电信科学 ›› 2016, Vol. 32 ›› Issue (3): 53-59.doi: 10.11959/j.issn.1000-0801.2016083

• 研究与开发 • 上一篇    下一篇

基于二维LMBP神经网络的ISM频段预测算法

万晓榆,胡盼,王正强   

  1. 重庆邮电大学,重庆 400065
  • 出版日期:2016-03-20 发布日期:2016-03-28
  • 基金资助:
    国家自然科学基金资助项目;重庆市基础与前沿研究计划基金资助项目;工业和信息化部软科学项目;重庆邮电大学博士科研启动基金资助项目;重庆邮电大学青年科学基金资助项目

Spectrum prediction algorithm in ISM band based on two-dimensional LMBP neural network

Xiaoyu WAN,Pan HU,Zhengqiang WANG   

  1. Chongqing University of Posts and Telecommunications,Chongqing 400065,China
  • Online:2016-03-20 Published:2016-03-28
  • Supported by:
    The National Natural Science Foundation of China;The Basic and Advanced Research Project of Chongqing;The Soft Science Project of Ministry of Industry and Information;Chongqing University of Posts and Telecommunications;The Science Research Project of Chongqing University of Posts and Telecommunications for Young Scholars

摘要:

随着短距离无线通信技术的快速发展及应用,ISM(2.4 GHz)频段的电磁干扰问题日益凸现,而利用频谱预测来预先获知频段的占用信息,已成为解决设备间兼容共存问题的有效途径。在验证ISM频段时域频域相关性的基础上,提出了一种时频二维LMBP神经网络,并将其应用于ISM频段的频谱预测。通过仿真和理论分析得到了最佳的时频训练组合点(△t=5、△f=2),在神经网络输入向量N=9的条件下,该点的预测准确度可达95%,相比Markov算法和时域LMBP神经网络分别提高了9%和4%的预测精度,且具有更优的训练收敛时间。

关键词: ISM频段时频相关性, BP神经网络, 时频二维LMBP神经网络, 频谱预测精度

Abstract:

With the rapid development and application of short-range wireless communications technology,the electromagnetic interference of ISM(2.4 GHz)band has become more apparent.Using the spectral prediction algorithm to predict the information of spectrum occupancy has become an effective way to solve the problem of compatible coexistence between devices.On the basis of verifying the time-domain and frequency-domain correlation of ISM band,an LMBP neural network of time and frequency domain was proposed and applied in the spectral prediction of ISM band.Through simulations and theoretical analysis,the best training combination of time-frequency point (△t=5,△f=2)was obtained.This point improves 95% of the spectrum prediction accuracy under the conditions of the input vector N=9 of the neural network.It increased 9% and 4% prediction accuracy compared with Markov algorithm and time-domain LMBP neural network and it had a better convergence time of training.

Key words: time-frequency correlation of ISM band, BP neural network, LMBP neural network of time and frequency domain, accuracy of the spectrum prediction

No Suggested Reading articles found!