Research On Distributed Adaptive Localization Algorithms

Passive localization of radio emitters based on time difference of arrival(TDOA) is an essential topic in signal processing. This technique has already been widely used in many practical engineering fieleds including radar, sonar, navigation, remote sensing and wireless sensor networks(WSN).TDOA localization can be classified into two-step methods and direct position determination(DPD) methods based on whether they need to estimate the TDOA values. Compared to the two-step methods, the DPD methods have better localization accuracy because it considers the constraint that all TDOA estitations must be consistent to a common emitter position. However, due to the grid searching model, the traditional maximum likelihood DPD(ML-DPD) algorithm suffers huge computational load. Compared to ML-DPD, the recently proposed adaptive direct position determination(ADPD) algorithm significantly reduces the computational complexity with only sacrificing little localization accuracy.With the development of sensor networks and the distributed signal processing theory, using a sensor network to locate and track a radio emitter has become a research hotspot in recent years. However, due to the centralized signal processing framework, the adaptive direct position determination method is unsutaible for large scale sensor networks. The main contribution of this article is to propose a distributed adaptive direct position determination of emitters using a sensor network based on the distributed adaptive theory. The details of the contribution in this article are summarized as following.Firstly, on the basis of centralized adaptive direct position determination, a partially distributed adaptive direct position determination algorithm using the well-known diffusion framework is proposed. With sacrificing little localization accuracy, the proposed approach distributes the computational load among each computing sensor in the sensor networks, which successfully solves the problem of huge computational load in a single sensor of the centralized counterpart.Secondly, the cost function argitecture of the centralized adaptive direct position determination is reconstructed, which makes it more fully use of the received signal samples. Based on this new cost function, a fully distributed adaptive direct position determination is proposed. This fully distributed method could not only avoid the multi-hop data transmission in centralized algorithm, but also improve the localization accuracy.Finally, the cost function of the fully distributed adaptive direct position determination is further analysed, which finds that the adaptive gain control schem could not commendably decouple the effect of the noise in received signals. Based on this, a novel cost function based on the idea of unbiased LMS algorithm is put forward, followed by an improved fully distributed algorithm proposed. Computer simulations demonstrate that this improved algorithm could not only improve the localization accuracy in colored signals situation, but also has the same localization accuracy with the original fully distributed counterpart.