Source Azimuth Research Articles

Low-Power Log-Domain CMOS Filter Bank for 2-D Sound Source Localization

This paper proposes the design of a novel current-mode front-end for the extraction of localization spectral cues from two audio signals, together with test results. The front-end consists of two parallel filter banks, envelope extraction and comparison circuitry, together with an AGC loop. The extracted cues are intended to be further processed in order to determine the source azimuth and elevation. A current-mode log-domain implementation using subthreshold MOS operation is used for micropower operation while still achieving a good bandwidth and linearity. A current-mode solution is also preferred because of the ease of implementation of certain mathematical operations. The front-end splits the input signals into different frequency bands and computes monaural and interaural spectral cues from the resulting signal envelopes for each band. The front-end has been optimized to operate at a supply voltage of 1.8 V and most blocks have been designed using a differential architecture. To our knowledge, this is the first log-domain implementation of a front-end for 2-D localization cues extraction. The design has been carried out using a standard double-poly double-metal 0.8 μm CMOS process with VT e 0.8 V. The bandpass filters which form the main core of the chip exhibit a measured dynamic range of 62 dB corresponding to 1.9% THD, while the total power dissipation is 890 μW.

An AUV underwater acoustic array

An eight-element hydrophone array has been installed within the shroud of an Odyssey IIb AUV. Each array channel is digitized and recorded with 10-kHz bandwidth by an autonomous, PC-104+ data acquisition system. To determine the effects of the AUV body itself on the acoustic field measured by the array, tones at low (200–800 Hz) and mid- (2–8 kHz) frequencies were transmitted in a large water tank to the AUV by a source continuously varying in azimuth but fixed range. Bartlett beamforming for source azimuth using both plane wave replicas and replicas calculated using a 2-D time domain, finite-difference code that accounts for the scattering from the two glass spheres in the AUV shows that (1) not accounting for scattering results in 2- to 6-dB signal gain degradation, but (2) scattering moves the ambiguity surface sidelobes closer to the main lobe location, thus improving localization capability. The effect of scattering from the AUV on an azimuthally isotropic noise field, estimated by averaging over pings at all azimuths, is a decrease in spatial coherence at low frequencies over that in free space, thus increasing detection capability. [Work supported by ONR.]

Encoding of sound motion by binaural brainstem units in a Horseshoe bat.

In order to study how and if single brainstem units respond to moving compared with stationary sounds, radially moving sound sources were presented to the bat, Rhinolophus ferrumequinum. This time-variant binaural stimulation was simulated dichotically through earphones (closed-acoustic-field for the virtual azimuth range of +/-40 degrees from the midline). Neurophysiologically recorded responses primarily showed a function of interaural intensity difference (IID) which is considered a direct correlate of the sound source's azimuth angle. However, this is only true for the stationary case. Unit's response did not remain unaffected by the dynamic stimulus cues of sound source movement (velocity and direction). Maximal discharge rate became a function of motion velocity as well as the slopes of the response profiles. Hence, coding of IID became ambiguous as, depending on the unit, the response profiles and therefore a unit's receptive field, became spatially shifted with respect to one another when the direction of the sound source movement was reversed. Shifts within the movement direction (hysteresis) as well as against it (termed here 'advance') were observed: hysteresis is typical for units with non-monotonic, stationary rate/intensity functions, whereas those units with monotonic functions predominantly show advances. Further dynamic response features in form of transient peaks and troughs, superimposed on the response profiles, were registered. It appears that the ongoing firing rate no longer represents azimuth position alone, but vigorously reproduces the dynamic cues (velocity and movement direction), too. With respect to the neural mechanisms leading to dynamic response features, it is proposed that, as long excitation and inhibition act with similar short time constants, neural activity can rapidly and faithfully follow changing IIDs. Different time constants for excitation, inhibition, facilitation, and depression may be responsible for the dynamic 'features' such as transient responses and hysteresis/advance. They may provide biologically relevant information for nocturnally hunting bats to efficiently guide their flight maneuvers.

