David Starer

Artificial Neural Nets: Cerebrally Smart, but Lamentably Dumb

is an analyst of investment technology with Lend Lease Investment Management in Sydney, Australia. ME Starer holds a B.Sc. in electrical engineeringfrom the University o f Cape Town, and M.S., M.Phil., and Ph. D. degrees in electrical engineering +om Yale University. His primary interest is the application of stochastic signal processing and control engineering techniques to investment management, and he has published widely in the signal processing literature.

Passive localization of near-field sources by path following

A new algorithm for passively estimating the ranges and bearings of multiple narrow-band sources using a uniform linear sensor array is presented. The algorithm is computationally efficient and converges globally. It minimizes the MUSIC cost function subject to geometrical constraints imposed by the curvature of the received wavefronts. The estimation problem is reduced to one of solving a set of two coupled 2D polynomial equations. The proposed algorithm solves this nonlinear problem using a modification of the path-following (or homotopy) method. For an array having m sensors, the algorithm reduces the global 2D search over range and bearing to 2(m/spl minus/1) independent 1D searches. This imparts a high degree of parallelism that can be exploited to obtain source location estimates very efficiently. >

Adaptive pole estimation

AN adaptive algorithm is developed for online estimation of the poles of autoregressive (AR) processes. The method estimates the poles directly from the data without intermediate estimation of the AR coefficients or polynomial factorization. It converges rapidly, is computationally efficient, and attains the Cramer-Rao bound (CRB) asymptotically. A closed-form expression for the asymptotic CRB is provided. Convergence to the true solution is proved, and methods are discussed for extending the algorithm for use with more general (e.g. autoregressive moving-average) models. Numerical examples are presented to demonstrate the performance of the algorithm. >

Long-Short Portfolio Management: An Integrated Approach

With the freedom to sell short, an investor can benefit from stocks with negative expected returns as well as from those with positive expected returns. The authors explain that the benefits of combining short positions with long positions in a portfolio context, however, depend critically on the way the portfolio is constructed. Only an integrated optimization that considers the expected returns, risks, and correlations of all securities simultaneously can maximize the investors ability to trade off risk and return for the best possible performance. This holds true whether or not the long–short portfolio is managed relative to an underlying asset class benchmark. Despite the incremental costs associated with shorting, the authors argue that a long–short portfolio, with its enhanced flexibility, can be expected to perform better than a long–only portfolio based on the same set of insights.

Direction-of-arrival estimation in applications with multipath and few snapshots

This paper analyzes two direction-of-arrival (DOA) estimation algorithms used in the presence of multipath propagation and with very few snapshots. The conditional maximum likelihood (CML) algorithm and the method of direction estimation (MODE) are discussed. The estimates provided by these algorithms are shown to coincide for large number of snapshots or large signal-to-noise ratio. Necessary and sufficient conditions are derived for the algorithms to yield unique estimates. It is shown that their uniqueness conditions coincide with the minimal uniqueness condition on the array, that is independent of the algorithm used (if the array does not satisfy this minimal condition, no DOA estimation method can give unique estimates). Numerical examples are presented to demonstrate the theoretical results.

Path-following algorithm for passive localization of near-field sources

Presents a new algorithm for passive localization of multiple narrow-band near-field sources using a uniform linear array. The algorithm is computationally efficient and has global convergence. It minimizes the MUSIC cost function with respect to the source ranges and bearings, subject to geometrical constraints imposed by the curvature of the received wavefronts. It is shown that, when the cost function is expanded in a two-dimensional power series in the Fresnel region, the estimation problem reduces to one of solving a set of two coupled two-dimensional polynomial equations. The proposed algorithm solves this nonlinear problem using a modification of the path-following (or homotopy) method.<<ETX>>

Adaptive polynomial factorization by coefficient matching

A polynomial factorization algorithm is presented which updates all roots simultaneously and efficiently in response to coefficient perturbations. The algorithm requires approximately 2n/sup 2/ complex floating point operations to update all roots of nth order polynomial. Close to the true root vector, the algorithms convergence rate is quadratic. The root update requires only the solution of two sets of structured linear equations and a convolution. The algorithm can be used to track the roots of time-varying polynomials which is useful for application in adaptive signal processing. >

Polynomial factorization algorithms for adaptive root estimation

The authors present two new polynomial factorization algorithms suitable for use in adaptive signal processing applications where it is required to track the movements of roots. Their distinguishing feature is that they provide methods for updating the roots optimally (and efficiently) in response to coefficient perturbations. This is useful, for example, for online estimation and tracking of time-varying roots. A Gauss-Newton type algorithm that requires approximately 2n/sup 2/ floating-point operations to update all roots of an nth-order polynomial is presented. An eigenvalue-based method that requires approximately 6n floating-point operations per individual root update is also proposed. Close to the true root vector, the rate of convergence of both algorithms is quadratic or faster. Unlike available direct pole estimation algorithms, the methods proposed here are not restricted to estimation of the roots of ARMA (autoregressive moving average)-like processes, but can be used in any situation where online estimates of polynomial coefficients are available.<<ETX>>

Consistency of direction-of-arrival estimation with multipath and few snapshots

An analysis is made of the consistency of two direction-of-arrival estimation algorithms used in the presence of multipath propagation and with very few snapshots. The conditional maximum-likelihood (CML) algorithm and the method of direction estimation (MODE) are discussed. The cost functions of these algorithms are shown to coincide for a very large number of snapshots or very large signal-to-noise ratio. Necessary and sufficient conditions are derived for the algorithms to yield unique estimates. It is shown that their uniqueness conditions coincide with fundamental uniqueness conditions for the array that are independent of the algorithm used.<<ETX>>