This paper compares the performance of a MIMO system applying a Mismatched Maximum Likelihood Detector (MMLD) with that applying a Generalized Likelihood Ratio Detector (GLRD) in the presence of channel estimation errors. Two sources of error are considered: errors due to noise and errors due to time variations induced by Doppler effects. The MMLD relies on the available channel estimate whereas the GLRD combines the channel estimate with a block of received data and uses a joint channel-symbol estimation approach to solve for the transmitted symbols. The GLRD is shown to outperform the MMLD whenever errors in channel estimation are incurred and the performance gap depends on the amount of error. Furthermore, the GLRD provides a unified receiver structure which can also be applied to differentially encoded MIMO systems and to situation where no channel estimate is available as in blind MIMO. The improved performance of the GLRD comes with increased computational complexity. This problem is addressed through search efficient algorithms based on the branch-estimate-bound optimization framework. We analyze the performance of the GLRD and provide a tight upper bound on the BER of the system at high SNR. Simulation results are shown for a 2 × 2 MIMO system.

We consider the problem of communicating through a time-varying channel whose parameters (gains, impulse response coefficients) are unknown, but have known statistics. Several detection schemes are developed for the signals received at the channel output. The performance of these detectors is evaluated as a function of the rank of the channel covariance matrix. It is demonstrated that effective communication can be achieved provided that the channel covariance matrix has a low rank.

The partly-filled nonuniform linear arrays (PFNLA) are presented and the DOA estimation problem for multipath signals is investigated. A new approach is proposed for DOA estimation in nonuniform linear arrays (NLA) based on array interpolation. A Wiener formulation is used to improve the condition number of the mapping matrix as well as the performance for noisy observations. An initial DOA estimate is obtained by using the uniform part of the PFNLA. This initial estimate is used in array interpolation and a new covariance matrix is found which improves the DOA estimation significantly. Noniterative and iterative algorithms are developed for the DOA estimation in multipath. The proposed approach overcomes some of the limitations of the conventional array interpolation. It is shown that the DOA performance is close to the CRB and the method is robust for a variety of source and DOA scenarios.

The performance of an antenna array in a distributed fading environment depends on how the signal gets distributed in space. It is known that the scattering in an urban area mainly happens within a 100 wavelength radius around a mobile. Assuming that the base station antennas are above the rooftop in urban areas, and approximately 1-2 kilometers from the mobile, the angular spread of the signal impinging on the array is on the order of 5-10 degrees, for a 1.8 GHz system. This angular spread results in an array response where the fading of the array elements is correlated along the array. We study how correlation effects the ability of the array to separate cochannel signals, and propose estimation algorithms which exploit this correlation. We derive the approximate theoretic performance of the array as a function of both angular spread and input SNR. Performance results are presented that compare the performance of different estimators, as well as the proposed theoretic performance, as a function of the input SNR.

Synchronous code-division multiple-access (CDMA) techniques possess intrinsic protection against cochannel interference when orthogonal spreading codes are used. However, in the presence of multipath propagation, the signals lose their orthogonality property, leading to increased cross correlation. In these cases, channel estimation may be needed in order to improve detection. We propose and compare several algorithms for channel estimation of a synchronous CDMA point-to-multipoint link (downlink) that uses aperiodic spreading waveforms. We compare by simulation and analysis a subspace approach, a pilot-aided approach, and a decision-based approach.

Some quadratic time-frequency representations (TFRs) may be called time-varying spectrum estimators. They are derived from first principles, and they turn out to be time-varying multiwindow spectrum estimators. In special cases they are time-varying spectrograms that may be written as Fourier transforms of lag-windowed, time-varying correlation sequences or as spectrally smoothed time-varying periodograms. These are not ad-hoc variations on stationary ideas to accommodate time variation. Rather, they are the only variations one can obtain for time-varying spectrum analysis.

We consider the problem of detecting a subspace signal when there is uncertainty in the subspace. Such uncertainty usually causes a mismatch between the detector and the signal to be detected, which may lead to significant loss in performance. To improve the robustness of the detection procedure we apply robust adaptive subspace detectors based on extending the dimension of the signal subspace. We consider two types of adaptive constant false alarm rate (CFAR) detector structures for the extended subspace detectors: CFAR generalized likelihood ratio detector (CFAR GLR) and CFAR matched subspace detector (CFAR MSD). Using Monte-Carlo simulations, we study the performance of the robust adaptive subspace detectors for space time processing.

