Threshold parameter estimation in nonadditive non-Gaussian noise

Threshold or weak-signal locally optimum Bayes estimators (LOBEs) of signal parameters, where the observations are an arbitrary mixture of signal and noise, the latter being independent, are first derived for “simple” as well as quadratic cost functions under the assumption that the signal is present a priori. It is shown that the desired LOBEs are either a linear (simple cost function) or a nonlinear (quadratic cost function) functional of an associated locally optimum and asymptotically optimum Bayes detector. Second, explicit classes of (threshold) optimum estimators are obtained for both cost functions in the coherent as well as in the incoherent reception modes. Third, the general results are applied to amplitude estimation, where two examples are considered: (1) coherent amplitude estimation in multiplicative noise with simple cost function (SCF) and (2) incoherent amplitude estimation with quadratic cost function (QFC) of a narrowband signal arbitrarily mixed with noise. Moreover, explicit estimator structures are given together with desired properties (i.e. efficiency of the unconditional maximum likelihood (ML) estimator) and Bayes' risks. These properties are obtained by employing contiguity-a powerful concept in modern statistics-implied by the locally asymptotically normal character of the detection algorithms     

Minimum variance filters and mixed spectrum estimation

This paper presents a spectral density estimator based on a normalized minimum variance (MV) estimator as the one proposed by Lagunas. With an equivalent frequency resolution, this new estimator preserves the amplitude estimation lost in Lagunas one. This proposition comes from a theoretical study of MV filters that highlights this amplitude lost. Two signal types are taken into account: periodic deterministic signals (narrow-band spectral structures) and stationary random signals (broad-band spectral structures). Without selecting a smoothing window, the proposed estimator is an alternative to Fourier-based estimator and, without modeling the signal, it is a concurrent to high-resolution estimators.     

Optimal blind nonlinear least-squares carrier phase and frequency offset estimation for general QAM modulations

This paper introduces a family of blind feedforward nonlinear least-squares (NLS) estimators for joint estimation of the carrier phase and frequency offset of general quadrature amplitude modulated (QAM) transmissions. As an extension of the Viterbi and Viterbi (1983) estimator, a constellation-dependent optimal matched nonlinear estimator is derived such that its asymptotic (large sample) variance is minimized. A class of conventional monomial estimators is also proposed. The asymptotic performance of these estimators is established in closed-form expression and compared with the Cramer-Rao lower bound. A practical implementation of the optimal matched estimator, which is a computationally efficient approximation of the latter and exhibits negligible performance loss, is also derived. Finally, computer simulations are presented to corroborate the theoretical performance analysis and indicate that the proposed optimal matched nonlinear estimator improves significantly the performance of the classic fourth-power estimator.     

Optimal blind carrier recovery for MPSK burst transmissions

The paper introduces and analyzes the asymptotic (large sample) performance of a family of blind feedforward nonlinear least-squares (NLS) estimators for joint estimation of carrier phase, frequency offset, and Doppler rate for burst-mode phase-shift keying transmissions. An optimal or "matched" nonlinear estimator that exhibits the smallest asymptotic variance within the family of envisaged blind NLS estimators is developed. The asymptotic variance of these estimators is established in closed-form expression and shown to approach the Cramer-Rao lower bound of an unmodulated carrier at medium and high signal-to-noise ratios (SNR). Monomial nonlinear estimators that do not depend on the SNR are also introduced and shown to perform similarly to the SNR-dependent matched nonlinear estimator. Computer simulations are presented to corroborate the theoretical performance analysis.     

