Distributed averaging in wireless sensor networks with triplewise gossip algorithms

This paper proposes triplewise gossip algorithm (TGA), a novel gossip algorithm for fast convergence in wireless sensor networks. During the operation of TGA, the active node selects two of its neighbors to perform triplewise averaging at each iteration. Compared with pairwise gossip algorithms, more transmissions are required at each iteration, however the total number of radio transmissions is reduced because fewer iterations are needed to get to convergence. The convergence rate of TGA is studied theoretically and verified by simulations. Similar to pairwise gossip algorithms, the proposed algorithm is a fundamental gossip algorithm. Many optimization methods that are designed for pairwise gossip algorithms can also be applied to triplewise gossip algorithms and would usually achieve better performance. Greedy TGA is presented as an improvement of the proposed algorithm, which is proved to be able to significantly accelerate convergence by numerical simulation results.