Daniel Gilio Tiglea. Um algoritmo adaptativo de baixo custo computacional para amostragem e censura em redes de difusão. Master’s Dissertation, Electrical Engineering, University of São Paulo, 2020 (in portuguese)

Abstract

In recent years, diffuse adaptive networks and graph adaptive filters have become topics of strong interest in the signal processing community. Diffuse adaptive networks have consolidated themselves in the literature as interesting tools for distributed signal processing, presenting advantages over centralized solutions and other diffusion techniques. Graph adaptive filters, in turn, have attracted attention for their ability to deal with situations in which there are large amounts of data that are related through irregular structures. In both cases, techniques have been proposed to reduce the amount of information measured and transmitted over the networks, which enables a reduction in the computational cost and in the energy consumption. Such techniques usually affect the performance of the original solutions, but are important for extending the network lifetime. In this work, we propose an adaptive sampling mechanism for distributed adaptive solutions and graph adaptive filters. The proposed sampling algorithm uses more nodes when the magnitude of the error throughout the network is high and less nodes otherwise. Thus, a significant reduction in computational cost is achieved while the impact on performance is mitigated. It is also shown that, with a small modification, it can be used to reduce the number of transmissions between nodes, enabling a reduction in energy consumption. Furthermore, we conduct a theoretical analysis of the proposed mechanism, which enables a better understanding of its functioning and allows more suitable choices for its parameters.

Downloads

• PDF