This paper considers the problem of estimating the parameters of two-dimensional moving average random fields. We first address the problem of expressing the covariance matrix of a moving average random field, in terms of the model parameters. Assuming the random field is Gaussian, we derive a closed form expression for the Cramer-Rao lower bound on the error variance in jointly estimating the model parameters. A computationally efficient algorithm for estimating the parameters of the moving average model is developed. The algorithm initially fits a two-dimensional autoregressive model to the observed field, then uses the estimated parameters to compute the moving average model.

The problem of separating and estimating signals received by an array whose array manifold has an unknown structural form is usually referred to as the blind signal copy problem. In this paper we consider the blind signal copy problem for polynomial phase signals. By deriving the Cramer Rao bound we evaluate the optimal performance achievable by any unbiased estimator. To gain additional insight into this problem we compare the CRB to the bound for the case where the functional form of the array manifold is known. We derive a computationally efficient approximate maximum-likelihood (ML) algorithm and compare its performance with the bound.

We consider non-homogeneous 2-D signals which can be represented by a constant modulus polynomial-phase model. In previous papers we developed a computationally efficient estimation algorithm for the parameters of this model, and a novel phase unwrapping method which is based on this estimation algorithm. In this paper we analyze the performance of the algorithm and derive expressions for the mean squared error of the estimated coefficients. Assuming high signal to noise ratio (SNR), we show that the estimates are unbiased, and derive a rule for optimal selection of the algorithm parameters. The theoretical results are verified by Monte-Carlo simulations for selected examples. Finally, we present an approximate error analysis of the estimates for an arbitrary SNR. This analysis is carried out for a specific set of the algorithm parameters, which is selected based on the high SNR analysis.

Practical sensor arrays are characterized by certain parameters, related to the array shape and its electrical properties. Accurate direction finding requires calibration of these parameters to a relatively high accuracy. Off-line calibration of the array parameters is performed by transmitting signals from known locations and measuring the array response. This paper relates the parameter estimation accuracy achieved in off-line calibration to the direction-of-arrival errors during on-line operation. It then proposes a criterion for choosing the off-line calibration points so as to meet on-line accuracy requirements.

Signal cancellation causes the performance of conventional adaptive beamformers to severely deteriorate in environments containing interference sources that are correlated with the signal. We propose an algorithm which overcomes the signal cancellation problem for wideband signals. The proposed algorithm applies array manifold interpolation to perform frequency domain averaging, which reduces the correlation between the signals and the interference sources. In contrast to other algorithms that apply frequency averaging, the algorithm proposed does not require any preliminary estimation or prior information on the interference environment.

We consider nonhomogeneous 2-D signals that can be represented by a constant modulus polynomial-phase model. A novel 2-D phase differencing operator is introduced and used to develop a computationally efficient estimation algorithm for the parameters of this model. The operation of the algorithm is illustrated using an example.

Any regular stationary random processes can be represented as the sum of a purely indeterministic process and a deterministic one. This paper considers the achievable accuracy in the joint estimation of the parameters of these two components, from a single observed realization of the process. An exact form of the Cramer-Rao bound (CRB) is derived, as well as a conditional CRB. The relationships between these bounds, and their relations to the previously derived asymptotic bound, are explored by analysis and numerical examples.

The high-order ambiguity function (HAF) is a nonlinear operator designed to detect, estimate, and classify complex signals whose phase is a polynomial function of time. The HAF algorithm, introduced by Peleg and Porat (1991), estimates the phase parameters of polynomial-phase signals measured in noise. The purpose of this correspondence is to analyze the asymptotic accuracy of the HAF algorithm in the case of additive white Gaussian noise. It is shown that the asymptotic variances of the estimates are close to the Cramer-Rao bound (CRB) for high SNR. However, the ratio of the asymptotic variance and the CRB has a polynomial growth in the noise variance.

