The Impact Of Mobility And Non-ideal Channel On Gossip Algorithms

Motivated by applications to sensor, peer-to-peer, and ad hoc networks, westudy distributed algorithms, also known as gossip algorithms, are widelyresearched for exchanging information and computing in an arbitrarily connectednetwork of nodes. Since require no specialized routing, gossip algorithms areattractive for distributed signal processing in wireless sensor networks. There is nobottleneck or single point of failure, and they are robust to unreliable wirelessnetwork conditions.This paper proposes an improved broadcast gossip algorithm with betterconvergence accuracy. In this proposed algorithm, we utilize a series of companionvariables to maintain the sum as much as possible in every iterative step. In thesimulation, we can see that convergence error of the algorithm is decreasingobviously with increasing a companion variable. In a practical application, thenumber of companion variables can be chosen freely according to the requirementof convergence precision and communication cost. So our algorithm seems to bemore practical than other gossip algorithms.In the next, we analyze the convergence, the error of convergence and the rateof convergence for pairwise gossip algorithms and broadcast gossip algorithms instatic networks and mobile networks with intermittent links. Since we don’t needthat each node is activated in i.i.d. and each link has identical distribution, theresearch results are more suitable for application. By study, we prove that variousgossip algorithms can converge in static networks and mobile networks withintermittent links as long as the digraph is jointly connected. In addition,simulations illustrate that intermittent links make gossip algorithms convergeslowly and movement can help gossip algorithms to converge fast.Finally, in order to reveal the practical application value of our algorithm, weanalyzed the performance in the non-ideal network model. Considering the limitedcomputing power of sensor nodes and narrow channel bandwith, the quantitativeanalysis of data is added on signal processing in each node. By the analysis andsimulations, we find that our algorithm can also converge and maintain goodperformance. In summary, we believe that the proposed algorithm has a goodpractical application value.