Covariance shaping least-squares estimation

A new linear estimator is proposed, which we refer to as the covariance shaping least-squares (CSLS) estimator, for estimating a set of unknown deterministic parameters, x, observed through a known linear transformation H and corrupted by additive noise. The CSLS estimator is a biased estimator directed at improving the performance of the traditional least-squares (LS) estimator by choosing the estimate of x to minimize the (weighted) total error variance in the observations subject to a constraint on the covariance of the estimation error so that we control the dynamic range and spectral shape of the covariance of the estimation error. The presented CSLS estimator is shown to achieve the Cramer-Rao lower bound for biased estimators. Furthermore, analysis of the mean-squared error (MSE) of both the CSLS estimator and the LS estimator demonstrates that the covariance of the estimation error can be chosen such that there is a threshold SNR below which the CSLS estimator yields a lower MSE than the LS estimator for all values of x. As we show, some of the well-known modifications of the LS estimator can be formulated as CSLS estimators. This allows us to interpret these estimators as the estimators that minimize the total error variance in the observations, among all linear estimators with the same covariance.     

Estimation of series system reliability for exponentially distributed component life times

The reliability of a series system with P components which have exponential life times is estimated using Type II censored samples. The series system reliability is a function of , where λ_j is the hazard rate constant for the j^th component, j=1,2,…P. An estimator of μ( ) which dominates the MLE in terms of the risk is derived. This improved estimator of μ( ) is used for estimating the series system reliability. Monte Carlo simulation is used to estimate the risks of the proposed estimators, and comparisons with the ML estimators are made.     

On the robustness of stride frequency estimation

The robustness of stride frequency estimation (location and spread) from stride period data is investigated using influence functions. Theoretical analysis reveals that stride frequency estimates by Stokes et al. and by direct calculation have unbounded influence functions and zero breakdown points, implying a lack of both local and global robustness. Comparison of estimates obtained from an ensemble of pathological gait stride time series shows that on average, differences among estimators are not statistically significant (p > or = 0.59) for long time series (hundreds of strides). Specific circumstances under which nonrobust estimates depart from robust estimates are investigated in terms of outlier influence. We recommend some heuristic rules-of-thumb for prudent selection of nonrobust stride frequency estimators for a given stride time series. The theoretical and empirical estimator comparisons suggest that in general, more research on estimator robustness in quantitative gait analysis is warranted.     

On the Design of Classifiers for Crop Inventories

Crop proportion estimators that use classifications of satellite data to correct, in an additive way, a given estimate acquired from ground observations are discussed. A linear version of these estimators is optimal, in terms of minimum variance, when the regression of the ground observations onto the satellite observations is linear. When this regression is not linear, but the reverse regression (satellite observations onto ground observations) is linear, the estimator is suboptimal but still has certain appealing variance properties. In this paper we derive expressions for those regressions which relate the intercepts and slopes to conditional classification probabilities. These expressions are then used to discuss the question of classifier designs that can lead to low-variance crop proportion estimates. Variance expressions for these estimates in terms of classifier omission and commission errors are also derived.     

Optimal recursive estimation with uncertain observation

In classical estimation theory, the observation is always assumed to contain the signal to be estimated. In practice, certain observations, or sequences of observations, may contain noise alone, only the probability of occurrence of such cases being available to the estimator. An example is trajectory tracking where the signal is first detected and then the estimator is allowed to process it for tracking purposes. However, any detection decision is associated with a false-alarm probability, which is the probability that the detected signal contains only noise. Minimum mean-square estimators are derived for two different forms of this problem; 1) when it is possible that the observation at any sample time contains signal or is noise alone, independent of the situation at any other sample, and 2) when the entire sequence of observations contains signal or is only noise. The estimators derived are of recursive form. A simple example is given for illustration.     

Wavelet-based estimators of scaling behavior

Various wavelet-based estimators of self-similarity or long-range dependence scaling exponent are studied extensively. These estimators mainly include the (bi)orthogonal wavelet estimators and the wavelet transform modulus maxima (WTMM) estimator. This study focuses both on short and long time-series. In the framework of fractional autoregressive integrated moving average (FARIMA) processes, we advocate the use of approximately adapted wavelet estimators. For these "ideal" processes, the scaling behavior actually extends down to the smallest scale, i.e., the sampling period of the time series, if an adapted decomposition is used. But in practical situations, there generally exists a cutoff scale below which the scaling behavior no longer holds. We test the robustness of the set of wavelet-based estimators with respect to that cutoff scale as well as to the specific density of the underlying law of the process. In all situations, the WTMM estimator is shown to be the best or among the best estimators in terms of the mean-squared error (MSE). We also compare the wavelet estimators with the detrended fluctuation analysis (DFA) estimator which was previously proved to be among the best estimators which are not wavelet-based estimators. The WTMM estimator turns out to be a very competitive estimator which can be further generalized to characterize multiscaling behavior     

