Distributed Agent Consensus Performance in Resilient Communication Networks

This paper discusses distributed average consensus (DAC) problems for a set of dynamic agents within a wireless ad hoc communication network with specific emphasis on the performance of distributed average consensus in k-resilient ad hoc topologies. We present work that models and analyzes the following cases of interest: 1) the performance of DAC in static k-resilient networks, 2) the performance of DAC when local consensus state exchanges extend beyond direct neighbor diffusion, 3) the performance of DAC in mobile k-resilient topologies and random receiver-based link loss models, and 4) the use of a connected dominating set (CDS) within a distributed network as an additional DAC optimization. We demonstrate that DAC performance, measured as convergence rate or consensus delay, is significantly improved in mobile ad hoc network topologies with moderate k-resilience (e.g., k=2,4). We also demonstrate that the use of j-hop message diffusion in the DAC process can significantly reduce the consensus delay even when j=2. We present similar experiments using mobility and receiver loss models and demonstrate that similar DAC performance trends hold compared to those in the static k-resilient topology cases. We also briefly discuss the CDS adaptation of DAC and show that it further reduced consensus delay in the distributed wireless networks studied.