Multi-Rate Asynchronous Distributed Filtering under Randomized Gossip Strategy

In this paper, we consider the extension of synchronous distributed filtering under randomized gossip strategy to multi-rate asynchronous sensor network. To deal with asynchronous measurements due to multi-rate sampling, two processing schemes are proposed, i.e., a batch one and a sequential one. In batch processing, the common fusion period is chosen as the least common multiple of the sampling periods of all sensors. Each measurement of a sensor is propagated to the nearest common fusion time and saved locally. At the common fusion time, multiple propagated measurements of a sensor are compressed to a single measurement and then exchanged with its neighbors. Whereas in sequential processing, the common fusion period is chosen as the greatest common divisor of the sampling periods of all sensors to sequentially process each measurement once it is received. Depending on whether a real measurement is available locally, three exchanging strategies are developed. It is found that unlike in the traditional distributed fusion with a common fusion center, the batch and sequential processing schemes are not equivalent. Pros and cons of these two schemes are analyzed. Numerical experimental results further verify the effectiveness of the proposed schemes.