A Framework for Real-Time Collection, Analysis, and Classification of Ubiquitous Infrasound Data

Abstract:

Traditional infrasound arrays are generally expensive to install and maintain. There
are ~10^3 infrasound channels on Earth today. The amount of data currently
provided by legacy architectures can be processed on a modest server. However, the
growing availability of low-cost, ubiquitous, and dense infrasonic sensor networks
presents a substantial increase in the volume, velocity, and variety of data flow.
Initial data from a prototype ubiquitous global infrasound network is already
pushing the boundaries of traditional research server and communication systems,
in particular when serving data products over heterogeneous, international network
topologies. We present a scalable, cloud-based approach for capturing and analyzing
large amounts of dense infrasonic data (>10^6 channels). We utilize Akka actors
with WebSockets to maintain data connections with infrasound sensors. Apache
Spark provides streaming, batch, machine learning, and graph processing libraries
which will permit signature classification, cross-correlation, and other analytics in
near real time. This new framework and approach provide significant advantages in
scalability and cost.