C. Richard Johnson, Jr., Ph. D. - US grants

Adaptive control algorithms presently dominating the theoretical literature have recently been demonstrated to behave poorly when assumptions used to prove their idealized behavior are violated. Though such violations undoubtedly occur in practice, adaptive controllers have been repeatedly documented as successful in a variety of applications. This research will focus on eliminating the theory-practice gap due to the impractical, current theoretical limitation that the controller structure be sufficient to arbitrarily modify all modes of the plant behavior. Thus, the aim of this research is the development of a theoretical framework for describing, and subsequently improving, the robustness of restricted complexity adaptive controllers. The major directions to be pursued are the characterization of undesirable parameter estimate drift in restricted complexity adaptive control as well as generation and evaluation of algorithm fixes given improved understanding of stability/instability mechanisms in existing algorithms.

Blind adaptive equalizers are key receiver components in high performance, high density, digital communications systems. In less than fifteen years since the conception of blind adaptive equalization, various algorithms, frequently tailored to a particular source constellation class, have been proposed, implemented, tested and installed. Convergence, despite an arbitrarily poor initialization, to a parameterization with nearly optimum achievable performance is a seemingly desirable attribute of an acceptable blind adaptive equalizer algorithm, or of any parameter adaptive system for that matter. Apparently no blind adaptive algorithm for a finite-impulse-response equalizer is known to exist that can promise this, even for an idealized situation with perfect recovery capability and the simplest digital source of plus and minus ones. yet, no theorem proves that such a globally convergent algorithm does not exist, even in an idealized setting. Indeed, the available literature pertinent to this issue leaves a mixed impression due to some (seemingly) contradictory claims. The research addresses this existence question, at least among the class of schemes with memoryless error functions which subsumes the most popular blind adaptive equalizers currently in use. Within the proscribed class of applications and algorithms that do not posses admissibility (i.e., global asymptotic optimality), the severity of the practical consequences of this inadmissibility will be investigated by examining the geometry of the associated error surface. For schemes lacking global asymptotic optimality, the utility of equalizer parameter initialization schemes will also be investigated. Finally, the potential for admissibility of "new" schemes that are outside the class established which lacks global asymptotic optimality will be investigated, e.g., adaptive algorithms that utilize error functions with memory.

There are many practical engineering applications which can be described as an adaptive feedback system. The distinguishing feature of an adaptive feedback system is that one component in the feedback loop has its parameters adapted according to a prescribed algorithm. Adaptive control in manufacturing systems and adaptive echo cancellers in telephone lines are two such examples. Adaptive feedback systems are nonlinear because the recursive parameter adaption laws are driven by signals that are functions of the past adapted parameters. Due to this nonlinear nature, different selections of design parameters, as well as different selections of adaptive algorithms, bring about qualitatively different types of behavior. We propose to initiate an effort to discern and classify the types of behavior from a system dynamic point of view. The tools to be employed are primarily standard dynamics analysis techniques that we, and others, have utilized before for adaptive systems analysis. The expected significance of a carefully composed classification is in its provision of the fundamental knowledge for the development of needed qualitative design guidelines, e.g. leakage factor selection for avoidance of bounded instabilities such as bursting. Good design is always preceded by good understanding of the problem and its "solution" candidates. A well-structured taxonomy is critical to well-informed design choices. Such a behavior/design-oriented knowledge base will be vital for the future success of emerging intelligent controllers. To contribute to such an expansion of adaptive feedback systems theory, we propose to clarify and propound our notion of a qualitative classification of the behavior of poorly excited adaptive feedback systems, and to develop some prototypes for elements of this nascent taxonomy.

The central theme of this research is the development of a fuller theoretical description of the behavior of memoryless-error- function (or Bussgang) type blind equalization algorithms, in particular CMA, as used in practice with finite-length, fractional spacing, higher-order (non-constant modulus, non-uniformly distributed) source constellations, (deterministic and stochastic) correlated source time series, and modest channel noise. The initial target is the behavior of fractionally-spaced CMA with a correlated (or periodic) source, which appears capable of causing misbehavior in practice. We plan to use analytical tools from topologically based, algebraic-geometry (in particular Bernstein's theorem, relaxation methods, and Morse theory) non typically associated with the study of adaptive filters in addition to averaging theory tools popular in adaptive systems analysis. The ultimate intent is to convert an expanded theory into situation dependent design guidelines for stepsize, length, fractional- spacing, initialization selection, and failure recovery tricks for the use of CMA as a start-up scheme (with a subsequent switch to decision-direction) for high data throughput applications. We plan to translate our analytical advances on CMA to fractionally-spaced realizations of other blind equalizer algorithms of the memoryless error function class (including refinements to CMA), for which actual operating data can be obtained industrially as for CMA. We also plan to use our behavior insights into CMA to help provide fair comparisons on actual operating data to competing schemes such as symbol-spaced decision feedback equalizers and similar complexity algorithms based on second-order correlation statistics of single-input, multiple- output models of fractionally-spaced equalization.

