Decentralised seismic tomography computing in cyber-physical sensor systems

This paper presents an innovative decentralised seismic tomography computing paradigm in cyber-physical sensor systems, where each sensor node computes the tomography based on its partial information and through gossip with local neighbours only. The key challenge is the potential high communication overhead due to limited knowledge of each node about the entire network topology and information. The aim of this paper is to develop efficient algorithms for maximising the quality of tomography resolution while minimising the communication cost. We reformulate the conventional seismic tomography problem and exploit the alternating direction method of multipliers method to design two distributed algorithms. One is a synchronous algorithm and the other is asynchronous and more fault-tolerant and scalable. We theoretically prove that both proposed algorithms can reach their convergent solutions in a linear rate in terms of the number of communication rounds. Extensive evaluations on both synthetic and real data-sets validate the superior efficiency of the proposed algorithms. They not only achieve near-optimal (compare to centralised solution) high-quality tomography but also retain low communication cost even in sparse networks.