Binaurality and azimuth tuning of neurons in the auditory cortex of the big brown bat

By using a combined closed and free-field stimulation system, binaurality and azimuth tuning of the neurons in the auditory cortex of the big brown bat,Eptesicus fuscus, were studied. A variety of azimuth-tuning functions were demonstrated for the binaural neurons. The large majority of EE (contralateral and ipsilateral excitatory) neurons exhibited azimuth selectivity with the best azimuths (BA) at contralateral 30°–40°, some at ipsilateral 20°-40° and preferred azimuth ranges (PAR, response amplitude ⩾75% of maximum) between 8° and 40°. Sound source azimuths strongly modulate spike counts with a mean modulation depth of 83.8% for EE neurons. EI (contralateral excitatory and ipsilateral inhibitory) neurons have simple azimuth tuning with BA located at contralateral 20°-40° and a broad PAR ranged from 30° to 55°. The present results suggest that azimuth-tuning characteristics of binaural neurons in the auditory cortex of the bat are of significance for acoustic behaviour.

Cortical representation of auditory space: information-bearing features of spike patterns.

Previous studies have demonstrated that the spike patterns of cortical neurons vary systematically as a function of sound-source location such that the response of a single neuron can signal the location of a sound source throughout 360 degrees of azimuth. The present study examined specific features of spike patterns that might transmit information related to sound-source location. Analysis was based on responses of well-isolated single units recorded from cortical area A2 in alpha-chloralose-anesthetized cats. Stimuli were 80-ms noise bursts presented from loudspeakers in the horizontal plane; source azimuths ranged through 360 degrees in 20 degrees steps. Spike patterns were averaged across samples of eight trials. A competitive artificial neural network (ANN) identified sound-source locations by recognizing spike patterns; the ANN was trained using the learning vector quantization learning rule. The information about stimulus location that was transmitted by spike patterns was computed from joint stimulus-response probability matrices. Spike patterns were manipulated in various ways to isolate particular features. Full-spike patterns, which contained all spike-count information and spike timing with 100-micros precision, transmitted the most stimulus-related information. Transmitted information was sensitive to disruption of spike timing on a scale of more than approximately 4 ms and was reduced by an average of approximately 35% when spike-timing information was obliterated entirely. In a condition in which all but the first spike in each pattern were eliminated, transmitted information decreased by an average of only approximately 11%. In many cases, that condition showed essentially no loss of transmitted information. Three unidimensional features were extracted from spike patterns. Of those features, spike latency transmitted approximately 60% more information than that transmitted either by spike count or by a measure of latency dispersion. Information transmission by spike patterns recorded on single trials was substantially reduced compared with the information transmitted by averages of eight trials. In a comparison of averaged and nonaveraged responses, however, the information transmitted by latencies was reduced by only approximately 29%, whereas information transmitted by spike counts was reduced by 79%. Spike counts clearly are sensitive to sound-source location and could transmit information about sound-source locations. Nevertheless, the present results demonstrate that the timing of the first poststimulus spike carries a substantial amount, probably the majority, of the location-related information present in spike patterns. The results indicate that any complete model of the cortical representation of auditory space must incorporate the temporal characteristics of neuronal response patterns.

Azimuthal variation of the P phase in Icelandic receiver functions

SUMMARY A curious observation has been made on radial receiver functions calculated from teleseisms recorded by 29 broad-band seismometers distributed over Iceland. The arrival time of the direct P phase of the radial receiver functions depends critically upon the azimuth of the teleseismic source. For a seismic station in West Iceland, the direct P phase of the radial receiver function arrives consistently later for easterly source azimuths than for westerly source azimuths. The reverse applies for stations in East Iceland. In the original seismograms, the delayed P phase of the receiver function appears up to 450 ms later on the radial than on the vertical component. The seismometer locations in East and West Iceland are separated by the Neovolcanic Zone, a constructive plate boundary. The delayed P phases occur for seismic rays travelling across this zone. However, it is not obvious how wave propagation across the plate boundary zone could cause the observed delays. The tentative explanation proposed here involves the regional dip of the Icelandic lava sequences towards the Neovolcanic Zone. A dipping interface at shallow depth results in a P–S converted phase arriving shortly after the P phase. These phases cannot be separated in the radial receiver functions, given the bandwidth of the observed signals. However, a calculation of receiver functions from estimates of the P, SV and SH wavefields clearly reveals a P–S converted phase at about 500 ms for easterly source azimuths in West Iceland and for westerly source azimuths in East Iceland. The amplitudes of the direct P phase and the P–S phase converted at a dipping interface would be expected to vary strongly with azimuth in accordance with the observed behaviour.

