论文部分内容阅读
信道信息作为刻画无线信道状态的重要信息,在新一代基于大规模多天线技术的系统中扮演了极其重要的角色.而在超蜂窝架构中,业务基站需要动态休眠以提高系统的能效.当基站休眠时,传统的基于导频的信道信息获取方法无法使用.本文针对这一问题提出了基于机器学习的信道信息获取策略,通过活跃基站多维度的信道信息来推测休眠基站的信道信息.我们以用户选择业务基站为例,利用人工神经网络架构设计了一套具体的信道学习解决方案.与位置辅助信道估计方案不同,该方案采用控制基站侧的信道信息作为输入,避免了用户定位过程中存在的一些问题.通过基于随机地理散射物的信道模型,我们对该方案进行了仿真验证,发现其预测准确率优于K近邻等其他预测算法,性能接近于已知准确用户地理信息的位置辅助估计方案.
Channel information, as important information to describe the status of wireless channel, plays an extremely important role in the new generation of large-scale multi-antenna technology-based system, and in the super-cellular architecture, the service base station needs to sleep dormant to improve the energy efficiency of the system.When the base station At the time of dormancy, the traditional pilot-based channel information acquisition method can not be used.In this paper, a machine learning-based channel information acquisition strategy is proposed to predict the channel information of the dormant base station by active multi-channel channel information of the base station. The user chooses the base station of the business as an example, has designed a set of concrete channel learning solution by artificial neural network framework. Different from the location assisting channel estimation scheme, this scheme adopts and controls the base station side channel information as the input, has avoided the user localization process exists We simulate the scheme based on the channel model based on stochastic geographic scatterers and find that the prediction accuracy is better than other prediction algorithms such as K-nearest neighbor and the performance is close to the location-aided estimation of the known accurate geographic information of users Program.