MAT2DSP is a MATLAB toolbox, currently under development, whose function is to estimate the implementation requirements of algorithms specified in the form of a MATLAB program. This toolbox is aimed at providing researchers developing advanced signal and image processing algorithms, a quick and convenient way of estimating what would be needed to implement their algorithm on a specified processor; MAT2DSP analyzes the user program and generates reports on its computational requirements.

The paper develops a method of error analysis for Fourier-transform based sinusoidal frequency estimation in the presence of nonrandom interferences. A general error formula is derived, and then specialized to the cases of additive and multiplicative interferences. Approximate error formulas are derived for the case of additive polynomial-phase interference. Finally, an application to error-analysis in estimating the parameters of multiple polynomial-phase signals is discussed in detail.

We consider the problem of separating and estimating the waveforms of superimposed signals received using an array of uncalibrated sensors. The array elements are assumed to have the same unknown gain pattern, up to an unknown multiplicative factor. The phases of the elements are arbitrary and unknown. In this paper we analyze the quality of the estimated signal in terms of the output signal to interference ratio (SIRO) and output signal to noise ratio (SNRO). It is shown that uncalibrated arrays can be used successfully for signal separation and estimation using only second order moments. The analysis is verified by Monte Carlo experiments using an algorithm for steering vector estimation.

This paper attempts to assess the potential performance gain of spatial-temporal processing relative to conventional spatial processing, for signals obeying a deterministic parametric model. The Cramer-Rao bound (CRB) on the estimates of the source directions of arrival (DOA) is used to quantify this gain. Spatial-temporal processing does not yield any such gain in the single source case, or for multiple coherent signals. However, significant gains can be achieved for multiple non-coherent signals.

We consider non-homogeneous 2-D signals which can be represented by a constant modulus polynomial-phase model. We develop a computationally efficient estimation algorithm for the parameters of this model. The algorithm is based on a phase differencing operator which is introduced in this paper. The basic properties of the operator are derived and used to develop the estimation algorithm.

In recent work the author (1990) studied the sensitivity of two direction finding algorithms (MUSIC and maximum likelihood) to gain and phase errors. Here, he generalizes the analysis to any consistent estimator based on the covariance matrix of the received data. The analysis is deterministic in the sense that the gain/phase perturbations are assumed to be fixed during the data collection interval; also, the exact covariance matrix of the received data is assumed to be known.>

A new direction finding algorithm is derived for multiple wideband signals received by an arbitrary array, and its performance is analyzed. Using an interpolation technique, a set of virtual arrays is generated, each for a direct frequency band, having the same array manifold. The covariance matrices of these arrays are added up to produce a composite covariance matrix. Direction of arrival estimates are obtained by eigendecomposition of this composite covariance matrix using the narrowband MUSIC algorithm or its variants. By choosing the virtual arrays to be linear and equally spaced, one can apply the root-MUSIC algorithm to this problem. Closed-form expressions for the asymptotic covariance matrix of the direction of arrival (DOA) estimation errors are derived using a perturbation analysis.>

The detection of transient signals is often performed in the time-frequency transform domain. An analytical framework is developed within which the performance of different detectors based on linear transforms can be easily compared, for different classes of signals. This framework is used to evaluate and compare the performance of detectors based on the Gabor (1946) transform and on the short-time Fourier transform.>

The performance of direction finding systems in a correlation signal environment utilizing diversely polarized antenna arrays is investigated. The Cramer-Rao bound (CRB) is used to evaluate the accuracy of the estimated directions of arrival. Compact closed form formulas are presented for the CRB corresponding to the joint estimation of the directions of arrival, signal covariance matrix, signal polarization parameters, and noise variance. The CRB is evaluated numerically for selected examples, to provide insights into the potential improvement in direction finding accuracy due to polarization diversity.>

