Analytical and Computational Evaluation of Scalable Distributed Fusion Algorithms

Abstract

The theoretical fundamentals of distributed information fusion have been developed over the past two decades and are now fairly well established. However, practical applications of these theoretical results to dynamic sensor networks have remained a challenge. There has been a great deal of work in developing distributed fusion algorithms applicable to a network centric architecture. In general, in a distributed system such as ad hoc sensor networks, the communication architecture is not fixed. In those cases, the distributed fusion approaches based on pedigree information may not scale because of limited communication bandwidth. In this paper, we focus on scalable fusion algorithms and conduct analytical performance evaluation to compare their performance. The goal is to understand the performance of these algorithms under different operating conditions. Specifically, we evaluate the performance of channel filter fusion, naïve fusion, Chernoff fusion, Shannon fusion, and Bhattacharyya fusion algorithms. We also compare their performance to "optimal" centralized fusion under a specific communication pattern. The results show that the channel filter fusion, representing a first order approximation to the information graph fusion, is the only "consistent" fusion algorithm.

Fast Consistent Chernoff Fusion of Gaussian Mixtures for Ad Hoc Sensor Networks

“…where denominator is the common information pdf between and ; this follows from the factorization (4) where is the exclusive information at node (similarly for ). 'Optimal' exact Bayesian DDF thus requires explicit tracking and removal of the common information pdf from the product of and .…”

“…'Optimal' exact Bayesian DDF thus requires explicit tracking and removal of the common information pdf from the product of and . However, this requires heavy computational expense in networks without tree communication topologies [1], [2], [4]. As such, exact Bayesian DDF is generally infeasible in dynamic ad hoc sensor networks.…”

“…By using the concept of [1], the outage secrecy capacity with multiple eavesdroppers was studied in [3] for quasi-static Rayleigh fading channel where transmitter, legitimate receiver and each eavesdropper are equipped with single antenna. In [4], Gopala et al studied the secrecy capacity of block ergodic-fading channel. With the MRC diversity, the probability of positive secrecy capacity and the outage probability of secrecy capacity were studied in [5] for Rayleigh fading channel assuming all the diversity branches of main channel and eavesdropper channel are independent and identically distributed (i.i.d.…”

Delay estimation based on kinematic track information without time stamps

“…Lokal birimlerin kendi verilerini işlemeleri merkezi bir sistemde füzyon merkezinde harcanacak hesaplama yükünün bir kısmının bu birimlere aktarılmasını sağlayarak üst düzey füzyon merkezinin hesaplama yükünü azaltır [5]. Lokal düzeyde verinin işlenmesi sensörlerden gelebilecek pek çok yanlış alarmın lokal sistemde elenmesini sağlayacağından sistemdeki iletişim gereksinimlerinin de azalmasına yardımcı olur.…”

Efficient distributed sensing using adaptive censoring based inference

“…Graphical model based algorithms are only guaranteed to work well on acyclic networks, because in that case there is only one path between any two agents, which guarantees that the messages are not duplicated. For an arbitrary network, one needs to use approximate algorithms (e.g., [19][20][21][22][23][24]), or implement additional algorithms to restructure the network into an acyclic network [25], which brings in extra complexity.…”