Propagation of Crustal-Waveguide-Trapped Pg and Seismic Velocity Structure in the South Indian Shield

Short-period vertical-component digital recordings of the seismic wave field generated by mine tremors and explosions recorded at the Gauribidanur array (GBA) reveal relatively high-frequency secondary energy, with a group velocity of about 6 km/sec accompanied by a long, incoherent coda. We interpret this part of the wave field including the coda as resulting from a crustal-waveguide-trapped Pg , and it is observable from a distance of about 150 km behind the first arrival Pn phase. These observations and extensive modeling results reveal the fine structure of the crust beneath the Archaean eastern Dharwar craton in the south Indian shield. A waveguide with random successions of high- and low-velocity layers in the upper crust, possibly with considerable lateral extent beneath the Closepet granite, explains the main features of the waveguide Pg , such as coda length and group velocity. Reflectivity synthetic seismograms for 1D lamellae structures in the upper crust are found to reproduce the observed features of a record section assembled from GBA seismograms extending to a distance of 240 km. The preferred models of P - and S -velocity structure reveal lower than normal V P / V S and Poisson9s ratios: respectively, 1.63–1.65 and 0.20–0.21 in the upper crust, including the lamellae structure, and 1.70 and 0.235 in the lower crust. A Q P / Q S ratio nearly equal to, or even slightly less than, 1 in the upper crust is found to generate the synthetics consistent with the observed seismogram section. F-K spectral analysis of the observed P coda following the waveguide Pg , for travel paths mostly through the Closepet granite, convincingly reveals two dominant energy peaks, at near-on-azimuth as well as relatively off the source azimuth, both with an apparent velocity of 6.1 km/sec appropriate to the upper crustal depths. The inferred upper crustal structure modeled by alternating high and low velocity lamellae, together with the derived physical properties and the F-K analysis of the P coda, suggests a scattering waveguide at 5–15 km depth in this region. The scattering waveguide appears to be related to the large Closepet granite intrusion, which is believed to be a Precambrian suture between the western and eastern Dharwar cratons in the south Indian shield. Possible lateral extent of the modeled lamellae structure receives added support from a recently proposed hypothesis of a late Archaean mantle plume centered below the Closepet granite. The inferred crustal P -velocity model for the eastern Dharwar craton, revealing a transitional Moho at a shallow depth of 34–36 km, is consistent with the models for other Archaean cratons. The observed Pn velocity of 8.2 km/sec and its gradient 0.013 sec –1 compare fairly well with similar values found for other shield regions. V P / V S and Poisson9s ratios are found to increase respectively from 1.71 and 0.24 below the Moho to 1.78 and 0.27 at about 60 km depth in the upper mantle.

Localization of multiple sound sources with two microphones.

This paper presents a two-microphone technique for localization of multiple sound sources. Its fundamental structure is adopted from a binaural signal-processing scheme employed in biological systems for the localization of sources using interaural time differences (ITD). The two input signals are transformed to the frequency domain and analyzed for coincidences along left/right-channel delay-line pairs. The coincidence information is enhanced by a nonlinear operation followed by a temporal integration. The azimuths of the sound sources are estimated by integrating the coincidence locations across the broadband of frequencies in speech signals (the "direct" method). Further improvement is achieved by using a novel "stencil" filter pattern recognition procedure. This includes coincidences due to phase delays of greater than 2pi, which are generally regarded as ambiguous information. It is demonstrated that the stencil method can greatly enhance localization of lateral sources over the direct method. Also discussed and analyzed are two limitations involved in both methods, namely missed and artifactual sound sources. Anechoic chamber tests as well as computer simulation experiments showed that the signal-processing system generally worked well in detecting the spatial azimuths of four or six simultaneously competing sound sources.