The problem of detecting short-duration nonstationary signals, which are commonly referred to as transients, is addressed. Transients are characterized by a signal model containing some unknown parameters, and by a 'model mismatch' representing the difference between the model and the actual signal. Both linear and nonlinear signal models are considered. The transients are assumed to undergo a noninvertible linear transformation prior to the application of the detection algorithm. Examples of such transforms include the short-time Fourier transform, the Gabor transform, and the wavelet transform. Closed-form expressions are derived for the worst-case detection performance for all possible mismatch signals of a given energy. These expressions make it possible to evaluate and compare the performance of various transient detection algorithms, for both single-channel and multichannel data. Numerical examples comparing the performance of detectors based on the wavelet transform and the short-time Fourier transform are presented.>

The interpolated spatial smoothing algorithm is a computationally efficient method for estimating the directions of arrival (DOAs) of signals, some of which may be perfectly correlated. It extends the spatial smoothing method to arbitrary array geometrics. A performance analysis is outlined for this algorithm. Closed form expressions for the covariance matrix of the DOA estimation errors are derived using a perturbation analysis.>

A direction-finding technique is developed which uses the outputs of a virtual array computed from the real array using a linear interpolation procedure. The geometry of the virtual array is under the control of the designer. Using a linear virtual array, an extension of the root-MUSIC algorithm to arbitrary array geometries is developed.>

Two types of high-resolution direction-finding algorithms which use the fourth-order cumulants of the array data are derived. One is a MUSIC-like technique based on eigendecomposition of a suitably defined cumulant matrix. The other is an optimal (asymptotically minimum variance) estimator based on minimization of a certain cost function.>

An algorithm for detecting moving targets by imaging sensors and estimating their trajectories is proposed. The algorithm is based on directional filtering in the frequency domain, using a bank of filters for all possible target directions. The directional filtering effectively integrates the target signal, resulting in an improved signal-to-noise ratio. Working in the frequency domain facilitates a considerable reduction in computational requirements compared to time-domain algorithms. The algorithm is described in detail, and its false alarm and detection probabilities are analyzed.>

A first-order analysis of the sensitivity of the MUSIC algorithm to differences between the true and assumed array manifolds is presented. A formula is derived for computing the amount of phase error that will cause the algorithm to fail to resolve two sources, as a function of source separation and other system parameters. The failure threshold is shown to be proportional to the square of the source separation.>

The authors present several algorithms for estimating the parameters of MA (moving average) and ARMA (autoregressive moving average) non-Gaussian processes from sample high-order moments. These algorithms use explicitly the second-order statistics of the sample moments, which is estimated from the measurements. The asymptotically minimum-variance algorithms are shown, by numerical simulations, to perform close to theoretical predictions. The optimal weighted least-squares algorithms do not reach their theoretical performance, but they still offer some improvement over simpler algorithms. Since the computational load for the minimum variance algorithm is similar to that of the weighted least-squares algorithm, while its statistical accuracy is considerably higher, it is preferable to the weighted least-squares for most applications. The main disadvantage of the minimum variance algorithm is its more complex implementation (programming), especially the need for an iterative optimization procedure.>

An algorithm for blind adaptive equalization of digital communication channels for QAM (quadrature amplitude modulation) signals is derived. The algorithm uses the fourth-order statistical moments of the symbol sequence explicitly. The algorithm is linear, and is fundamentally different from common gradient-type algorithms intended for the same purpose.>

Results are presented concerning the performance of moving-average (MA) estimation algorithms based on high-order moments. A general lower bound is presented for the variance of estimates based on high-order sample moments. Then, an expression is given for the variance of weighted least-squares estimates, of the type recently reported in the literature. The existence of an optimal-weight matrix for such estimates is exhibited. The analytic results are verified by Monte-Carlo simulations for some specific test cases.>

An eigenstructure-based method for direction finding in the presence of sensor gain and phase uncertainties is presented. The method provides estimates of the directions of arrival (DOA) of all the radiating sources as well as calibration of the gain and phase of each sensor in the observing array. The technique is not limited to a specific array configuration and can be implemented in any eigenstructure-based DOA system to improve its performance.>

The authors are concerned with the classification of transient signals. Spectral ratio distance measures operating on the parametric spectra of the transient signals are used to perform classification. Performance of the classification algorithm is studied analytically and experimentally using both synthetic and real data. In addition to spectral features, the classification system also utilizes some side information such as transient duration and other temporal features. It was found that the symmetrized Itakura distance operating on high-order autoregressive spectra yields recognition rates of 87% on a library of 23 transients of six types.>