Admissible, Minimax and Equivariant Estimators of Life Distributions from Type II Censored Samples

In life testing, the unique minimum variance unbiased estimator (MVUE) ? is often used when it exists. However it has been shown for certain distributions that an estimator of the form k? with uniformly smaller mean square error exists. Such extimators are derived here for a class of life distributions and are shown to be admissible, minimax, and (in most cases) equivariant. The underlying distribution from which the samples are drawn follows a generalized life model (GLM) which includes a model proposed by Epstein & Sobel, Weibull, exponential, and Rayleigh distributions as special cases. Results are also given for the Type II asymptotic distribution of largest values, Pareto, and limited distributions. In addition, admissible linear estimators of the form a? + b are obtained and it is shown that they are a form of locally best estimators for some portion of the parameter space. Both k? and a? + b could be used in nonrepetitive estimation problems where bias causes no difficulty.     

A Bayesian estimation approach for speech enhancement using hiddenMarkov models

A Bayesian estimation approach for enhancing speech signals which have been degraded by statistically independent additive noise is motivated and developed. In particular, minimum mean square error (MMSE) and maximum a posteriori (MAP) signal estimators are developed using hidden Markov models (HMMs) for the clean signal and the noise process. It is shown that the MMSE estimator comprises a weighted sum of conditional mean estimators for the composite states of the noisy signal, where the weights equal the posterior probabilities of the composite states given the noisy signal. The estimation of several spectral functionals of the clean signal such as the sample spectrum and the complex exponential of the phase is also considered. A gain-adapted MAP estimator is developed using the expectation-maximization algorithm. The theoretical performance of the MMSE estimator is discussed, and convergence of the MAP estimator is proved. Both the MMSE and MAP estimators are tested in enhancing speech signals degraded by white Gaussian noise at input signal-to-noise ratios of from 5 to 20 dB     

Robust methods for model order estimation

Model order estimation is a subject in time series analysis that deals with fitting a parametric model to a vector of observations. This paper focuses on several model order estimators known in the literature and examines their performance under small deviations of the probability distribution of the noise with respect to a nominal distribution assumed in the model. It is demonstrated that the standard estimators suffer from high sensitivity to deviations from the nominal distribution, and a drastic performance degradation is experienced. To overcome this problem, robust estimators that are insensitive to small deviations from the nominal distribution are developed. These estimators are based on a composition between model order estimation methods and robust statistical inference techniques known in the literature. In addition, a new estimator based on a locally best test for weak signals is presented both in nonrobust and robust versions. The proposed robust model order estimators are developed on a heuristic basis, and there is no claim of optimality, but experimental results indicate that they provide significant improvement over the standard estimators     

NAR estimators of spatial covariance matrices for adaptive arraydetection

The theory of noise-alone-reference (NAR) power estimation is extended to the estimation of spatial covariance matrices. A NAR covariance estimator insensitive to signal presence is derived. The SNR (signal-to-noise ratio) loss incurred by using this estimator is independent of the input SNR and is less than that encountered with the maximum likelihood covariance estimator given that the same number of uncorrelated snapshots is available to both estimators. The analysis assumes first a deterministic signal. The results are extended and generalized to signals with unknown parameters or random signals. For the random signal case, generalized and quasi-NAR covariance estimators are presented     

One-Order-Statistic Conditional Estimators of Shape Parameters of Limited and Pareto Distributions and Scale Parameters of Type II Asymptotic Distributions of Smallest and Largest Values