Source tracking with a gradient-based eigenstructure algorithm

A new gradient-based eigenstructure algorithm to locate and track the azimuth and elevation angles of unknown sources in a 3-dimensional environment is proposed and investigated. Starting from initial estimates of the source locations from, say, a coarse search of the multiple signal classification (MUSIC) spectrum, the algorithm obtains better estimates and tracks sources in a repetitive cyclical manner. To refine a particular source location estimate: (1) a preprocessor is designed to remove the effects of the other sources; (2) the eigenvector for the largest eigenvalue of the covariance matrix after preprocessing is found and its difference from the ideal value is determined; and finally (3) a gradient calculation is used to obtain an estimate for the difference of the assumed source location from the actual position. The advantages of using the proposed technique rather than performing a thorough MUSIC search in the 2-dimensional spectrum are: (1) the implementation complexity can be reduced considerably, and (2) the algorithm is well suited to be employed for continuous adaptive filtering purposes in a tracking scenario.

Acoustic vector-sensor processing in the presence of a reflecting boundary

We consider the passive direction-of-arrival (DOA) estimation problem using arrays of acoustic vector sensors located in a fluid at or near a reflecting boundary. We formulate a general measurement model applicable to any planar surface, derive an expression for the Cramer-Rao bound (CRB) on the azimuth and elevation of a single source, and obtain a bound on the mean-square angular error (MSAE). We then examine two applications of great practical interest: hull-mounted and seabed arrays. For the former, we use three models for the hull: an ideal rigid surface for high frequency, an ideal pressure-release surface for low frequency, and a more complex, realistic layered model. For the seabed scenario, we model the ocean floor as an absorptive liquid layer. For each application, we use the CRB, MSAE bound, and beam patterns to quantify the advantages of using velocity and/or vector sensors instead of pressure sensors. For the hull-mounted application, we show that normal component velocity sensors overcome the well-known, low-frequency problem of small pressure signals without the need for an undesirable stand-off distance. For the seabed scenario, we also derive a fast wideband estimator of the source location using a single vector sensor.

An ASIC-chip for stereoscopic depth analysis in video-real-time based on visual cortical cell behavior.

In a stereoscopic system both eyes or cameras have a slightly different view. As a consequence small variations between the projected images exist ("disparities") which are spatially evaluated in order to retrieve depth information. We will show that two related algorithmic versions can be designed which recover disparity. Both approaches are based on the comparison of filter outputs from filtering the left and the right image. The difference of the phase components between left and right filter responses encodes the disparity. One approach uses regular Gabor filters and computes the spatial phase differences in a conventional way as described already in 1988 by Sanger. Novel to this approach, however, is that we formulate it in a way which is fully compatible with neural operations in the visual cortex. The second approach uses the apparently paradoxical similarity between the analysis of visual disparities and the determination of the azimuth of a sound source. Animals determine the direction of the sound from the temporal delay between the left and right ear signals. Similarly, in our second approach we transpose the spatially defined problem of disparity analysis into the temporal domain and utilize two resonators implemented in the form of causal (electronic) filters to determine the disparity as local temporal phase differences between the left and right filter responses. This approach permits video real-time analysis of stereo image sequences (see movies at http://www.neurop.ruhr-uni-bochum.de/Real- Time-Stereo) and a FPGA-based PC-board has been developed which performs stereo-analysis at full PAL resolution in video real-time. An ASIC chip will be available in March 2000.

Binaural cross-correlation and auditory localization in the barn owl: a theoretical study

The barn owl is a nocturnal predator that is able to capture mice in complete darkness using only sound to localize prey. Two binaural cues are used by the barn owl to determine the spatial position of a sound source: differences in the time of arrival of sounds at the two ears for the azimuth (interaural time differences (ITDs)) and differences in their amplitude for the elevation (interaural level differences (ILDs)). Neurophysiological investigations have revealed that two different neural pathways starting from the cochlea seem to be specialized for processing ITDs and ILDs. Much evidence suggests that in the barn owl the localization of the azimuth is based on a cross-correlation-like treatment of the auditory inputs at the two ears. In particular, in the external nucleus of the inferior colliculus (ICx), where cells are activated by specific values of ITD, neural activation has been recently observed to be dependent on some measure of the level of cross-correlation between the input auditory signals. However, it has also been observed that these neurons are less sensitive to noise than predicted by direct binaural cross-correlation. The mechanisms underlying such signal-to-noise improvement are not known. In this paper, by focusing on a model of the barn owl's neural pathway to the optic tectum dedicated to the localization of the azimuth, we study the mechanisms by which the ITD tuning of ICx units is achieved. By means of analytical examinations and computer simulations, we show that strong analogies exist between the process by which the barn owl evaluates the azimuth of a sound source and the generalized cross-correlation algorithm, one of the most robust methods for the estimate of time delays.

