Tuning of Automatic Signal Detection Algorithms for IMS Style Infrasound Arrays

摘要

Design specifications for infrasonic arrays that comply with IMS requirements stipulate at least 4 sensors in a centered triangle configuration with maximal separation between sensors of between 1 to 3 km. Such a design is considered to be optimal for the detection of infrasound signals from explosive sources with yield down to around 1kT that are within a distance of around 3000km from the receiver. Automatic algorithms used for signal detection need to be optimized for both the signals of interest and the particular geometry of the IMS style array. A research program nearing completion at the Center For Monitoring Research has been designed to produce automatic algorithms that are optimal with respect to signals and array geometry. Synthetic waveforms generated using Pierce's normal mode algorithm, designed to be representative of the infrasound signal from a 1-kT blast several thousand km from the receiver, were used to establish the Receiver Operator Curves (ROC curves) for the infrasonic signal detection algorithm in use at the Prototype International Data Centre (PIDC) in Arlington Virginia. The synthetic signals were implanted into conventional array channel data using an implant strategy that allowed the signal to noise ratio (SNR) of the signal to be specified in a fashion that accounts for the spectral composition of the background. SNR values ranging from 0.01 to 100 were used in the study to map out the ROC curves. The PIDC infrasonic detection algorithm is known as a 'coincidence-detector' which relies on both the Fstat (Fischer Statistic), and the ratio of the L1 norms of a Short Term Average (STA) data window to that of a Long Term Average (LTA) data window, exceeding pre-determined threshold values simultaneously before a detection is declared.