传感器网络中分布式盲信号估计研究

Recently, the study on sensor networks has become a new research hot topic and received more and more attention from industrial, academic and military communities world-widely, due to their wide potential applications. Distributed processing is a major tenet of sensor networks. By exploiting local computation resources at each node, it is possible to reduce the amount of data that needs to be transmitted over the sensor networks, thereby saving bandwidth and energy, extending the network lifetime, and reducing latency. Besides, it is of high scalability and robustness. Owing to these merits, distributed in-network processing has been widely used in many areas, ranging from precision agriculture to environmental monitoring and transportation. Currently, distributed in-network processing has aroused considerable interest of the researchers both in scientific and engineering field. In the traditional distributed in-network processing, for example, in the context of distributed demodulation or distributed power spectrum estimation, in order to get the estimate of the source signal or its power spectrum, we need to know the communication channel in prior. To achieve this, we must transmit a series of training data at first. Then, based on the training data and the desired output, we can get the estimate of the channel model. However, the use of training data may cause many problems, such as more data communication, more bandwidth and energy consumption. So, it is more important and meaningful if we can estimate the signal of interest only based on the information of the received signal and some stochastic properties of the input by resorting to some blind signal processing technique, which is termed as distributed blind signal estimation. On the other hand, in the context of blind signal processing, the existing methods are performed either based on the collection of a single sensor without cooperation or that of multiple sensors using the centralized processing. So far, the research on distributed blind signal estimation has not yet been reported. In the light of current development, in this project, we would like to establish the framework of distributed blind signal estimation based on a combination of distributed processing and blind signal processing, study some key problems in distributed blind signal estimation over sensor networks and also try to deal with some related practical engineering problems.

传感器网络是目前世界各国学术界、工业界和军界共同关注的研究热点。分布式处理是传感器网络进行信息处理的一种重要方式。它具有能耗低、灵活性好、可靠性高、鲁棒性强的特点，在许多领域具有重要应用，因而激起了科学和工程领域研究者极大的研究兴趣。在传统的分布式处理中，如在信号解调或信号功率谱估计中，我们通常需要预先传输一定的训练序列以获得信道模型。但训练序列的使用具有明显的不足。如果能够在仅给定观测信号和输入信号的一些统计特性的情况下，分布式地估计出我们感兴趣的信号，即分布式盲信号估计，必然具有更重大的科学意义和更广泛的应用价值。另一方面，已有的盲信号估计要么是基于单节点的处理，要么是基于多节点的集中式处理。迄今为止，国内外还鲜见报道有关分布式盲信号估计的研究。鉴于国内外研究现状，本项目研究拟建立一套分布式盲信号处理的理论，研究分布式盲信号估计中的一些关键问题，为一些工程问题的解决提供坚实的理论基础。

分布式处理是传感器网络进行信息处理的一种重要方式，在许多领域具有重要应用，因而激起了科学和工程领域研究者极大的研究兴趣。在传统的分布式处理中，如在信号解调或信号功率谱估计中，我们通常需要预先传输一定的训练序列以获得信道模型。但训练序列的使用具有明显的不足。如果能够在仅给定观测信号和输入信号的一些统计特性的情况下，分布式地估计出我们感兴趣的信号，即分布式盲信号估计，必然具有更重大的科学意义和更广泛的应用价值。. 本项目研究正是从这一点出发，针对分布式场景下盲信号处理中的一些关键问题进行研究，包括分布式盲信号估计、分布式盲信道均衡、分布式稀疏盲信号估计、分布式多任务学习、安全的分布式估计、分布式非线性系统预测、分布式盒形粒子滤波、分布式机器学习等几个部分进行了深入的研究，建立了一套分布式盲信号处理的理论。具体研究路线是，首先通过考虑待估计的盲信号的常模特性、稀疏性等，融合邻居节点信息设计可去中心处理的代价函数，采用优化算法推导出一系列分布式盲信号估计、分布式盲信道均衡、分布式稀疏以及联合稀疏盲信号估计的算法。其次，充分考虑节点之间任务不同但相似的特点，提出了一些基于自适应耦合权重的多任务分布式学习算法。再次，考虑到网络可能受到攻击的情况，提出了安全的分布式估计算法，保证了分布式合作的可靠性。然后，考虑到线性模型的局限性，提出了分布式非线性预测模型，能够更好地对复杂的动力学系统进行建模和预测。最后，作为本项目研究的后续扩展，针对分布式场景中一些机器学习问题，包括半监督学习、主动学习等进行了研究。本项目依照研究计划执行，各项研究内容进展顺利。本项目丰富和发展了分布式盲信号处理的理论，并且为一些工程问题的解决如信号谱估计、通信信号恢复、目标跟踪、时间序列预测、图像处理等提供坚实的理论基础。

内容获取失败，请点击重试