Effects of sensor placement on acoustic vector-sensor array performance

We consider the role played by the sensor locations in the optimal performance of an array of acoustic vector sensors, First we derive an expression for the Cramer-Rao bound on the azimuth and elevation of a single far-field source for an arbitrary acoustic vector-sensor array in a homogeneous wholespace and show that it has a block diagonal structure, i.e., the source location parameters are uncoupled from the signal and noise strength parameters. We then derive a set of necessary and sufficient geometrical constraints for the two direction parameters, azimuth and elevation, to be uncoupled from each other. Ensuring that these parameters are uncoupled minimizes the bound and means they are the natural or canonical location parameters for the model. We argue that it provides a compelling array design criterion. We also consider a bound on the mean-square angular error and its asymptotic normalization, which are useful measures in three-dimensional bearing estimation problems. We derive an expression for this bound and discuss it in terms of the sensors' locations. We then show that our previously derived geometrical conditions are also sufficient to ensure that this bound is independent of azimuth. Finally, we extend those conditions to obtain a set of geometrical constraints that ensure the optimal performance is isotropic.

Azimuth and elevation direction finding using arbitrary array geometries

An ESPRIT-based algorithm is proposed to estimate the azimuth and elevation angles of multiple independent sources. The algorithm uses cumulants and imposes no geometric constraint on the array. Only one third of the hardware is needed for our algorithm as compared with covariance-based two-dimensional (2-D) ESPRIT. Our algorithm can estimate azimuth and elevation angles of M-1 sources using M sensors. Simulation results show that for several array configurations, our algorithm works well.

S‐wave structure of the American Southwest and Mexico from single‐station broadband data

We investigate S velocity structure beneath the Southwestern U.S. and Mexico using upper mantle turning S waves and surface waves. The data are tangential component records from Baja California, Mexican and Central American events recorded at broadband station ANMO (Albuquerque, New Mexico). Study geometry allows investigation of the variation of S structure with source‐receiver azimuth for sources from Baja to Yucatan. The narrow backazimuthal sector studied comprises a transition from oceanic to tectonic to more ‘continent‐like’ structure. We document this transition in the lithosphere and upper mantle to roughly 450 km depth.

Three-dimensional auditory localization of nearby sources

Auditory localization performance has been measured for a point- source stimulus within 1 m of a listener’s head. Broadband noise bursts, roved in intensity to eliminate loudness as a distance cue, were presented at random locations ranging from 0°–180° in azimuth, 60°–60° in elevation, and 15 cm–1 m in distance. The listeners responded to these stimuli by moving a response wand to the perceived location of the sound. Response errors were then calculated in azimuth, elevation, and distance. Directional localization error for nearby sources was comparable to that previously reported in localization experiments at distances greater than 1 m. Distance accuracy was found to depend on source azimuth, with best performance to the listener’s side and worst performance directly in front of the listener. Distance judgments were slightly more accurate when the amplitude of the stimulus was not randomized to eliminate distance effects. The improvement in accuracy for the fixed-amplitude stimulus was greater in front than to the side of the listener. Distance accuracy was substantially degraded when one of the listener’s ears was occluded by an earplug and muff. These results suggest the importance of binaural cues to near-field distance perception. [Work supported by AFOSR.]

Across-frequency nonlinear inhibition by GABA in processing of interaural time difference

The barn owl uses the interaural time difference (ITD) to determine the azimuth of a sound source. Narrowband ITD-sensitive neurons cannot distinguish a given ITD from those that produce the same interaural phase difference (phase ambiguity). Neurons in the external nucleus of the inferior colliculus (ICx) resolve the ambiguity by gathering ITD information across many frequencies, thereby suppressing false responses (side peaks, SP) relative to the true ITD (the main peak, MP) in a response versus ITD curve. This process was quantitatively studied by comparing the ITD curve for a pair of tones presented simultaneously (two-tone curve) to the simple sum (predicted curve) of the individual ITD curves for the same tones presented separately. Sixteen of the 39 neurons tested did not show a significant difference in MP and SP responses between these curves (category I); 14 neurons showed significant SP suppression (category II). During iontophoretic application of bicuculline methiodide, a GABAA antagonist, most (n=7/8) category II neurons lost nonlinear SP suppression and became linear, whereas category I neurons retained linear summation (n=3/3). Thus, the nonlinear cross-frequency interaction of ITD responses in ICx neurons was mediated mostly by GABAergic inhibition, which enhanced SP suppression, and helped resolve phase ambiguity.

