Statistical Inference-Based Distributed Blind Estimation in Wireless Sensor Networks

To realize the Internet of Things, one of the essential elements is wireless sensor networks which can sense the physical conditions of the environment. The ubiquitous sensing is achieved by a large number of spatially dispersed sensors and distributed estimation technology. However, the low-cost sensors are insufficient to support conventional distributed estimation schemes. Since most conventional schemes include channel training process, the resource consumption of which is enormous. Thus, one key challenge in designing a feasible distributed estimation scheme is to reduce resource consumption from channel training. We tackle the challenge by proposing a distributed blind estimation scheme. The proposed scheme consists of two components: random transmission and statistical inference. Specifically, assuming sensors contain only two states that are active and inactive. The random transmission strategy turns the sensing value into a parameter to govern the sensor states. At the fusion center, statistical inference method is used to recover the sensing value. The specific design of the inference method involves the distribution approximation and clustering, which are accomplished by Gaussian mixture model and expectation-maximization principle. By the proposed scheme, the channel information is no longer needed in distributed estimation. Therefore, it is more energy-efficient and more applicable to the complicated wireless environment compared with conventional schemes. Besides, we investigate the impacts of the number of sensors and quantization on the estimation performance. Finally, simulation results demonstrate the effectiveness of the proposed blind estimation scheme.