Optimized distributed average consensus and its applications in cloud detectio

Distributed information processing has attracted more interests recently because of its advantages in distributed network than centralized method. Some of the tools of distributed signal processing are the distributed consensus algorithms. If the consensus algorithms are to find the average value over the network, the algorithms are called distributed average consensus (DAC) algorithms, which are matrix iterative algorithms to find the dominant eigenvector of the matrix. Generally, the DAC algorithms with optimized convergence rate can return the result more quickly so that the distributed system can have higher signal processing speed. Therefore, many efforts have been devoted into its optimization. However, most of the optimizations are centralized methods so that the system can not be optimized in a distributed network. In addition, the existing distributed optimization algorithm converges very slowly. Consequently, we proposed a distributed real-time optimization for the DAC algorithms. The optimization has advantages not only in short computation time but also can work simultaneously with the consensus algorithms. Simulation results show that the DAC algorithms using centralized optimization and proposed optimization have similar performance in convergence rate, when floating point number in double format is used and the network size is less than 32. Later, an application of cloud detection is presented where optimized DAC algorithms are applied to perform the hypothesis testing. In addition, the performance of the detection system with different number of sensors is evaluated using relative operating characteristic curves. It is shown that detection system with more sensors can have better performance.