无线传感器网络（Wireless Sensor Network, WSN）由大量嵌入式无线传感器节点构成，节点具有一定的感知、计算、存储和通信能力。无线传感器网络的基本功能是将源节点采集的感知数据传输到sink节点。其应用领域包括灾难救助、环境监控、智能楼宇、建筑安全等。在无线传输过程中，数据可能由于信号衰减、外界干扰、网络拥塞等原因而丢失，在资源受限的传感器网络中尤为严重。如何保障数据在传输过程中的可靠性一直是研究人员关注的重点。本文围绕传感器网络数据传输可靠性目标，分析了已有研究的缺陷和不足，结合典型的应用场景，针对不同的服务质量需求，研究并提出了多项数据传输可靠性保障机制和系统。本文的创新性工作和贡献如下：

1. 应急场景下基于无线传感器网络的语音通信系统研究：设计了高质量语音收集系统QVS和语音分发系统ASM，提出了多种新颖的可靠性保障机制，包括基于树状拓扑结构的机会路由、基于冗余分组和前向纠错（Forward Error Correction，FEC）的丢包恢复机制、基于本地剩余容量估计的分布式准入控制机制。在语音通信节点SenEar组成的测试平台上，实现和分析了提出的机制和系统，并进行了深入的实验，验证了系统能够收集和分发高质量的语音流。

2. 城市环境下车载感知信息收集过程中的转发策略研究：利用不同车辆的移动规律，提出了车辆间数据转发的优化机制，提高了数据转发的成功率。设计了一种根据车辆日常行驶规律预测车辆移动趋势的方法，提出基于移动趋势预测优化下一跳车辆选择的转发算法Seer；根据公交车行驶轨迹固定的特点，提出了基于公交车线路间相遇频率的数据转发机制R2R。基于生成和真实的trace数据对Seer和R2R机制进行了模拟仿真，验证了所提出机制能够显著提高数据传输的可靠性。

3. 大范围、高精度、能量高效的时钟校准方法研究：在带宽严格受限的情况下，时间同步是提高带宽利用率，实现大数据流可靠传输的前提之一。本文提出了一种基于FM广播数据系统（Radio Data System, RDS）的时钟校准协议ROCS，该协议利用RDS信号中带有的周期性，校准节点自身的晶体钟。在Senshoc节点组成的原型系统上测量并验证了ROCS协议，证明其具有能量高效、精度高、覆盖性好等优点。

Wireless Sensor Networks (WSN) consist of a large number of embedded sensing devices, namely nodes. Each node has a certain capability of sensing, computing, storage, and communication. Recent years have witnessed WSNs in various applications such as disaster recovery, environment surveillance, intelligent building, structure monitoring, and etc. WSN typically transmits packetized sensing data from source nodes to sinks, during which packets may get lost due to signal attenuation, fading, interferences, and congestion, especially for resource constrained sensor networks. Consequently, reliable data transmission attracts attentions from the research community. In this thesis, we focus on improving reliability during data transmission. We review existing work and point out their limitations, which motivates us to investigate the following topics: 

1. WSN-based voice communication system under emergency: we propose two systems, i.e., QVS for voice gathering and ASM for voice dissemination. Both systems are composed of a number of novel schemes to ensure reliability, which include a tree-based opportunistic routing protocol, loss concealment schemes based on duplicate packets or Forward Error Correction (FEC), a distributed admission control algorithm based on local capacity. We evaluate QVS and ASM on our hardware testbed with SenEar nodes. The results show our proposed systems can deliver quality-assured voice streams.

2. Vehicle-based information gathering in metro-areas: we optimize data forwarding process by exploiting vehicular mobility models. With the intuition that personal cars usually take repeat trips, we design a forwarding method named Seer based on driving trend prediction; we also propose a data forwarding protocol called R2R for bus-based networks on the basis that buses have fixed routes. We evaluate Seer and R2R with traces in a simulator. The results show Seer and R2R can significantly improve the reliability.

3. Large-scale, highly accurate, energy efficient clock calibration: time synchronization is essential for reliable bulk data transmission especially for resource-constrained sensor networks. We propose to exploit the FM Radio Data System (RDS) as an external clock source to calibrate the native crystal oscillator for each sensor node. We build our proto-type system with Senshoc nodes and show our proposed algorithm can achieve large-scale, highly accurate, and energy efficient clock calibration.