Evaluation of asynchronous average consensus algorithms in pure broadcasting infrastructure-free networks

Distributed and cooperative algorithms are of preponderant importance for the correct operation of multiagent systems. In particular, average consensus algorithms represent an appealing alternative for combining measurements in large-scale networks of low-capable sensors, due to their low computational cost and strong convergence properties. However, the actual performance of average consensus algorithms in real scenarios, where the interaction between agents involves a communication network introducing stochastic delays, sequential transmissions and receptions, and unreliability in the information exchanging process, is yet to be investigated. This work presents an evaluation on a pure broadcasting infrastructure-free sensor network of two popular average consensus strategies: the broadcast gossip algorithm (which can be regarded as an asynchronous version of the discrete-time average consensus algorithm), and the push-sum algorithm (also known as double linear iterations). To understand the operating principles behind the algorithms, a hybrid model is first introduced that is used to conduct numerical simulations. An implementation in microprocessor-based development boards is then presented to evaluate the performance in a real environment. Results of the evaluation show that the push-sum algorithm outperforms the broadcast gossip algorithm for practical values of the reception probability.
Heuristic algorithms, Sensors, Convergence, Algorithm design and analysis, Mathematical model, Broadcasting, Stochastic processes