Event-Based Statistical Signal Processing

In traditional time-based sampling, the sampling mechanism is triggered by predetermined sampling times, which are mostly uniformly spaced (i.e., periodic). Alternatively, in event-based sampling, some predefined events on the signal to be sampled trigger the sampling mechanism; that is, sampling times are determined by the signal and the event space. Such an alternative mechanism, setting the sampling times free, can enable simple (e.g., binary) representations in the event space. In real-time applications, the induced sampling times can be easily traced and reported with high accuracy, whereas the amplitude of a time-triggered sample needs high data rates for high accuracy. In this chapter, for some statistical signal processing problems, namely detection (i.e., binary hypothesis testing) and parameter estimation, in resource-constrained distributed systems (e.g., wireless sensor networks), we show how to make use of the time dimension for data/information fusion, which is not possible through the traditional fixed-time sampling.