Covariance matrices of noise models are used in signal and array processing to study the effect of various noise fields and array configurations on signals and their detectability. Here, the asymptotic eigenvalues of noise covariance matrices in 2-D and 3-D attenuating media are derived. The asymptotic eigenvalues are given by a continuous function, which is the Fourier transform of the infinite sequence formed by sampling the spatial coherence function. The presence of attenuation decreases the value of the large eigenvalues and raises the value of the smaller eigenvalues (compared to the attenuation free case). The eigenvalue density of the sample covariance matrix also shows variation in shape depending on the attenuation, which potentially could be used to retrieve medium attenuation properties from observations of noise.

The performance benefits of multiple input multiple output (MIMO) antenna systems in increasing the throughput of a communication system depends on the type and quality of the information sent to the transmitter. In this paper, we examine the loss in throughput due to a delay in receiving information at the transmitter. Specifically, we examine the impact on the throughput when the channel has a delayed channel matrix estimate, as well as when delayed channel statistical information is provided. Additionally, the effect of both delayed transmit and receiver channel state information is explored. The performance is examined using measured data in several environments. Two analytical channel models are synthesized using parameters estimated from the measured data. The usefulness of the analytical models in predicting the same performance when delays are present is then explored.

This paper investigates the capabilities of analytical MIMO channel models in correctly capturing the underlying correlation structure of geometrically-based channel structures. The limitations and performance differences of two correlation-based analytical models, the Kronecker model and a structured eigenbasis model are examined. Additional insight into the limitations of using either model in accurately predicting system performance is provided through an examination of common channel metrics and the impact of modeling mismatch on those metrics.

In this paper, we examine an adaptively modulated Open Loop MIMO (OL-MIMO) system that uses a modified V-BLAST transmission technique with limited feedback and a Closed Loop MIMO system (CL-MIMO) that employs a waterfilling strategy. The performance of these systems are evaluated using measured MIMO channel matrices and compared with the performance predicted by two analytical models, a link-end correlation-based model known as the Kronecker model and an eigen-based model developed by Weicheselberger [1]. Results show that the Weichselberger model accurately predicts the average channel behavior while the Kronecker model severely underestimates the throughput of the channel at high SNRs.

For selecting optimal signaling strategies in MIMO systems, a variety of metrics are used for determining the correct mode of operation. However, many of the strategies that have been developed rely on a particular correlation-based channel model, known as the Kronecker model. In this paper, we investigate the impact of the choice of channel model on commonly used channel characterization and feedback metrics. We compare two different models, an eigen-based model introduced by Weichselberger and the more commonly used link- end based correlation model, the Kronecker model.

An active sonar scenario in which the sonar array is placed at a relatively shallow depth and is prone to surface interference is considered. The array depth is either known or unknown, with known probability density function. The performance of two detectors representative of the detection approaches discussed is examined using a reverberation simulation generator (REVSIM). The performance comparison is made in terms of detector receiver operating characteristic (ROC) curves.>

The use of full wavefield (matched field) processing techniques in a multipath environment is illustrated. Experimental data consist of the arrivals at a hydrophone array from an explosive source in shallow water. A coherent recombination of the multipath arrivals is performed in order to provide enhanced signal-to-noise ratio for source detection and range/depth localization. Thus, signal-to-noise ratio is enhanced. Range/depth localization information is made available by searching for a peak in the surface formed by plotting beamformer output at all range/depth cells of interest. The results are compared to theoretical predictions of range/depth localization resolution based on the characteristics of the oceanic waveguide.>

Data from a vertical array experiment is used to illustrate the impact of seismic profiling on ambient noise spectrum level and vertical directionality in the 25-300 Hz frequency region. It is shown by the analysis results that seismic profiling has a substantial impact on the spectral and spatial characteristics of ambient noise. For these data, the contamination is most severe in the region below 50 Hz and in the band 150-200 Hz. Spatially, the seismic profiler energy arrives within +or-15 degrees of the horizontal and produces up to 8-dB deviations from the underlying vertical directionality of the profiler-free data.>

An alternative to the point scattering model approach for simulating reverberation is discussed. Reverberation is synthesized by passing a vector white noise process through a multichannel IIR (infinite-impulse response) filter whose time-varying transfer function matches the time-evolving spectral characteristics of the expected auto and cross range-Doppler maps.>

A pole-zero adaptive noise cancelling structure is described and its relevance to the adaptive beamforming problem discussed. An embedding approach is taken where a scalar ARMA (autoregressive moving-average) process is embedded into a two-channel AR process. Drawing on previously developed recursions for the joint process, multichannel, least-squares lattice, the recursions for the pole-zero adaptive noise canceller are implemented. A performance comparison is made between the resulting pole-zero adaptive noise cancelling structure and the all-zero least-squares lattice adaptive noise cancelling structure. The ability to implement adaptive poles is shown to be particularly advantageous in the region where the reference channel has a null.>

The application of adaptive least-mean square (LMS) and least-squares lattice (LSL) algorithms to the rejection of sea surface reverberation is discussed. A specific dual channel noise cancelling structure is considered along with the results of processing a ping selected from an experimental ocean reverberation data set. A time-evolving eigenvalue/eigenvector decomposition is used to provide insight into the performance differences between the two adaptive algorithms in this highly dynamic physical geometry.

