Robust Dynamic Corrective Consensus over Lossy Networks with Gilbert-Elliott Channels

In this paper we propose a distributed algorithm to track the average of time-varying reference signals at nodes (i.e. dynamic average consensus) over networks with heterogeneous, asymmetric Gilbert-Elliott (GE) type lossy channels. The algorithm is robust to GE type network disruptions, (re)-initialization errors and agents joining, faulty or leaving the network during time. Necessary and sufficient conditions are formulated to assess almost sure and mean bounded dynamic average consensus according to the characteristics of reference signals. The proposed conditions can be used for design purpose as they relate algorithm step size, network connectivity and channel quality parameters (e.g. recovery and failure rate) to the algorithm convergence. Simulations confirm the effectiveness of the proposed algorithm and design conditions in two representative validation scenarios.