The authors provide an analytic performance evaluation of the errors of the direction-of-arrival estimates obtained by the MUSIC algorithm for uncorrelated sources. Explicit asymptotic formulas are given for the covariances of the estimates. The covariances are then compared to the Cramer-Rao lower bound. It is shown that, for a single source, the MUSIC algorithm is asymptotically efficient. For multiple sources, the algorithm is not efficient when the SNRs (signal-to-noise ratios) of all sources tend to infinity.>

Sensor location uncertainty can severely degrade the performance of direction finding systems. An exact maximum-likelihood method for simultaneously estimating directions of arrival (DOA) and sensor locations is developed to alleviate this problem. The case of nondisjoint sources, i.e., sources observed in the same frequency cell and at the same time, is emphasized. The algorithm is iterative and converges to the global maximum of the likelihood function is the initial conditions are good enough. Selected numerical examples demonstrate that the proposed technique is capable of correcting severe errors in the DOA estimates due to sensor location uncertainty.>

Recently, Kumaresan and Tufts (KT) presented a method for estimating the parameters of damped exponential waveforms in additive white noise. The KT method uses singular value decomposition (SVD) of the data matrix, with truncation and backward prediction to improve the accuracy of the estimates. The KT method was demonstrated to have a very good performance, in comparison with traditional methods used for the same problem. In this paper we provide a quantitative accuracy analysis of the KT method. The analysis is based on first order Taylor series approximations of the estimated parameters around their true values. Results of the analysis were illustrated by some numerical examples in [3]. These results confirm the good performance of the KT method, and show the effect of the user chosen parameters on the accuracy of the estimates.

This paper proposes to use the Gabor representation for the detection of transient signals with unknown arrival times. We introduce a one-sided exponential window function, which seems to the most appropriate for transient modeling. Explicit expressions for the Gabor coefficients are given for this window function. When the given signal is random, so are the coefficients. The second-order moments of the Gabor coefficients are computed for a white noise signal. These are then used to introduce a detection statistic based on the Gabor coefficients. The proposed detector is capable of separating transients having different arrival times, even in the case where their waveforms partially overlap.

A three step method for obtaining nearly maximum likelihood ARMA spectral estimates is presented. The computational complexity of the algorithm is comparable to Yule-Walker methods, but the method gives asymptotically efficient estimates. The implementation of the algorithm is discussed, and numerical examples are presented to illustrate its performance.

This paper proposes a new algorithm for detecting moving targets by mosaic sensors and estimating their trajectories. The algorithm is based on directional filtering in the frequency domain, using a bank of filters for all possible target directions. The proposed algorithm is described in detail, and its performance is illustrated by a numerical example.

A method is presented for acoustic passive multipath target tracking in the ocean. The method accounts for the inhomogenuity of the medium by integration of a prefilter into the tracking algorithm. The prefilter uses the output of an eigenray acoustic model computed over a two dimensional (2-D) grid of possible target positions (depth,range). The function is inverted and interpolated to form a transfer function from propagation time delay differences to range and depth. A three dimensional (3-D) maneuvering target estimator is used and its performance is evaluated. The results demonstrate the strong effects of the medium on the tracking concept.

The paper addresses the issue of estimating the parameters of stationary time series from finite or infinite sets of sample covariances. Several known results concerning finite sets of sample covariances are briefly reviewed. It is then shown that for certain Gaussian processes an efficient estimation of the process parameters is possible if the number of sample covariances used by the estimation algorithm tends to infinity. Next we examine stationary processes whose innovations are non-Gaussian. It is shown that the achievable accuracy of the parameter estimates based on the sample covariances is the same as that for the Gaussian case. Thus, parameter estimates based on sample covariances are insensitive to the process distribution and as a result, these estimates are generally statistically inefficient.

The asymptotic properties of the high-order Yule-Walker (HOYW) estimator of sinusoidal frequencies are analyzed. An explicit formula for the covariance matrix of the HOYW frequency estimation error is derived. The effects of the number of equations, the model order and a certain weighting matrix are investigated analytically. The optimally weighted HOYW estimator is described. The analysis motivates using an overdetermined and high-order Yule-Walker estimator.

