The ability to determine the topology of worldwide telephone networks offers the promise of substantially improving their operating efficiency. This paper explores the problem of identifying the topology of a telephone network using observations made within the network. Using tomographic methods inspired by medical imaging, we consider measurements made by transmitting probes (e.g., phone calls) between network endpoints. In general, these measurements alone do not suffice to reconstruct a unique network, and in fact, there are many network topologies from which the set of measurements could have been generated. We propose a topology reconstruction algorithm based on correlating measurements collected at different internal nodes, and identify conditions under which correctness of the inferred topology is guaranteed.

This paper outlines the goals, structure, and technology elements of the INFINITY Project, a joint effort between university educators, industrial partners, and civic leaders to introduce a signal-processing-based engineering curriculum at the high school educational level. Implementation issues of the program are addressed, including class textbook and laboratory creation, teacher training, and online classroom support. Initial responses to the effort from teachers and students at fourteen different high schools are highly positive, and plans for further expansion of the program are given.

We address the deconvolution of MIMO linear mixtures. The approach is based on the construction of a hierarchical family of composite criteria involving CM criterion and second order statistics constraint. Although, the criteria are based on fourth order statistics, we give a complete proof of convergence of this structure. We show that each cost function leads to the restoration of one single source. Moreover the approach is naturally robust with respect to the channels order estimation. An adaptive algorithm is derived for the simultaneous estimation of all sources.

This paper examines the problem of demodulating time dispersed digitally modulated signals with particular emphasis on two aspects: the all-digital implementation of such demodulators and the use of "blind" algorithms for initializing the demodulator in the absence of explicit training by the transmitter. The factors impacting the architecture of blind demodulators are discussed, followed by a description of several practical implementations.

Provides an abstract for each of the exprt presentations and a brief professional biography of each presenter. The complete presentations were not made available for publication as part of the conference proceedings.

The length and complexity of a linear equalizer filter is highly dependent on the nature of the channel effects it must mitigate. The governing design rules are typically stated in terms of the channel's temporal characteristics, i.e. impulse response duration. Equalizer implementational complexity is a principal limiting factor for high bandwidth data communication applications, and consequently there is motivation for reexamining accepted design guidelines. It was previously demonstrated that for relatively benign conditions on the effective channel and transmitter pulse shaping, there exists a linear equalizer that perfectly mitigates intersymbol interference, and whose span matches that of the composite distortion. This paper examines the implications of the minimal-length equalizer in light of accepted design rules, and shows that a tangible loss in performance can be assigned to this complexity reduction. Actual line-of-sight microwave radio channels are used to demonstrate the nature of the performance loss.

The gradient-descent-based constant modulus algorithm (CMA) is commonly thought to converge much more slowly than its least mean square (LMS) counterpart, particularly for quadrature-amplitude-modulated (QAM) signals. Experiments shown in this paper indicate that in fact there is no substantial difference in the convergence rates of the two methods in the important special case of constant envelope signals (e.g., FM and FSK). For both CMA and LMS algorithms the convergence for nontrivially frequency-modulated signals depends on the same eigenvalue disparity problem that affects all gradient-descent techniques.

In the noise-free case, fractionally-spaced equalization by constant modulus algorithm (FSE-CMA) was shown to be robust to the channel identifiability condition. The effect of additive channel noise on FSE-CMB steady-state behavior is analyzed in terms of the perturbation of the equalizer convergence settings with respect to the convergence settings of the noise-free case. The loss of performance is characterized by the additional residual mean square error induced by the noise. We show that, in noisy conditions, FSE-CMA exhibits an interesting "smoothing effect" that leads to a trade-off between zero forcing (ZF) equalization and output noise enhancement (NE). We also show that the main loss is due to the NE and not to residual intersymbol interference.

Under certain conditions on the equalizer length and on the channel dynamics, temporal or spatial diversity allows one to achieve blind equalization perfectly by means of second-order statistics only. Loss of channel disparity causes perfect equalization to no longer be achievable. The achievable channel-equalizes combination then depends only on the part of the multichannel transfer function lacking disparity. We show that the fractionally spaced equalizer adapted by the constant modulus algorithm (FSE-CMA) still achieves "reasonable" equalization. Its performance equals that of the non-fractional CMA, with a slightly shorter baud-length equalizer than the FSE, applied to the part of the channel lacking disparity.

An efficient and effective method is presented for identification of the signal constellation of a voiceband or RF modem from a small number of data samples. The Radon (11917) transform based method has been shown to be more efficient than modulus based methods for identifying rectangular and hexagonal constellations with known carrier frequency and more tolerant of low SNR. When the carrier frequency is uncertain, the drift that the Radon transform method can tolerate depends on SNR, constellation size and rate of drift. The Radon transform based method is modified to greatly extend the range of carrier drift which can be tolerated with a modest increase in computation.>

