高了通信质量。笔者设计出新的传递函数和代价函数,利用最陡梯度下降法推导出输出层和隐层单元权值的迭代公式。通过对QAM信号进行计算机仿真,笔者提出的新算法与同类算法相比,具有均方误差收敛速度加快、误码率降低。

In this paper, a new complex--valued blind equalization algorithm based on the feedforward neural network （N--FNN） is proposed. Because the traditional equalization technology requests to transmit the training sequence constantly, the effective information rate of the system is low. The new algorithm has changed this status. It can eliminate Intersymbol Interference（ISI） effectively and improve the communication quality. The new transmission function and cost function are designed in this paper. At the same time, weight iteration formula of the output layer and hidden layer are deduced using the steepest descent method . Results of the simulation for QAM signals show that the proposed algorithm has faster mean square error convergence speed, and lower bit error rate, smaller steady state mean square error than the other similar algorithms.