Dedicated hardware implementations of neural networks promise to provide faster, lower power operation when compared to software implementations executing on processors. Unfortunately, most custom hardware implementations do not support intrinsic training of these networks on-chip. The training is typically done using offline software simulations and the obtained network is synthesized and targeted to the hardware offline. The FPGA design presented here facilitates on-chip intrinsic training of artificial neural networks. Block-based neural networks (BbNN), the type of artificial neural networks implemented here, are grid-based networks neuron blocks. These networks are trained using genetic algorithms to simultaneously optimize the network structure and the internal synaptic parameters. The design supports online structure and parameter updates, and is an intrinsically evolvable BbNN platform supporting functional-level hardware evolution. Functional-level evolvable hardware (EHW) uses evolutionary algorithms to evolve interconnections and internal parameters of functional modules in reconfigurable computing systems such as FPGAs. Functional modules can be any hardware modules such as multipliers, adders, and trigonometric functions. In the implementation presented, the functional module is a neuron block. The designed platform is suitable for applications in dynamic environments, and can be adapted and retrained online. The online training capability has been demonstrated using a case study. A performance characterization model for RC implementations of BbNNs has also been presented.

This dissertation is concerned with dynamical modeling, estimation and identification of wireless channels from received signal measurements. Optimal power control algorithms, mobile location and velocity estimation methods are developed based on the proposed models. The ultimate performance limits of any communication system are determined by the channel it operates in. In this dissertation, we propose new stochastic wireless channel models which capture both the space and time variations of wireless systems. The proposed channel models are based on stochastic differential equations (SDEs) driven by Brownian motions. These models are more realistic than the time invariant models encountered in the literature which do not capture and track the time varying characteristics of the propagation environment. The statistics of the proposed models are shown to be time varying, and converge in steady state to their static counterparts. Cellular and ad hoc wireless channel models are developed. In urban propagation environment, the parameters of the channel models can be determined from approximating the band-limited Doppler power spectral density (DPSD) by rational transfer functions. However, since the DPSD is not available on-line, a filterbased expectation maximization algorithm and Kalman filter to estimate the channel parameters and states, respectively, are proposed. The algorithm is recursive allowing the inphase and quadrature components and parameters to be estimated on-line from received signal measurements. The algorithms are tested using experimental data, and the results demonstrate the method’s viability for both cellular and ad hoc networks. Power control increases system capacity and quality of communications, and reduces battery power consumption. A stochastic power control algorithm is developed using the so-called predictable power control strategies. An iterative distributed algorithm is then deduced using stochastic approximations. The latter only requires each mobile to know its received signal to interference ratio at the receiver.

The research in this dissertation has focused upon image segmentation and its related areas, using the techniques of partial differential equations, variational methods, mathematical morphological methods and probabilistic methods. An integrated segmentation method using both curve evolution and anisotropic diffusion is presented that utilizes both gradient and region information in images. A bottom-up image segmentation method is proposed to minimize the Mumford-Shah functional. Preferential image segmentation methods are presented that are based on the tree of shapes in mathematical morphologies and the Kullback-Leibler distance in information theory. A thorough evaluation of the morphological preferential image segmentation method is provided, and a web interface is described. A probabilistic model is presented that is based on particle filters for image segmentation. These methods may be incorporated as components of an integrated image processed system. The system utilizes Internet Protocol (IP) cameras for data acquisition. It utilizes image databases to provide prior information and store image processing results. Image preprocessing, image segmentation and object recognition are integrated in one stage in the system, using various methods developed in several areas. Interactions between data acquisition, integrated image processing and image databases are handled smoothly. A framework of the integrated system is implemented using Perl, C++, MySQL and CGI. The integrated system works for various applications such as video tracking, medical image processing and facial image processing. Experimental results on this applications are provided in the dissertation. Efficient computations such as multi-scale computing and parallel computing using graphic processors are also presented.

