Network Topology Optimization for Accelerating Consensus Algorithms under Power Constraints

Abstract

The average consensus algorithm is a well known distributed process in which the nodes iteratively communicate with the nodes within their communication range in order to obtain an estimation of the global average. These repeated communications, when performed in a uniformly randomly deployed network, such as a Wireless Sensor Network, lead to several nodes consuming much more power than others, thus reducing the lifetime of the whole network. This paper proposes a fully distributed method that allows the network nodes to suitably decide which subset of communications provides the best performance during the consensus process in terms of convergence time and power efficiency. Our method simultaneously improves both the convergence of the consensus algorithm and the lifetime of the whole network. Moreover, as a benchmark, we propose a convex optimization problem whose results can be compared with those obtained by our distributed approach. Simulation results are presented to show the efficiency of our proposal, comparing our two methods with existing approaches in the related literature.

Distributed Two-Step Quantized Fusion Rules Via Consensus Algorithm for Distributed Detection in Wireless Sensor Networks

“…We will also show via simulations that there is an optimum K 1 (for both decision fusion (37) and (38)) such that the system detection performance is maximized. Then, we propose a sub-optimum but simple solution to find this optimum K 1 .…”

“…In [36] a simplified relaxed (centralized) solution was presented, where the energy minimization problem was formulated as a convexconcave fractional programming. Another approach was followed in [37], where a distributed algorithm to decide which subset of communication links provides the optimum power consumption and the best network lifetime (i.e., minimizing simultaneously both the total power consumption and the maximum power consumption per SN) was developed. While both ( [36] and [37]) improve the total power consumption and/or the whole network lifetime, they also assume that the exchange of information among SNs is ideal.…”

“…Another approach was followed in [37], where a distributed algorithm to decide which subset of communication links provides the optimum power consumption and the best network lifetime (i.e., minimizing simultaneously both the total power consumption and the maximum power consumption per SN) was developed. While both ( [36] and [37]) improve the total power consumption and/or the whole network lifetime, they also assume that the exchange of information among SNs is ideal. But here we propose to quantize with q i bits at SN i before transmitting to SN j and to satisfy the capacity constraint between SNs i and j we require:…”