9528363 Johnson We propose a research program in the active control of acoustic noise (and related problems) using adaptive identification and control methodologies. The motivating applications are the control of acoustic noise in a one dimensional or duct noise scenario (e.g. heating, ventilation and air conditioning or HVAC applications) as well as the control of noise in a three dimensional or enclosure scenario (e.g. automobile or helicopter interiors). Open problems to be addressed in this research program concern (a) the systematic stability analysis of adaptive techniques used in the presence of intrinsic feedback loops which are caused in these applications by acoustic feedback and not by the usual closed loop control configurations assumed in adaptive control theory and (b) the development and efficient implementation of adaptive filters especially in the MIMO enclosure problem. The research program will clarify the pertinent aspects of adaptive control theory and practice and thus allow the systematic application of adaptive techniques. The research program will be conducted in collaboration with and will utilize experimental facilities of the United Technologies Research Center in East Hartford, CT. It is expected that the multidisciplinary team thus assembled by this collaboration will allow for wide-ranging and practical research results to be achieved and broadly disseminated thus making contributions to both education and commercial interests. ***

This project supports -- under the ``Faculty in Industry'' activity of NSF's GOALI program -- the visit of Professor C. Richard Johnson, Jr. (School of Electrical Engineering, Cornell University) to Applied Signal Technology (Sunnyvale, CA) for the 12-month period from August 1998 to July 1999. As co-PI, Dr. Michael G. Larimore of Applied Signal Technology monitors, expedites, and collaborates in all of the major activities of the visit, including presentation of a short course on adaptive equalizer theory and design, development of a monograph on blind equalizer design (based on over 12 years of experience at Applied Signal Technology and 10 years of theoretical work at Cornell), and participation in a product development team utilizing adaptive digital signal processing in an emerging communication system. The benefits of this bilateral technology and pedagogy exchange include a heightening of the relevance (and commensurate shrinkage of the theory- practice gap) in Professor Johnson's future research and teaching efforts and broad dispersion of course and reference materials on adaptive equalization theory and design tuned to the needs of the practicing communication systems development engineer.

9811297
Johnson
Wireless systems for mobile telephony and data transmission are currently undergoing very rapid development due to recent international agreements opening new regions in the radio spectrum to such services. These new systems, so-called third-generation wireless systems, will incorporate considerable signal-processiilg intelligence in order to provide advanced services such as multimedia. In order to make optimal use of available bandwidth for these services and to provide maximal flexibility, many such systems will operate as multiple-access systems, in which channel bandwidth is shared my many users on a random-access basis using protocols such as code division multipleaccess (CDMIA) signaling. Moreover, in order to support the high data rates inherent in such services, ratios of bit rates to bandwidths will be pushed to their limits. In general, wireless channels can be be very hostile media through which to communicate. Physical impairments such as multiple-access interference, co-channel interference, multipath transmission, amplitude fading, and dispersion due to limited bandwidth, all contribute to make it difficult to transmit data reliably and quickly through wireless channels. Moreover, the dynamism resulting from user mobility and the above-noted random-access nature of mobile channels, amplify the effects of these impairments, and make them much more difficult to ameliorate. Solutions to these difficulties lie in the use of
advanced adaptive systems to perform advanced signal processing functions, and this proposal addresses such solutions.

Specifically, we plan to consider three general research problems within the area of adaptive equalization of multiple-access wireless channels: optimal methods for joint adaptive equalization and multiuser detection; adaptive linear and nonlinear methods for joint adaptive equalization and multiuser detection; and iterative methods for joint adaptive equalization and multiuser detection on coded channels. By addressing these research issues we can answer key questions about the equalization of wireless multiple-access channels. In particular, the consideration of optimal methods will establish baseline results by which other methods can be benchmarked. Furthermore, by considering low complexity adaptive linear and nonlinear methods, we can develop an understanding of mechanisms by which adaptivity can brought to bear on the problem at hand, and moreoever we. can examine the dynamics of adaptation in a tractable setting. And, finally, by consideration of iterative, and other adaptive nonlinear methods, we can seek adaptive techniques that can approach the ideals - of high performance and low complexity - established via the analytical results of the two other areas.

