CSR--EHS: DEFT (Distributed Embedded Fault-Tolerant Control of Resource Constrained Sensor Networks)

真正的移动、自适应和自主的传感器网络，包括移动健康监测和环境监测系统，资源极其有限。对于扩展操作，这样的系统需要使用最少的电力、通信或计算资源的先进的实时控制方法。资源受限传感器网络的分布式嵌入式容错控制(DIFT)涉及到嵌入式无线传感器网络的实时分布式控制算法的开发，该算法受到体积、计算、功耗、存储和带宽的严格限制。DEFT使用一种新的信息论卡尔曼滤波(KF)方法来表示通过传感器提取的信息中的不确定性作为每个传感器的采样率和传感器观测之间的互相关的函数。这在分布式马尔可夫决策过程(DMDP)控制器中使用，以导出最佳采样率，同时确保不违反系统约束。DMDP内置了容错功能，可处理间歇性或永久性传感器降级。带有KF估计器的DMDP公式非常有效地处理和控制多个变量，包括传感器采样率和决定何时应与其他传感器交换信息(通过无线通信)。DIFT系统适用于使用传感器网络的一系列应用，从监控消防员、野战士兵、警察、糖尿病患者，到用于海洋、太空栖息地和战场多目标跟踪的环境监测传感器网络。该软件和算法有望在各种无功、低功率系统中具有潜力，包括用于建筑物和结构中的振动分析和补偿的无功传感器网络。

Truly mobile, adaptive and autonomous sensor networks, including mobile health monitoring and environmental monitoring systems, have extremely constrained resources. For extended operation, such systems require advanced real-time control methodologies with minimal usage of power, communication or computing resources. DEFT (Distributed Embedded Fault-Tolerant Control of Resource Constrained Sensor Networks) involves the development of real-time distributed control algorithms for use in embedded wireless sensor networks that are severely constrained by size, computation, power, storage and bandwidth.DEFT uses a novel information theoretic Kalman Filter (KF) approach to express the uncertainty in the information extracted through the sensors as a function of the sampling rate of each sensor and the cross-correlations between sensory observations. This is used within a distributed Markov Decision Process (DMDP) Controller to derive the optimal sampling rates while ensuring that system constraints are not violated. Fault tolerance is built into the DMDP to handle intermittent or permanent sensor degradation. The DMDP formulation with the KF estimator very efficiently handles and controls multiple variables, including sensor sampling rate and deciding when information should be exchanged with other sensors (via wireless communication). The DEFT system is applicable to a host of applications that use sensor networks, ranging from monitoring firefighters, soldiers in the field, police, diabetics, to environmental monitoring sensor nets for oceanography, space habitats, and multitarget tracking in battlefields. The software and algorithms are expected to have potential in a variety of reactive, low power systems, including reactive sensor networks for vibration analysis and compensation in buildings and structures.