Many communications receivers employ a cascade of two or more filters. They are typically designed and implemented separately, each accomplishing its own specific function. Some are fixed in their spectral characteristics, some are selectable, and some, such as data equalizers, are data-adaptive. The incredible growth in the speed and capability of digital signal processing (DSP) devices allows cost-effective digital implementations of many of these filters. This growing digitization would appear to permit the combination of many heretofore separate functions into the same physical filter realization, leading to yet further savings size, weight, power consumption, and cost. The authors examine the case of a digitally implemented tally implemented demodulator which combines the digital filters used for conversion from real sample to complex-valued samples, predecimation filtering, and adaptive equalization. It is shown that it is feasible to combine these functions into a single adaptive filter, but that computational savings are not always attained and that memory requirements almost always grow.>

Two algorithms are described for determining the signal constellation quadrature amplitude modulation in use on a (QAM) modulator, one based on Radon transform ideas and the other based on histogramming received signal radius information. The detection performances of the two specific algorithms are compared as a function of the signal-to-disturbance power ratio and the number of samples available. Generally speaking, it is shown that, when the carrier frequency is known with reasonable accuracy, the Radon transform method performs much better than the specific radius- or magnitude-only method tested, in the sense that for the same signal-to-disturbance ratio the former can reliably determine the constellation size and orientation with about one-tenth of the data samples required by the latter. It should be noted, however, that neither of the two algorithms has been optimized, nor have the comparisons taken into account other issues such as the details of particular applications, the relative computational requirements, and the added complexity needed by the Radon transform method to deal with uncertainties in the carrier frequency.>

It is shown that the transport delay present in the feedback path of a frequency-domain adaptive filter consisting of the combination of a CMA (constant modulus algorithm) with a transmultiplexer reduces the filter's maximum attainable convergence rate. An upper bound for the convergence rate is developed in terms of the amount of group delay and hardware pipeline delay present in the adaptive system under analysis. These limitations are shown to be fundamental in nature. They relate directly to the amount of spectral resolution desired in the adaptive filter and not to the particular nonlinear error function employed.>

In a previous work the authors examined the effect of imperfect equalization of a single-pole time-varying channel on the mean-squared error seen at the receiver in a communications system (M. G. Larimore, et al. Proc. ICASSP, pp. 2157-2180, 1987). Here, those results are extended by first addressing more complicated channel propagation models, and by examining the implications for digital transmission, i.e. the symbol error rate (SER) for a time-varying channel.>

The problem of identifying the type of signal constellation in use by a data modem without the active cooperation of the transmitting modulator is considered. This problem arises in so-called 'multi-drop' networks in which nodes may independently enter and leave the network. It is shown that Radon transform techniques can be used to recognize the type of rectangular or cross constellation in use, and to determine its orientation, even if the received signal-to-noise ratio is quite poor. This method is successful even when there is incomplete equalization and some residual carrier remains. With these estimates a decision-directed demodulator can be approximately initialized. The technique itself and examples of its performance over a range of constellation size, noise power levels, and number of input samples are described.>

This paper addresses the problem of adaptively equalizing a slowly time-varying linearly dispersive propagation channel. Owing to the time variation, the adaptive filter rarely has exactly the proper impulse response, and the quality of the output signal will almost always be poorer than that of a time-invariant, adaptively equalized channel. In order to gain a fundamental understanding of the underlying mechanisms, this paper examines the signal degradation introduced by the time-varying combination of a propagation channel with a single complex pole and an equalizer with a single adaptable complex zero. The effects on transmission mean squared error (and hence NPR and BER) are quantified in terms of the type of variation in the pole's position and the updating strategy of the zero. A set of curves is developed which quantifies the intuitive concept that more rapid updating produces progressively better matching between the channel and the equalizer and hence progressively better signal quality.

Increasing use of digital transmission under a wide variety of propagation conditions has prompted continued investigation of equalization techniques. Certain RF propagation effects can limit the performance of conventional adaptive equalization algorithms that require local synchronization. Discussed in this paper is a "predetection" adaptive technique, i.e. one that operates with no initial synchronization requirements. An adaptive filter based on the Constant Modulus Algorithm has proven useful for equalization of severe channel distortion. Observations regarding the utility and limitations of this algorithm are presented.

In communication applications, the Constant Modulus Algorithm has proven an effective means for adjusting coefficients of an adaptive filter. However, under certain conditions, it may generate a response that discriminates against the signal component in a preference for broadband noise for sustained periods. Work has been done toward characterizing and modeling this phenomenon.

A class of adaptive filtering algorithms has recently been developed which appears to be very useful for removing multipath and interference from communication signals. Practical application of this class of algorithms requires that approximate convergence time of the algorithms be predicted given signal properties and filter parameters. This paper develops a set of formulas which predict the filter convergence rate for one or more sinusoidal inputs. Simulation work verifies the accuracy of the formulas. The formulas are shown to provide an interesting contrast with the performance of an LMS-based adaptive filter.

Within current RF allocations, increases in communication traffic have resulted in a trend toward spectral crowding. Effects of adjacent channel interference have traditionally been minimized by geographically isolating channel assignments, or by over-spacing channel centers, inefficiencies which will be unacceptable for future needs. As a result of increased density of communication signals, future receiver hardware will require digital signal processing techniques for suppressing adjacent channel interference. Over the last few years, there have been research efforts aimed at solutions for this type of interference, especially in FM communications Of specific interest here is a mathematical structure developed in [7] addressing suppression of such interfering components after discrimination, where bandwidth reduction results in savings in computation requirements. Initial attempts to apply adaptive filtering techniques to this structure indicated that fundamental problems existed with the formulation. In this paper, this technique is closely scrutinized, and we demonstrate that significant baseband improvements are unlikely for general modulating signals, contrary to published claims.

