An image coding scheme combining subband coding and the established energy compaction technique of the discrete cosine transform (DCT) with trellis coded quantization (TCQ) is introduced. Image spectrums are split into 16 subband images using a quadrature mirror filter bank, and the DCT is performed on the lowest subband. The data is quantized using TCQ, transmitted and recombined at the receiver. It is shown that quantizing the subband data with TCQ decreases the mean-squared error (MSE) incurred in the quantization step, versus that of a Lloyd-Max scalar quantizer.

Dynamic bandwidth allocation plays an important role in integrated services networks. Effective bandwidth management can improve the overall performance of the network significantly. In integrated services networks, different types of traffic coexist. To serve these different types of traffic satisfactorily, hybrid switching is commonly used. Hybrid switching is an integration of circuit switching and packet switching in the same network. In this thesis, dynamic bandwidth allocation is studied to provide satisfactory hybrid switching performance. Specifically, we will study how to allocate bandwidths for continuous bit rate (CBR) traffic served by circuit switching and variable bit rate traffic (VBR) served by packet switching. The performance measure is based on the blocking probability of CBR traffic and the time delay of VBR traffic. Therefore, the objective of the dynamic bandwidth allocation is to maintain balance between blocking probability and delay under various traffic conditions. To accomplish this, the dynamic bandwidth allocation algorithms consists of traffic intensity estimation, traffic performance prediction, and performance optimization. A simulation is written to demonstrate the algorithm for various traffic scenarios.

An investigation into the feasibility of designing a monolithic high frequency voltage controlled oscillator is performed. With design constraints of an oscillator Q between 5 and 10 and minimal chip area, the resonant LC tank and negative resistive cross coupled differential amplifier circuit is analyzed and design guidelines are developed. Analysis of the circuit encompasses both linear and non-linear modes of operation of the circuit, predicts the fundamental frequency of oscillation, and highlights design limitations for the resonant elements in terms of meeting the Q specifications at higher frequencies. Experimental results on a fixed frequency version of the VCO circuit yielded good agreement with theoretical analysis. For the parameters tested, the error was on the order of 10% in most cases.

The arrival of high-speed, large-area IC's forces development of packages with high-density, small cross-sectional wiring. This trend is particularly clear for packages housing many chips (MCM's). It is shown how electrical, thermal, and fabricational constraints control line geometry in MCM's forcing the use of lossy lines. To obtain high speeds with these lossy lines, a generalized impedance-matched design is to combine load, line, and driver for overshoot-controlled performance. Using overshoot control, tradeoffs among power, area, and performance are derived for low-impedance buffers and drivers useful for lossy lines in MCM's.

The time-domain finite difference (TD-FD) method is applied to obtain the frequency-dependent characteristics of shielded multiconductor microstrip line interconnect structures. A Gaussian pulse is used as the excitation source and the reflected, transmitted, and coupled pulses on the various lines are used for the calculation of the frequency-dependent elements in the scattering matrix representation of the structure. In order to correctly specify the frequency range of applicability of the results, the cutoff frequencies of the higher-order modes of the structures under consideration need to be found. This is done by a two-dimensional application of the TD-FD method over the cross-section of the multiconductor transmission line, based on the simple observation that modes at cutoff are characterized by the fact that ∂/∂z = 0 where z is the direction of propagation along the axis of the line. Numerical results are obtained for the dispersive characteristics of coupled microstrip lines and the frequency-dependent scattering parameters of a microstrip open-end line, a 90° bend, a right-angle via, and the discontinuity which occurs when one of the two lines is terminated in a coupled symmetric microstrip geometry.

A constant frequency diode-clamped series resonant power converter (CFCSRC) is proposed as a solution to problems associated with frequency-controlled resonant converters. This converter has two resonant frequencies, and control is achieved by varying the relative time per switching cycle spent at each resonant frequency. Two zero-current-switching (ZCS) modes are examined and plotted in the output plane. Operating and mode boundaries are found and also plotted in the output plane. The output equation for the main mode is shown to be hyperbolic. Peak voltages are shown to be less than or equal to the input voltage, and peak currents are shown to be less than those of the frequency-controlled diode-clamped series resonant converter over a large operating range. A design procedure is given in the form of a design example. Data from a prototype converter is plotted with theoretical data in the output plane and good agreement with the theoretical model is obtained.

A buck converter with a given output filter is operated with pulse-width modulated and quasi-resonant switching schemes at the same nominal load and switching frequency. Electromagnetic interference generated by the natural switching action of the converter is examined by Fourier spectral analysis. Interference caused by excitation of parasitic elements is examined experimentally. Quasi-resonant converters are found to have a lower switching frequency harmonic bandwidth than the equivalent pulse-width modulated converter. The most significant parasitic responses are the turn-on current and turn-off voltage of the catch diode, and the gate charge current of the MOSFET switch. A significant decrease in radiated and conducted noise is obtained when the gate drive voltage rise and fall times are increased, which is possible without loss of efficiency using quasi-resonant switching.

As VLSI chip sizes and packing densities continue to escalate, electromigration failures have become a primary reliability concern. The issues concerning electromigration testing are addressed and an effective test for characterizing electromigration failures is presented. It is shown that this test is a reliable and sensitive measure for determining electromigration resistance based upon the construction of an electromigration database using this test. Lastly, the implementation of an electromigration process control test based upon the characterization is discussed.

The gain-bandwidth (GB) product of an operational amplifier (op amp) can effect the filtering characteristics of an infinite gain, multiple feedback, low pass active RC filter. In this thesis, the characteristics of ideal op amps and actual op amps are compared. The op amp is characterized by a dominant single pole, which is used to develop a third order voltage transfer function that describes the op amp's gain-bandwidth effects on the filter. Optimization theory is used to find adjusted values for the filter's elements to compensate for these GB effects. Of the sixty-five compensation design methods examined with the optimization program GOSPEL, twelve methods produced useful compensation results. The adjusted element values are tabulated and displayed in design graphs for several values of GB and quality factor, Q. The twelve design techniques were verified by simulating the filter with the adjusted element values using PSpice.

Optical matched filters have been used for the recognition of patterns in a noisy background. Different types of matched filters have been proposed since the introduction of the VanderLugt matched spatial filter. A novel filter, the hybrid phase-only matched filter, is proposed which shows promise for better signal to noise ratio, correlation peak intensity and light efficiency compared to the recently proposed optimal phase-only filter. A neural technique for the design of space-domain binary filter for pattern recognition applications is developed. The method takes advantage of the similarity in the structure of the minimum squared error criterion for the construction of linear discriminant functions and the Lyapunov function of the Hopfield Neural Model.