An adaptive joint process structure is described which rejects under-ice reverberation by taking advantage of the spatial separation between acoustic backscatter returning from small elevation angles and transmitted energy reflected off the sea surface.

An application of the adaptive joint complex least-squares lattice algorithm to the rejection of sea surface reverberation is described. A specific dual channel noise cancelling structure is considered along with the results of processing experimental ocean reverberation data.

Both the water layer and bottom sediment act as transmission media for acoustic signals. In April 1981, the Marine Physical Laboratory (MPL) conducted an experiment investigating acoustic signal propagation in thick sediments. In this paper, the coherence between sediment borne and water path acoustic signals will be discussed. Data from the Monterey Fan experiment will be presented along with comments regarding the viability of coherently recombining the total acoustic field which is available at the receiving array.

The application of adaptive linear predictor structures to the prewhitening of acoustic reverberation data is discussed. The results of processing experimental ocean reverberation data constitutes the bulk of the material. In specific, a comparison between three adaptive linear predictor structures is made: (1) complex gradient transversal filter (CLMS), (2) complex gradient lattice (CGL), and (3) complex least-squares-lattice (CLSL). Figures include: (1) input range-Doppler map (conventional), (2) input range-Doppler map (inverse of the spectrum of the prewhitening filter), and (3) output range-Doppler map (conventional). Trade-offs between the three adaptive structures are illustrated as a function of their respective adaptation rate parameters.

Doppler sonars use the perceived shift in carrier frequency between the outgoing pulse and the returning echo to make an estimate of scatterer radial velocity at several ranges of interest. Scatterers suspended neutrally buoyant in the water column are used as tracers for the remote sensing of water mass motion. Inherently, the problem is one of high resolution (distinguishing small differences in velocity) spectral estimation across segments (range bins) which are too short to provide the desired resolution via conventional methods (FFT) of spectral analysis. In this paper, the application of pth-order autoregressive time series analysis models to the estimation of the time-evolving Doppler sonar reverberation power spectrum is reported. First, with the aid of a reverberation model, it is shown that the spatial transfer function characteristics of a transducer have a significant influence on the time-evolving shape of the reverberation power spectrum. Thus, it is suggested that the use of first-order models for the purpose of Doppler shift estimation simply is inadequate. Then, the results of an exploratory analysis of pings from a 67 kHz Doppler sonar are presented which appears to confirm this caution.

A greal deal of interest has been shown in the scalar adaptive lattice structure due to its rapid convergence properties even in the presence of a wide eigenvalue spread in the input data. Recursive update expressions for the lattice parameters (reflection coefficients) have been developed both as the approximate solution to the Wiener filtering problem and the exact solution to the deterministic least squares problem. Considered here is the multi-channel extension of the lattice structure and its application to adaptive array processing.

Design trade-offs in a programmable time-delay-and-sum digital beamformer built by MPL will be discussed. This dynamic beam-former permits the incorporation of slow changes in element positions and/or beam steering direction while the beamformer carries out the real-time formation of 1300 beams from 32 input sensors. The sensors are distributed in a random but known manner over an aperture diameter of up to 800 λ (at 400 Hz).

The approach to adaptive array procesing typically has taken one of two paths - time domain (wide band) or frequency domain (narrow band). In the former, the entire frequency range of interest is processed as a whole with a multichannel tapped delay line adaptive filter providing frequency selectivity. In the latter, the desired passband is broken into independent contiguous pieces each of which is processed independently via two quadrature weights per channel. This paper investigates the tradeoffs in each approach and the conditions under which they are equivalent.

Optimum in the mean-square error sense, the Wiener filter is the best linear filter which can be derived given known input statistics. Excellent results have been achieved for the filtering problem with the LMS adaptive filter when these same statistics are unknown. The gradient descent algorithm introduces an excess mean-square error which is proportional to the adaptive filter's length. In an adaptive array processor, the LMS filter can be configured as a noise canceller to partially remove a sidelobe interference source from a given receive beam. This paper derives the optimum length of the adaptive transversal filter such that the residual interference signal energy plus the excess mean-square error contributed by the LMS algorithm is minimized. Both narrowband and wideband interference signals are considered.

The performance of a likelihood ratio processor must often be determined via computer simulation. This correspondence suggests an effective method of calculating the receiver operating characteristic (ROC) that avoids emphasizing the upper tail region of the likelihood ratio's estimated probability density function. A specific example is presented.