William G. Frazier

Enhancing real-time 3D tracking of acoustic targets using elevated microphones and environmental parameter effects

Networks of unattended acoustic ground sensors can be used to detect and accurately track high-speed airborne acoustic sources. While it is possible, in principle, to estimate altitude using networks of two-dimensional microphone arrays, the high sensitivity of this configuration significantly limits performance. This work shows that the addition of elevated microphones and appropriate signal processing can significantly improve performance. Airborne source tracking results from a field experiment are compared for the use of a small orthogonal three-microphone ground plane array versus the same array with an elevated microphone added.

Acoustic detection, tracking, and characterization of three tornadoes

Acoustic data recorded at 1000 samples per second by two sensor arrays located at ranges of 1-113 km from three tornadoes that occurred on 24 May 2011 in Oklahoma are analyzed. Accurate bearings to the tornadoes have been obtained using beamforming methods applied to the data at infrasonic frequencies. Beamforming was not viable at audio frequencies, but the data demonstrate the ability to detect significant changes in the shape of the estimated power spectral density in the band encompassing 10 Hz to approximately 100 Hz at distances of practical value from the sensors. This suggests that arrays of more closely spaced sensors might provide better bearing accuracy at practically useful distances from a tornado. Additionally, a mathematical model, based on established relationships of aeroacoustic turbulence, is demonstrated to provide good agreement to the estimated power spectra produced by the tornadoes at different times and distances from the sensors. The results of this analysis indicate that, qualitatively, an inverse relationship appears to exist between the frequency of an observed peak of the power spectral density and the reported tornado intensity.

Application of extended Kalman filtering to tracking of airborne high‐speed broadband acoustic sources

A technique for tracking airborne high‐speed (subsonic) broadband acoustic sources is presented. This technique is based upon the familiar extended Kalman filter algorithm used in navigation, radar‐based tracking, and feedback control applications. In this application, acoustic sensor‐to‐source bearing measurements from two locations are used to estimate target position, speed, and heading in near real time. In particular, two strategies for compensating the variable delay in obtaining bearing information caused by sound propagation are developed and presented. One of these is based upon a simple linear predictor. While adequate for sources that are not too distant from the acoustic sensors, it is shown to perform inadequately in some longer‐range situations. A second approach, based upon reformulating the Kalman filter to work in conjunction with a generalized triangulation algorithm, is shown to be more robust. Advantageous features of employing a Kalman filter approach to tracking problems include the ability to systematically compensate for measurement noise and to accommodate target maneuverability constraints. Application of the technique to data obtained from a highly maneuverable low‐flying commercial aircraft are presented and compared to flight path data obtained via the Global Positioning System.

Exploiting both output channels of seismically decoupled infrasound sensors for enhanced transient detection and direction-of-arrival estimation

Some infrasound sensors have internal sensing capabilities that remove or permit subsequent removal of much of the effect of ground motion on the measured pressure fluctuations. This presentation explores possible enhancements to transient acoustic signal detection and direction-of-arrival estimation when data corresponding to the ground motion and pressure fluctuation signals are analyzed simultaneously using statistical array signal processing techniques. Data from three 4-element infrasound sensor arrays that were deployed to detect explosive events occurring from 10 km to 20 km distant are used as the basis for the study.

Noise adaptive real-time estimation of infrasound using a compact array

Previous investigations [Frazier, ASA Pittsburg (2015); ASA San Francisco (2013); ASA Kansas City (2012)] have demonstrated that exploiting the correlated statistical properties of wind noise on short spatiotemporal scales with a compact array permits enhanced detection and reduced estimation error of infrasound transients resulting from explosions. The approach was shown to be effective in real-time using non-parametric wind noise models. More recently [Frazier, ASA Boston (2017); ASA Salt Lake City (2016)], the basic ideas of the approach were extended to estimation of quasi-stationary infrasound signals such as microbaroms using parametric wind noise models and was demonstrated to be effective, but the computational requirements were too demanding for real-time implementation. This presentation will highlight modifications that enable real-time implementation for quasi-stationary signal estimation.

