短波认知ALE系统中基于深度学习-GP混合模型的多维谱预测方法研究

项目名称： 短波认知ALE系统中基于深度学习-GP混合模型的多维谱预测方法研究

项目编号： No.61461013

项目类型： 地区科学基金项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 唐智灵

作者单位： 桂林电子科技大学

项目金额： 45万元

中文摘要： 在短波认知ALE系统中利用多维谱预测不仅能够提高频谱利用率,还能支持ALE将频谱空洞进行聚合，根据短波通信业务的需要灵活地为用户分配带宽。本项目拟研究短波ALE中基于深度学习-高斯过程（GP）混合模型的多维谱预测新方法，能够预测下一时段内的空间谱空洞和频谱空洞。我们拟在以下四个方面展开研究。首先，研究短波ALE系统中多维谱预测深度学习-GP混合模型结构；其次，研究深度学习-GP混合模型的多维谱预测算法性能；第三，研究预测模型的预测长期状态的性能；第四，研究短波ALE系统中基于预测结果的多跳通信链路建立方法，分析其提高频谱利用率的作用。该研究得到的多维谱预测方法实现了深度学习与高斯过程的优势互补，可应用于下一代宽带短波ALE系统中提高频谱利用率并保证短波业务传输的质量。

中文关键词： 认知无线电；短波信道；神经网络模型；协作频谱预测；天波传播

英文摘要： Spectrum prediction in cognitive HF automatic link establishment(ALE) system not only can improve the spectral efficiency, but also supports aggregation of Spectrum Hole finished by ALE. Then ALE can assign bandwidth flexibly to user according to the needs of HF communications services.In this project, we will research a novel method of multidimensional spectrum prediction based on deep learning-Gaussian process hybrid model which can pridict space spectrum hole and frequency spetrum hole in next time duration.We will launch the research from following four aspects. Firstly,we will research the architecture of deep-learning - Gaussian process hybrid pridiction model in HF ALE system.Secondly, the performaance of the prediction model put forword by us will be researched. Thirdly, the peformence of our pridiction model for long time status will be researched. Lastly, the method of building multi-hop link by HF ALE based on prediction results and its improvement for spectrum efficiency will we researched.The spectrum pridiction method obtained from this project combine the relative sthenths of both deep learing and Gaussian process regression, which are helpful to improve spectral efficiency and can guarantee the quality of HF service in next generation wideband HF ALE system.

英文关键词： Cognitive Radio;HF Channel;Neural Network Model;Cooperative Spectrum Prediction;Sky-wave propagation