摘要
为了克服传统频谱感知的缺点,提升在低信噪比下的频谱检测性能,提出了一种改进的量子神经网络频谱感知算法。通过提取授权用户信号的特征参数,对量子神经网络进行训练,获取授权信号中数据的不确定性并加以存储、记忆,从而实现周围环境"频谱机会"的检测。为了提高量子神经网络的收敛性、稳定性,对算法进行改进,采用三层Josephson函数作为激励函数,缩短激励函数的饱和区,减少训练过程中出现"假饱和"现象;并在原有的学习目标函数中加入约束条件,使网络权值的调整和量子间隔的更新在学习过程中的相互影响降到最低。通过实验仿真得出,改进后的量子神经网络算法与改进前的算法、BP神经网络检测算法相比,不但在网络收敛速度和稳定性上有了明显提升,而且在低信噪比情况下具有更高的检测概率。
To overcome the shortcomings of traditional spectrum sensing and enhance performance at low SNR,this paper proposes a sensing algorithm based on improved quantum neural network.The basic idea is to extract characteristic parameters by an authorized user signals and train quantum neural network,then to access authorization data signals uncertainty and store ,to achieve ambient ＂spectrum opportunity＂ test.ln order to enhance the convergence and stability of quantum neural network ,the quantum neural network is improved. The new algorithm chooses the three-layer Josephson function as transfer function to shorten excitation of the saturation zone and reduce the ＂false saturation＂ phenomenon occurred during training; With constraints added in the original learning objectives functions, the interaction of network weights adjustment and updating quantum in during learning process decreases to a mlnimum. As a result, the experiment results show that the improved quantum neural network has a faster convergent speed and a higher stability compared with the quantum neural network and BP neural network.Also,the improved QNN has a hi^her detection probability at the low SNR environment.
出处
《无线电通信技术》
2015年第2期7-11,共5页
Radio Communications Technology
基金
国家自然科学基金项目(60902046)
