Distributed Parameter Estimation in Wireless Sensor Networks Using Fused Local Observations

Mohammad Fanaei, Minnesota State University Mankato
Matthew C. Valenti, West Virginia University
Natalia A. Schmid, West Virginia University
Marwan M. Alkhweldi, West Virginia University


The goal of this paper is to reliably estimate a vector of unknown deterministic parameters associated with an underlying function at a fusion center of a wireless sensor network based on its noisy samples made at distributed local sensors. A set of noisy samples of a deterministic function characterized by a nite set of unknown parameters to be estimated is observed by distributed sensors. The parameters to be estimated can be some attributes associated with the underlying function, such as its height, its center, its variances in different directions, or even the weights of its specific components over a predefined basis set. Each local sensor processes its observation and sends its processed sample to a fusion center through parallel impaired communication channels. Two local processing schemes, namely analog and digital, are considered. In the analog local processing scheme, each sensor transmits an amplified version of its local analog noisy observation to the fusion center, acting like a relay in a wireless network. In the digital local processing scheme, each sensor quantizes its noisy observation before transmitting it to the fusion center. A at-fading channel model is considered between the local sensors and fusion center. The fusion center combines all of the received locally-processed observations and estimates the vector of unknown parameters of the underlying function. Two different well-known estimation techniques, namely maximum-likelihood (ML), for both analog and digital local processing schemes, and expectation maximization (EM), for digital local processing scheme, are considered at the fusion center. The performance of the proposed distributed parameter estimation system is investigated through simulation of practical scenarios for a sample underlying function.