Although electrical busses are commonly used to connect computer subsystems, they are limited by loading to relatively small numbers of inputs and outputs. Optical interconnects may be used to connect much larger numbers of subsystems. This talk will describe the use of optical "slabs" as interconnects. Slabs have many more modes than fibers, permitting higher efficiency in broadcasting between many transmitters and many receivers. They also permit "mode division multiplexing," a way of transmitting multiple channels on the same slab. This may be combined with wavelength division multiplexing, so that the same optical equipment can support a very large number of different channels between many transmitters and receivers.

Common opinions respective to the nature of the reactive power, energy flow and oscillations, as well as the notion of the apparent power in single- and in three-phase systems are discussed. It is shown that some interpretations of powers and energy flow in linear, single-phase circuits are often generalized for more complex situations where these interpretations are no longer valid. Consequently, power phenomena in electrical systems are often misinterpreted. This relates to the reactive power which occurs in three-phase systems without energy oscillation between the supply source and the load, as well as it occurs in time-variant systems without energy storage capability. Also, it was demonstrated in the paper that the arithmetic and geometric apparent powers, commonly used in three-phase systems, do not characterize the supply loading correctly when the load is unbalanced.

Every few years we hear of a new technology that will revolutionize artificial intelligence (AI). After careful reflection it is found that the advance is within the framework of the Turing machine model and equivalent, in many cases, to existing statistical techniques. But this time, in quantum computing, we seem to be on the threshold of a real revolution--a "quantum" leap--because it is a true frontier beyond classical computing. But will these possibilities be realized any time soon? Classical computers work on classical logic and they may be viewed as an embodiment of classical physics. Quantum computers, on the other hand, are based on the superpositional logic of quantum mechanics, which is an entirely different paradigm. Conventional explanation sees consciousness arising as an emergent property of the classical computations taking place in the circuits of the brain, but this does not address the question of how thoughts and feelings arise. If brains perform quantum processing, then this may be the secret behind consciousness. Furthermore, it may explain several puzzling features of animal/human intelligence, as also provide a new direction to develop AI machines. In this brief survey I present the rationale for the convergence between quantum computing and AI, and prospects for the realization of the technology.

Wireless asynchronous transfer mode (WATM) has been proposed as a potential solution for broadband integrated wireless networks. Extending ATM services to the wireless environment allows for allocation of bandwidth on demand, quality of service (QoS) guarantees, as well as seamless transmission of ATM cells between wired and wireless networks without the need for protocol conversion.

ATM is designed for a fiber-based network where bandwidth is plentiful and can be allocated dynamically based on users' need, and transmission quality is very good (bit-error-rates of 10^-9). In ATM provision of quality of service (QoS) is achieved through call admission control and packet scheduling disciplines. In wireless ATM, however, due to the nature of the radio channel, design of call admission control and packet scheduling policies is more challenging. In the radio channel bandwidth is limited and must be shared by all users and the transmission quality is usually poor.

In this talk we derive the achievable QoS region for a family of channel-adaptive work-conserving, earliest-due-date (WC-EDD) scheduling policies operating in a time division multiple access (TDMA) system. The radio channel is modeled by a two state Gibert/Elliot channel model (a good state with no errors and a bad state with high bit error rate). The scheduling policies are adapted so as not to schedule transmissions for the terminals whose radio channels are in a bad state. From the achievable QoS region we can then derive the call admission control algorithm for the WATM network.

The electric power industry is the last of the major industries in the United States to undergo deregulation. Dr.\ Denny will discuss the past performance of the industry, the motivation for deregulation, the industry restructuring that is now taking place, and the future technical challenges that must be addressed.

The Cajun Electric Company and many other electric utility companies are making changes in their operations to prepare for industry deregulation, to utilize new technologies, and to become more efficient and competitive. Mr.\ David Breaux will address these subjects from his perspective as an engineer at the Cajun Electric Company.

The inverse of a dynamical system is intuitively understood to be a second dynamical system which when cascaded with the original system, produces as its output the input to the original system. This may be simply related to the exact tracking problem. Solutions have been obtained for certain cases but it is extremely difficult, in general, to extend these results to linear distributed parameter systems.

We consider a model for the approximate inverse of an input-output system represented by the wave equation in n dimensions, with a boundary input and distributed output. We develop two models: In the first model, we approximate the system and invert it calling this model the inverse of the approximated system. In the second, we invert the system first, then approximate, calling this model the approximated inverse of the system.

We also establish the input-output relation of the inverse system using Fourier series. By truncation of the transfer function, an approximation is obtained. To demonstrate the relative accuracy of the approximations, simulation examples are presented.

The flight control of NASA's X-33 Reusable Launch Vehicle (RLV) poses a challenge to conventional gain-scheduled flight controllers due to its large attitude maneuvers from liftoff to orbit and reentry. In addition, a wide range of uncertainties in vehicle handling qualities and disturbances must be accommodated by the attitude control system. Nonlinear tracking and decoupling control by trajectory linearization can be viewed as the ideal gain-scheduling controller designed at every point on the flight trajectory. Therefore it provides robust stability and performance at all stages of flight without interpolation of controller gains, and eliminates costly controller redesigns due to minor airframe alteration or mission reconfiguration. A prototype X-33 ascent flight controller was designed using the trajectory linearization method and tested with 3-DOF and 6-DOF simulations at the NASA Marshall Space Flight Center. It is noted that the 6-DOF results were obtained from the 3-DOF design with only a few hours of tuning, which demonstrates the inherent robustness of the design technique. It is this "plug-and-play" feature that is much needed by NASA for the development, test and routine operations of RLVs. Plans for further research will also be presented.

Instantaneous learning is a desirable feature in neural networks. This type of learning enables the network to be trained very quickly, typically in just one pass of the training set as opposed to hundreds or even thousands of passes for networks trained by an iterative process, such as the error backpropagation (BP) algorithm. This talk discusses several types of neural networks with the instantaneous learning property, including the CC4 Corner Classification neural network. Some comparison of generalization performance is presented.

It is well known that feedback can be introduced to stabilize an unstable system, to attenuate the response of a system to disturbance, and to reduce the effect of plant parameter variations and modeling error. On the other hand, feedback design is also known to be contingent on many performance considerations and physical constraints, and use of feedback may suffer from certain serious potential disadvantages. Design difficulty may arise due to the very nature of feedback structure as well as plant properties, which generally manifests itself as limitations on achievable performance goals, and tradeoffs to be made between conflicting design objectives. An important step in the design process, therefore, is to analyze how plant characteristics may inherently impose constraints upon design and thus may fundamentally limit the level of achievable performance. This task has become even more tangible nowadays, as current control design theory and practice relies heavily on optimization-based numerical routines and tools.

In this talk I will discuss classical as well as new results in performance limitation studies. The first part of the talk will concentrate on reviewing the classical work by Bode and its more recent extensions for single-input single-output systems. Interpretations of these results from control perspectives will be particularly emphasized. The second part will focus on multivariable generalizations of Bode and Poisson type integrals, together with a number of intrinsic limits on the best achievable performance measured by sensitivity and complementary sensitivity functions. These developments lead to the discovery of new phenomena which have no analog in single-input single-output systems. A central and prevailing theme is that in a multivariable system performance and design limitation intrinsically depend upon the locations as well as directions of nonminimum phase zeros and unstable poles, and in particular, on the mutual orientation of the zero and pole directions.