In order to accomplish these goals, it is necessary to take a novel approach to these problems, and in particular to synthesize technologies from disparate areas. Consequently, this studv will be conducted as a joint project among three institutions - ALGOREX, Inc., Cornell University, and Princeton University - bringing together considerable separate expertise in the fields of equalization, interference suppression, and adaptive system design to facilitate the necessary synthesis of ideas.
***

0233127
Johnson

This three-year award for US-France collaboration in wireless telecommunications involves students and researchers at Cornell University and the French National Institute for Telecommunications. C. Richard Johnson in the US and Phillip Regalia in France lead this cooperative project on the theory, design, and pedagogy of broadband telecommunications receivers. As communication systems become sophisticated and hardware advances allow for more complex algorithms in telecommunications receivers, these systems will become more critical to emerging applications. The investigators will address four topics: (1) adaptive multi-carrier equalization; (2) ultra wideband receivers; (3) turbo equalization; and (4) equalizer initialization strategies. The investigators have complementary expertise in adaptive system theory for control, signal processing and communications applications. Their expertise will be used in studies of the interplay of broadband receiver subsystems such as synchronization, equalization and coding. The French expertise in joint coding and equalization has potential for adaptation to wireless personal area networks under development in the US. Similarly, US expertise in multi-carrier and ultra wideband could be applied to problems of inefficient spectral band in local wireless links in Europe.

This award represents the US side of joint proposals to the NSF and the French National Center for Scientific Research (CNRS). NSF will cover travel funds and living expenses for the US investigator and graduate student. CNRS will support the visits of the French researchers and graduate students to the United States.

ABSTRACT
0310023
Cornell University
C. Richard Johnson

The Cornell University Blind Equalization Research Group (C.U. BERG) under the direction of Professor C. Richard Johnson, Jr. has spent the past decade and a half researching blind and semi-blind adaptive algorithms used in communication receiver design, usually for blind equalization and blind synchronization. The current focus of the BERG is the Broadband Adaptive Receiver Design (BARD) Project, which aims to develop and study adaptive algorithms for emerging broadband last-mile communications applications. One manifestation of thisw ork isthe BERG'scurren t e.ort in the development of blind, adaptive algorithmsfor
channel shortening.
Channel shortening is a generalization of equalization, and it can be applied to the design of receivers for multicarrier communication systems, the design of reduced-complexity maximum likelihood sequence estimation (MLSE), and the design of interference suppression for multiuser detectors. The BARD Project is taking an adaptive system-theory approach to the (blind) channel shortening problem. Through thisapproac h, the BARD Project hasrecen tly produced several blind, adaptive algorithms, leading to receivers with improved performance in timevarying environments, and receivers with greatly reduced complexity requirements. In addition, the BARD Project has analytically characterized several of the existing non-adaptive approaches, leading to both explanations of their (poor) performance in certain (time-invariant) situations, and dramatic reductionsof their implementation complexity.

PROJECT ABSTRACT
"Counting Van Gogh and Vermeer"

Automated thread counting algorithms, in development since 2007 by Professors Rick Johnson (Cornell University) and Don Johnson (Rice University), are poised for a profound impact on the practice of technical art history. This work is a pioneering effort in the emerging application of signal and image processing to the analysis of paintings. The approach is based on the utilization of spectral analysis of x-rays of paintings. X-rays can display the periodic intensity pattern due to the greater thickness of the lead white ground between the canvas threads. With the addition of innovative application-specific signal processing, local peaks in the spectrum of the x-ray data can reveal the numbers of threads per centimeter, i.e. a thread count, separately for (nearly) vertically and (nearly) horizontally oriented threads. Such thread counts have been used previously, but their manual acquisition proved too costly to be done thoroughly. The introduction of the capability to assemble previously unthinkable thread density and angle pattern maps of high detail across the entire surface of a painting leads to correlation-based identification of canvases sharing the same pattern in thread density variations. Such weave maps and matches can now be assembled across all of the paintings on canvas of a single artist or school, thereby significantly extending the art historian's capabilities in, for example, dating and attribution. The scale, breadth, and depth of such thread-counting projects represent a bold leap in the capabilities of technical art history. This grant supports the analysis phase of the projects of such grand scope: the thread counting and subsequent weave matching among (i) all of the paintings by Vincent van Gogh (for which data can be obtained by the end of 2010) and (ii) all of the paintings by the Delft School during the career of Johannes Vermeer (for which data can be obtained by the end of 2010). This grant places Professors Rick and Don Johnson in the center of the data fusion, data analysis and database creation activities that are scheduled to occur in Amsterdam during the spring of 2011. The archives being established of thread count reports, including weave density maps providing fingerprints for weave matching and angle maps for characterizing cusping, form a groundbreaking resource at a time when museums are just beginning to address the technological and cultural barriers to technical data sharing among museums and collaborating researchers outside the museum. This project is part of a long-term effort that aims to expand the utility of thread counting from x-rays to all suitable oil paintings on canvas, and to photos of unlined backs of old master paintings and of the raw canvas, for example, of the modern colorfield painters and of densely-woven, multi-pattern fabrics prominent in the design and decorative arts.