The sample variance is commonly used to estimate the variance of stationary time series. When the second-order statistics of the process are known up to a scaling factor, this estimator is generally inefficient. In the case of an autoregressive (AR) process with unknown parameters, the sample variance is shown to be asymptotically efficient. However, the sample variance of a moving-average (MA) process with unknown parameters is generally an inefficient estimator. Closed-form expressions are derived for the Cramer-Rao hound associated with the variance estimation problem and for the variance of the sample-variance estimator, for both AR and MA processes.

The accuracy of ARMA spectral estimation methods is investigated. A lower bound for spectral estimates based on sample covariances is derived. Numerical examples are presented to illustrate the potential loss of accuracy when using sample covariances rather than the data.

The paper discusses the problem of detecting transient signals of unknown waveforms in white Gaussian noise. The signals are modeled as impulse responses of rational transfer functions with unknown parameters. A generalized likelihood ratio test (GLRT) is proposed and its statistical properties are analyzed. The GLRT involves constrained maximum likelihood estimation of the signal parameters. The distribution of the likelihood ratio is shown to be a central quadratic form under H0, and a noncentral quadratic form under H1.

This paper proves global convergence of the Constant-Modulus Algorithm (CMA) for the case of a real channel when the model order is equal to or greater than that of the channel (the so-called "model-complete" case). The analysis is based on an exact fourth-order Taylor series representation for the cost function minimized asymptotically by the CMA.

The Cramer-Rao lower bound (CRLB) provides a useful tool for evaluating the performance of parameter estimation techniques. Several techniques for the computation of the CRLB for ARMA and AR-plus-noise models are presented. It is shown that the CRLB can be expressed as an explicit function of the model parameters.

The problem of estimating the power spectral density of stationary time series when the measurements are not contiguous is considered. A new ARMA method is proposed for this problem, based on nonlinear optimization of a weighted squared error criterion. The method can handle either regularly or randomly missing observations. As a special case, the method can handle the problem of missing sample covariances. The computational complexity is modest compared to exact maximum likelihood estimation of the same parameters.

The problem of estimating sinusoidal or narrowband signals with a time-varying center frequency is considered. The signal parameters are estimated by fitting an autoregressive model with time-varying coefficients to the data. The overdetermined modified Yule-Walker equations are used to estimate a set of constant model parameters. Some numerical examples illustrating the behavior of the estimator are presented, and its accuracy aspects are briefly discussed.

Two algorithms are proposed for the continuous tracking of multipath delay. The methods use an adaptive delay line interpolated by a first-order filter. The interpolation coefficient is explicitly estimated using the recursive Gauss-Newton algorithm.

Using a theorem due to Whittle, simple derivations of the Cramer-Rao lower bound are presented for some delay estimation problems related to a single source, multiple sources, and multipath. The problem of Doppler estimation is briefly discussed.

An adaptive notch filter is derived by using a general prediction error framework. The proposed infinite impulse response filler has a special structure that guarantees the desired transfer characteristics. The filter coefficients are updated by a version of the recursive maximum likelihood algorithm. The convergence properties of the algorithm and its asymptotic behavior are discussed, and its performance is evaluated by simulation results.

The problem of estimating the power spectral density of stationary time series when the measurements are not contiguous is considered. A new autoregressive moving-average (ARMA) method is proposed for this problem, based on nonlinear optimization of a weighted-squared-error criterion. The method can handle either regularly or randomly missing observations. As a special case, the method can handle the problem of missing sample covariances. The computational complexity is modest compared to exact maximum likelihood estimation of the same parameters. The performance of the algorithm is illustrated by some numerical examples and is shown to be statistically efficient in these cases.

An efficient algorithm is presented for computing the covariance sequence of a multichannel autogregressive process represented by a set of reflection coefficients. The covariance sequence is shown to be the impulse response of a certain lattice filter related to the optimal predictor.

A parametric model is developed for signals received from a moving source. An estimation technique for the delay, Doppler and spectral parameters based on minimizing the error between the sample spectral density function and the spectral model is presented.