This paper introduces the concept of designing adaptive filtering algorithms to exploit the presence of some invariant property possessed by a signal of interest. If this invariant property is disturbed by the interference or dispersion which degrades the signal's quality, and if the disturbance can be sensed, then it can be used to guide an adaptive algorithm. Certain analytic requirements on this type of algorithm are discussed as well as their relation to well-known least squares and MSE algorithms.

An earlier paper [1] presented the Constant Modulus Algorithm, an adaptive filtering algorithm which can be employed to remove interference and equalize propagation anomolies that impair communications via constant envelope signals, such as FM and QPSK. The technique presented in that paper employs complex waveforms and a FIR adaptive filter with complex coefficients so that signal properties such as the instantaneous amplitude (modulus) can be measured. Even so many practical applications of such equalization or interference reduction involve real signals and adaptive filters implemented with real rather than complex arithmetic. This paper presents a version of the constant modulus algorithm (CMA) which employs real signals and real arithemtic. The algorithm is developed in an evolutionary form from the version based on complex arithmetic. A key technical result is that certain error terms can be highly simplified due to the smoothing intrinsic to a gradient search algorithm.

An adaptive filter algorithm has been developed and introduced [1] for use with constant envelope waveforms, e.g., FM communication signals. It has proven capable of suppressing additive interferers as well as equalization, without the need for a priori statistical information. In this paper, aspects of dynamic convergence behavior are discussed, with conclusions supported by simulation.

Some applications of adaptive FIR filtering, such as multipath correction and interference rejection, can require very long impulse responses, and hence are often computationally expensive. This paper examines the concept of sparse filtering, that is, constraining a substantial number of the coefficients in an FIR filter to zero. This concept is particularly appropriate where a long impulse response duration is required but only a few degrees of freedom (e.g., notches) are needed. This paper examines two aspects, the effects of forcing coefficients to zero and the problem of how to select the taps to be "thinned."

In previous papers the authors have developed and characterized SHARF and HARF, algorithms for adaptive IIR digital filters. This paper considers a practical problem in which adaptive IIR filtering can be applied to advantage: the removal of multi-path effects from television video ("ghosting"). The problem is motivated, a solution based on adaptive IIR filtering is developed, and simulation results are presented. Also discussed are several analytical results developed in this inquiry, including the formulation and proof of a complex data and coefficient version of SHARF (and HARF), and a treatment of the effects of additive noise.

In previous work the authors have developed a class of provably convergent adaptive algorithms for digital IIR filters based on the concept of hyperstability. While this class of adaptive filters offers much promise in practical applications little has been done toward providing the characterization that would allow a designer to select filter order, adaptation constants, and other design parameters. This paper reports on the initial effort toward providing this information through an investigation of the local convergence behavior of SHARF, the simple hyperstable adaptive recursive filter.

The relatively cheap computation promised by the advent of advanced large-scale integrated (LSI) devices has made it possible to consider the use of complex joint parameter estimators in many signal processing applications. In certain cases however the performance of such an estimator is limited by the lack of "observability" of the signal generating process, where observability in this context implies the ability to discern all possible variations in the parameters contributing to a signal of interest. This concept is pursued by defining parameter observability and developing a mathematical test for it. It is then shown that the uniqueness of joint maximum likelilhood parameter estimates is directly related to the observability of tke signal generating process. With this background the observabiity of several typical signal generating structures is examined.

The application of an adaptive prediction error filter to autoregressive spectral analysis has been proposed by Lloyd Griffiths and is in use for the measurement of the frequencies of sinusoids It is well known that additive observation noise can bias these frequency estimates, thus limiting the useful applications of Griffiths technique to low noise environments. In the limit of white observation noise this paper develops a modified adaptive algorithm which incorporates a priori or measured noise power information so that the generated frequency estimates are unbiased. Unlike other bias removal techniques recently discussed, this technique can work with multiple sinusoids. The performance of this algorithm will be demonstrated through computer simulations and its convergence behavior will be discussed.

Several authors have suggested various methods of implementing an adaptive IIR filter. This paper suggests that hyperstability and associated positivity concepts can be used to determine the regions of assured convergence for adaptive IIR filter algorithms. This hyperstability interpretation is applied to the adaptive recursive LMS filter, for which it is demonstrated that convergence appears assured if the transfer function formed from the denominator (in z-1) of the transfer function effectively being modelled and a unity numerator is strictly positive real. Examples show that this plausibly sufficient condition, though not necessary, loosely approximates the region of applicability, which seriously restricts the utility of this attractively simple adaptive synthesis method. Slight modifications, suggested by the hyperstability approach, can favorably alter the positive-real "convergence" region with subsequent dramatic improvement in applicability and convergence speed.