One-order-statistic estimators are derived for the shape parameter K of the limited distribution function F1(x, ?, K) = 1 - (? - x)K and the Pareto distribution function F2(y, ?, K) = 1 - (y - ?)-K, given the location parameters ? and ?, respectively. Similar estimators are derived for the scale parameters v1 and Vn, of the Type II asymptotic distributions of smallest and largest values, F3(w, v1, K) = 1 - exp[-(w/v1)-K] and F4(z, vn K) = exp [-(z/vn)-K], given the shape parameter K and assuming the location parameter is zero. The one-order-statistic estimators are K?|? = -1/Cmn 1n(? - xmn) for the limited distribution, K?|? = 1/Cmn 1n(ymn - ?) for the Pareto distribution, ?1|K = Cmn-1/K Wmn and ?n|K = Cmn-1/K Zn-m+1,n for the Type II distributions of smallest and largest values, where Xmn, Ymn, Wmn, Zmn are the mth order statistics of samples of size n from the respective distributions and Cmn is the coefficient for a one-order-statistic estimator of the scale parameter of an exponential distribution, which has been tabled in an earlier paper. It is shown that exact confidence bounds can be easily derived for these parameters using exact confidence bounds for the scale parameter of the exponential distribution. Use of the estimators is illustrated by numerical examples.     

System reliability estimation using simulation combined with network reductions

This paper proposes a dynamic Monte Carlo sampling method, called the conditional minimal cut set (COMICS) algorithm, where all arcs are not simulated at each trial and all minimal cut sets need not be given in advance. The proposed algorithm repeats simulating a minimal cut set composed of the arcs which originate from the (new) source node and reducing the network on the basis of the states of simulated arcs until the s-t connectedness is confirmed. We develop the importance sampling estimator, the total hazard estimator and the hazard importance sampling estimator which are all based on the proposed algorithm, and compare the performance of these simulation estimators. It is found that these estimators can significantly reduce the variance of the raw simulation estimator and the usual importance sampling estimator.     

Best linear unbiased estimator of the Rayleigh scale parameterbased on fairly large censored samples

This paper tabulates: coefficients and relative s-efficiency of the best linear unbiased estimator (BLUE) of the scale parameter of the Rayleigh distribution for type II censored samples of size N=20(5)40, r=0(1)4 (number of observations censored from the left) and s=0(1)4 (number of observations censored from the right); and ranks, coefficients, variances, and relative s-efficiencies of the BLUE of a based on a selected few order statistics (k) for sample size N=20(1)40 and k=2(1)4. These estimators have the minimum variance among the BLUE of a based on the same number of order statistics. As compared to maximum likelihood estimators (MLE) and approximate MLE, the s-efficiency of BLUE of ρ is very high. When estimating the parameter using only a few observations, the k-optimum BLUE of ρ is the only choice, as the MLE of ρ is not available. Therefore, these tables for coefficients of BLUE of ρ based on censored samples and few observations for moderately large samples, have many applications     

Robust estimation via stochastic approximation

It has been found that robust estimation of parameters may be obtained via recursive Robbins-Monro-type stochastic approximation (SA) algorithms. For the simple problem of estimating location, appropriate choices for the nonlinear transformation and gain constant of the algorithm lead to an asymptotically min-max robust estimator with respect to a familymathcal{F} (y_p,p)of symmetrical distributions having the same masspoutside[-y_p,y_p], 0 < p < 1. This estimator, referred to as thep-point estimator (PPE), has the additional striking property that the asymptotic variance is constant over the familymathcal{F}(Y_p,p). The PPE is also efficiency robust in large samples. Monte Carlo results indicate that small sample robustness may be obtained using both one-stage and two-stage procedures. The good small-sample results are obtained in the one-stage procedure by using an adaptive gain sequence, which is intuitively appealing as well as theoretically justifiable. Some possible extension of the SA approach are given for the problem of estimating a vector parameter. In addition, some aspects of the relationship between SA-type estimators and Huber'sM-estimators are given.