摘要
考虑到BP网络易于陷入局部最小以及收敛速度慢的缺点,本文应用改进型BP网络算法自动识别无线电信号的调制类型。通过增加动量项可以避免陷入局部最小,但付出了训练时间长的代价;选择权向量的更新方向为误差函数的共轭梯度方向,可以显著地提高收敛速度。仿真结果表明,改进后的BP网络方法能够高效、可靠地收敛,实现通信信号的自动分类,获得满意的识别率。
A modified BP network to recognize the modulation type of communication signals automatically is presented in the paper, it is common that the shortcomings, such as falling into the local minimum and converging slowly, are caused by BP network. A piece of momentum is added to keep the correct convergence in adjusting the weight at the expense of long training time. In order to reduce the training time, the weight vector is changed along the conjugated grads of square error function. Better results have been achieved by simulation.
出处
《电路与系统学报》
CSCD
北大核心
2006年第2期136-140,共5页
Journal of Circuits and Systems
基金
预研基金资助项目(41501130304)