Wind noise reduction using a compact infrasound sensor array and a Kalman filter based on the Matérn Covariance Function

A method for real-time estimation of stationary infrasound signals such as microbaroms in wind noise at low signal-to-noise ratios using a compact infrasound sensor array is presented. A compact array is defined as a sensor array that has an aperture that is much smaller than the shortest infrasound wavelengths of interest and is unsuitable for estimation of direction-of-arrival. In this application, the spacing between sensors results in the measured wind noise being highly correlated, and therefore, simple averaging cannot be used to obtain a good estimate of the infrasound signal. However, by adequately modeling the spatiotemporal wind noise process, array gain can be realized. The method is based on using a Kalman Filter that is designed with the assumption that the measured wind noise can be adequately modeled as a dynamic Gaussian random field with a Matern covariance function (demonstrated previously at the ASA Meeting in Salt Lake City, Utah, May, 2016). The presentation describes how to design th...

A highly portable, lightweight windscreen for infrasound sensors

It is often desired to deploy infrasound sensors in remote locations for short periods of time. Because of noise created by wind flow past the sensor in the atmosphere at infrasound frequencies, it is also desirable to employ a mechanical windscreen with the goal of reducing the effect of the wind so as to enhance the acoustic signal-to-noise ratio. One of the challenges in wind screen design for temporary applications is to achieve reasonable portability in addition to adequate wind noise filtering. In this presentation, we report the performance a new windscreen design that is highly portable and that achieves performance on par with a commonly used dome-shaped windscreen. Performance comparisons of this newly designed windscreen, a typical porous hose windscreen, and a commonly used dome windscreen are investigated. Specifically, wind noise reduction of these three windscreens compared with a un-screened sensor are presented as a function of frequency and wavenumber using the measured wind speed as a p...

Method for effective implementation of an acoustic based Hostile Fire Detection System on moving vehicles

Due to background noise issues, effective implementation of an acoustic array as part of a Hostile Fire Detection System on moving vehicles has been a significant challenge to date. A new method utilizing both software and hardware design parameters has been developed and successfully demonstrated to be very effective at high speeds on various vehicles. This method has been shown to be effective against both flow based noise sources and acoustic wave based noise sources in low signal to noise ratio conditions. The technology will be discussed and results from measurements utilizing conformally mounted acoustic sensors on moving ground vehicles will be shown.

Using the Matérn covariance function to enhance detection and estimation of transient infrasound

Adaptive filtering techniques are well known to enhance detection of anomolous events in the presence of slowly changing (quasi-stationary), autocorrelated noise backgrounds such as those occurring in measurements of infrasound signals in the atmosphere. The noise in this case is dominated by the effect of the turbulent wind blowing over the sensing element. While mechanical windscreens can provide significant signal-to-noise ratio gains, additional benefits can be obtained by employing adaptive filters. In this presentation, a kernel-based adaptive filter based on the Matern Covariance Function is demonstrated to improve the probability of detection of transient infrasound signals as well as improve signal estimation errors without increasing false alarm rates in the presence of wind noise. The choice of the Matern Covaraiance Function to represent the wind noise process is motivated by its roots in fractional-order stochastic differential equations. Because the wind noise at infrasound frequencies can b...

Wind noise suppression using compact arrays of infrasound sensors

This presentation describes how a compact array (aperture much less than the shortest wavelengths of interest) of infrasound sensors can be used to significantly enhance signal detection and waveform estimation in the presence of high levels of wind noise without the use of unwieldy mechanical screening. The methodologys effectiveness is founded on the fact that wind noise can be highly correlated on short spatiotemporal scales. This correlation structure is adaptively estimated from data and is used to formulate a generalized likelihood ratio detection problem and a minimum mean squared error waveform estimation problem. The infrasoudnic waveform is explicitly represented by a user-definable, parametrically characterized, stochastic prior distribution. Choice of this prior can enhance detection of anticipated signals and suppress others and thus more than one prior can be utilized in order to perform some level of classification. The presentation provides typical performance results from a range of appl...