An Anatomical Basis for Visual Calibration of the Auditory Space Map in the Barn Owl’s Midbrain

The map of auditory space in the external nucleus of the inferior colliculus (ICX) of the barn owl is calibrated by visual experience during development. ICX neurons are tuned for interaural time difference (ITD), the owl's primary cue for sound source azimuth, and are arranged into a map of ITD. When vision is altered by rearing owls with prismatic spectacles that shift the visual field in azimuth, ITD tuning in the ICX shifts adaptively. In contrast, ITD tuning remains unchanged in the lateral shell of the central nucleus of the inferior colliculus (ICCls), which provides the principal auditory input to the ICX, suggesting that the projection from the ICCls to the ICX is altered by prism-rearing. In this study, the topography of the ICCls-ICX projection was assessed in normal and prism-reared owls by retrograde labeling using biotinylated dextran amine. In juvenile owls at the age before prism attachment, and in normal adults, labeling patterns were consistent with a topographic projection, with each ICX site receiving input from a restricted region of the ICCls with similar ITD tuning. In prism-reared owls, labeling patterns were systematically altered: each ICX site received additional, abnormal input from a region of the ICCls where ITD tuning matched the shifted ITD tuning of the ICX neurons. These results indicate that anatomical reorganization of the ICCls-ICX projection contributes to the visual calibration of the ICX auditory space map.

Three-dimensional noise field directionality estimation from single-line towed array data

For many reasons, the three-dimensional (3-D) arrival structure of the undersea ambient noise field is of interest to the research and development community. One reason is that the arrival structure can be used to estimate the beam noise of an array, which may be required to estimate the performance of the array as an operational Navy asset or a scientific measurement tool. Another reason is that there are clues inherent in the vertical arrival structure that relate to the nature of the acoustic propagation along the azimuths of the noise sources. Similarly, there are also clues in the horizontal arrival structure of the undersea ambient noise field that relate to the azimuthal distribution of the noise sources. Both of these classes of clues are important in the verification and validation of undersea ambient noise models. The ideal measurement tool to measure the 3-D arrival structure of the noise field is a high-resolution volumetric array sonar system. Unfortunately, such a system is not generally available. However, towed horizontal line arrays are available and can be used, even though they are far from ideal. The beam patterns of a line array are conically symmetric about the axis of the array. When the axis of a horizontal line array is tilted slightly from an elevation angle of 0°, the vertical character of the beam cones can be used to discriminate between various vertical arrival angles. Combined with measurements made on several different towed array headings, an estimate of the 3-D directionality of the noise field can be obtained. An algorithm that uses such a technique on single-line towed array data to generate an estimate of the 3-D directionality of the noise field is described herein. Some results using measured data from a recent noise measurement exercise are presented and discussed.

Spatial acuity of ultrasound hearing in flying crickets.

The minimum audible angle is the smallest angular separation at which two sounds are perceived as coming from distinct sources. To determine the spatial acuity of hearing in crickets, we measured minimum audible angles at various locations in azimuth and elevation. Crickets (Teleogryllus oceanicus) were able to discriminate between sources separated by 11.25 degrees (1/32 of a circle) in azimuth directly ahead of them; acuity declined to 45 degrees in azimuth for sound sources 90 degrees to the side and then improved to 33.75 degrees at the rear. Crickets were also able to discriminate between sources separated in elevation, although acuity was much poorer, ranging from 45 degrees at the front and rear of the animal to 90 degrees below the animal. A habituation-dishabituation test was used to test discrimination. This involved presenting a train of ultrasound pulses from one location, habituating the cricket's escape response. This train was followed by a test pulse of ultrasound from another location, after which a final pulse was presented from the original source. If the test pulse was discriminated from the habituating pulses, then the response to the final pulse was dishabituated. To determine the minimum audible angle, we repeated such tests while moving the two sound sources closer together until dishabituation no longer occurred.