Digital processing of Synthetic Aperture Radar (SAR) data requires very large amounts of computation. In this paper we consider the feasibility of a VLSI implementation of the azimuth compression part of a real-time SAR processor. The area/size requirements of such a processor are evaluated by means of a prototype design. Two processing architectures are compared in terms of their suitability for VLSI implementation.

An iterative frequency domain technique is presented for estimating autoregressive moving-average (ARMA) parameters. The technique is based on minimizing the error between the sample power spectrum and a spectral model. The variance of the estimation error is shown to be close to the Cramer-Rao bound for some examples.

The problem of estimating spatial and spectral parameters of multiple sources by sensor arrays is considered. A parametric model is derived for the multichannel measurement vector. Tbe model parameters are the time delays and the autoregressive moving-average parameters of all the sources. A suboptimal parameter estimation procedure is proposed, and simulation results are presented to illustrate its performance.

The modified Yule-Walker technique of ARMA spectral estimation is shown to be a special case of the instrumental variable method of system identification. Several recursive instrumental variable algorithms are proposed for adaptive spectral estimation. An efficient lattice algorithm is presented for solving the modified Yule-Walker equations in the overdetermined case.

Estimating the time-delay between two received signals is formulated as a parameter estimation problem for a certain spectral model. The model parameters are computed by a two step procedure: The modified Yule-Walker equations are used to estimate the autoregressive coefficients of the source signal, and a frequency domain squared error criterion is minimized to provide the delay estimate. The proposed technique is compared by simulations to the parametric techniques of Hamon-Hannan and Chan-Riley-Plant and is found to have better performance.

A spectral estimation technique is presented for autoregressive moving-average (ARMA) processes. The technique is based on a parameter estimation technique known as the rec ursive maximum likelihood (RML) method. The recursive spectral estimation algorithm is presented and its asymptotic properties are discussed. Simulation results are presented to illustrate the performance of the estimator for various types of data.

Optimal filtering of narrowband signals corrupted by noise requires an infinite impulse response (IIR) filter. A recursive maximum likelihood algorithm is proposed for adaptively adjusting the coefficients of an IIR line enhancer. The performance of this adaptive IIR filter is analyzed and its advantages compared to finite impulse filters are discussed.

Many least-squares deconvolution algorithms are based on fitting an autoregressive model to the observed data. An implicit assumption in these algorithms is that the data was generated by a zero mean white noise process driving a linear filter. In many applications such as speech analysis or seismic deconvolution this assumption does not hold, and the performance of conventional deconvolution algorithms is, therefore, degraded. This paper presents an approach that attempts to alleviate this problem. The proposed technique combines a joint-process lattice filter with a nonlinear estimator of the driving process.

In many applications it is desirable to use filters with linear phase characteristics. This paper presents an approach for developing such filters for adaptive signal processing. Several least-squares algorithms are derived for adjusting the coefficients of an adaptive linear phase filter. It is shown that smoothing filters, rather than the commonly used prediction filters, are a more natural choice if linear phase characteristics are required.

All-zero filters in tapped-delay-line or lattice implementations are commonly used for speech deconvolution. The analysis techniques are mostly non-recursive, operating on a block of data at a time. In this paper we present a recursive pole-zero lattice form for speech analysis. The algorithm is based on the recently developed square-root normalized lattice forms. A comparison between the performance of AR and ARMA lattice filters is presented, using synthetic data. Preliminary results using speech data are also discussed.

Square-root normalized ladder algorithms provide an efficient recursive solution to the problem of multichannel autoregressive model fitting. The so-called covariance case is presented here, with emphasis on two special cases, namely the growing memory and sliding memory covariance ladder algorithms. New ladder form realizations for the identified models are presented, leading to convenient methods for computing the model parameters from estimated reflection coefficients. Several application areas of the new algorithms are discussed.

A new framework is presented for the problem of estimating source location from a set of time difference of arrival (TDOA) measurements. First the conditions under which the source coordinates are identifiable are derived and geometrically interpreted. Then it is shown that the source location can be found, when it is identifiable, by estimating the coefficients of a set of linear relationships, with the number of equations being equal to the number of stations.