Combining Channel Estimation and Sensor Fault Protection in Wireless Sensor Networks Si-Yao Huang ∗ , Chia-Lung Wu § , Po-Ning Chen ∗ , Tsang-Yi Wang † and Yunghsiang S. Han ‡ ∗ Dept. of Electrical Eng., National Chiao-Tung Univ., Taiwan, ROC Email: SYHuang0114@gmail.com, poning@faculty.nctu.edu.tw § Mobile Devices Inc., Taiwan, ROC Email: clwu@banyan.cm.nctu.edu.tw † Institute of Comm. Eng., National Sun Yat-sen Univ., Taiwan, ROC Email: tcwang@mail.nsysu.edu.tw ‡ Dept. of Electrical Eng., National Taiwan Univ. of Sci. and Tech., Taiwan, ROC Email: yshan@mail.ntust.edu.tw Abstract—The ongoing development of wireless sensor net- works (WSNs) demands not only low-power sensors and less system cost but also good performance. Considering this back- ground, investigating a new technology to satisfy both require- ments is an important issue for current development of wireless sensor network systems. In this paper, we consider the situation that the sensor nodes in WSNs are deployed in a harsh environment such that both channel fading and unexpected sensor faults may occur. This work then proposes to combine channel estimation and sensor- fault protection using error-correcting coding technique so as to free the costly devices of channel estimation and equalization from fusion centers. Simulations, performed to examine the performance of our proposed blind-detection scheme, show that it can compete with the conventional training-sequence-based fusion in performance when the training sequences are retained for information bearing. By this new approach, the complexity in fusion centers can be reduced and the transmitted power of sensors decreased without sacrificing the performance. I. I NTRODUCTION A wireless sensor network (WSN) contains a number of sensors to detect the environmental variation. The sensor decisions are then transmitted to the base station. The base station, serving as a fusion center, subsequently determines what phenomenon has occurred after the collection of local decisions from sensor nodes. In real applications, the system cost is often required to be inexpensive with acceptably good performance. How to simultaneously achieve these two goals is seemingly a challenge for the design of a WSN. When a WSN is deployed in a harsh environment, the wireless links between sensors and the fusion center may suffer channel fading. Since resources such as energy and computation power are often limited at sensor nodes, it turns out that only simple local classification can be implemented to combat such fading effect. In particular, when some sensor nodes are cornered at the margin of the target environment to be sensed, their local decisions could become nearly irrelevant to the true phenomenon. Sensor nodes are also prone to dam- age owing to random deployment. Placing them in a harsh or inaccessible area may make them irreplaceable. Consequently, one of the requirements for WSNs in certain applications is the capability to function robustly under unexpected sensor faults, which are ideally defined as that the local decision sent by the faulty sensor is independent of the true phenomenon [1]–[4]. It was then shown in [5], [6] that in channels without fading, the incorporation of the error-control coding technique to the design of WSNs can provide the desired robustness against sensor faults that are unaware to the fusion center. One technique that can perhaps be used to prevent the performance degradation from faulty senor nodes is to transmit a sequence of training bits for faulty sensor detection, after which the decisions sent from faulty nodes could be ignored at the fusion center. In an environment with channel fading, these training bits could also serve their conventional role, i.e., being the basis for channel estimation. Nevertheless, the training- sequence-based technique can capture only the abnormality of local transmission modules but not the failure of computation modules or the inconsistent sensing of a node. Hence, when the local classification of a sensor happens to be independent of the phenomenon, its local decision that sends after a sequence of correctly transmitted training bits still degrade the global decision at the fusion center. Alternative technique is therefore required to prevent the performance degradation from such faulty nodes. Recently, some works for point-to-point communications [7]–[10] proposed to use computer-searched signals (codes or constellations) for combined channel estimation and error pro- tection in quasi-static fading channels. After the establishment of a search criterion, their signals were then searched by using the random search technique or gradient descent method. They found under the presumption of a fixed effective code rate that the best signals result in apparently better performance than a benchmark system with a certain number of training bits. Motivated by these works, the codes that combine channel estimation and sensor fault protection are developed in this pa- per. Upon the reception of coded transmissions from sensors, the fusion center performs the optimal blind detection about the true phenomenon. Simulations showed that without the effort of channel estimation and equalization, the developed