There has been a dramatic increase in wireless awareness among the user community in the past five years. The 2.4-GHz Industrial, Scientific and Medical (ISM) band is being used for a diverse range of applications due to the following reasons. It is the only unlicensed band approved worldwide and it offers more bandwidth and supports higher data rates compared to the 915-MHz ISM band. The power consumption of devices utilizing the 2.4-GHz band is much lower compared to the 5.2-GHz ISM band. Protocols like Bluetooth and Zigbee that utilize the 2.4-GHz ISM band are becoming extremely popular. Bluetooth is an economic wireless solution for short range connectivity between PC, cell phones, PDAs, Laptops etc. The Zigbee protocol is a wireless technology that was developed as an open global standard to address the unique needs of low-cost, lowpower, wireless sensor networks. Wireless sensor networks are becoming ubiquitous, especially after the recent terrorist activities. Sensors are employed in strategic locations for real-time environmental monitoring, where they collect and transmit data frequently to a nearby terminal. The devices operating in this band are usually compact and battery powered. To enhance battery life and avoid the cumbersome task of battery replacement, the devices used should consume extremely low power. Also, to meet the growing demands cost and sized has to be kept low which mandates fully monolithic implementation using low cost process. CMOS process is extremely attractive for such applications because of its low cost and the possibility to integrate baseband and high frequency circuits on the same chip. A fully integrated solution is attractive for low power consumption as it avoids the need for power hungry drivers for driving off-chip components. The transceiver is often the most power hungry block in a wireless communication system. The frequency divider (prescaler) and the voltage controlled oscillator in the transmitter’s frequency synthesizer are among the major sources of power consumption. There have been a number of publications in the past few decades on low-power high-performance VCOs. Therefore this work focuses on prescalers. A class of analog frequency dividers called as Injection-Locked Frequency Dividers (ILFD) was introduced in the recent past as low power frequency division. ILFDs can consume an order of magnitude lower power when compared to conventional flip-flop based dividers. However the range of operation frequency also knows as the locking range is limited. ILFDs can be classified as LC based and Ring based. Though LC based are insensitive to process and temperature variation, they cannot be used for the 2.4-GHz ISM band because of the large size of on-chip inductors at these frequencies. This causes a lot of valuable chip area to be wasted. Ring based ILFDs are compact and provide a low power solution but are extremely sensitive to process and temperature variations. Process and temperature variation can cause ring based ILFD to loose lock in the desired operating band. The goal of this work is to make the ring based ILFDs useful for practical applications. Techniques to extend the locking range of the ILFDs are discussed. A novel and simple compensation technique is devised to compensate the ILFD and keep the locking range tight with process and temperature variations. The proposed ILFD is used in a 2.4-GHz frequency synthesizer that is optimized for fractional-N synthesis. Measurement results supporting the theory are provided.

Cocaine is a well-known, illegal, recreational drug that is addictive due to its effects on the mesolimbic reward pathway in the human body. Accurate and real-time measurement of the concentration of cocaine in the body as a function of time and physiological factors is a key requirement for the understanding of the use of this drug. Current methods for such measurements involve taking samples from the human body (such as blood, urine, and hair) and performing analytical chemistry tests on these samples. This techniques are relatively expensive, time consuming, and labor intensive. To address this issue, a new implantable sensor for the automated detection and measurement of the relative cocaine concentration is presented here. The device is more economical and provides for higher sampling frequencies than the current methods. The active sensor elements consist of piezoresistive microcantilever arrays, which are coated with an oligonucleotide-based aptamer, i.e. a short sequence of RNA with high affinity for specific target molecules, as the cocaine receptor. A Wheatstone bridge is used to convert the biosensor signal into an electronic signal. This signal is transmitted wireless at an operating frequency of 403.55 MHz, which complies with the US Medical Implant Communication System (MICS) FCC 47CFR Part 95. The limit of detection for the in vitro experiment is found to be 1 ng/ml. The device has successfully measured the relative concentration of cocaine upon implantation in the subcutaneous interstitial fluid of male Wistar rats.

Field Programmable Gate Arrays (FPGAs) enable powerful performance acceleration for scientific computations because of their intrinsic parallelism, pipeline ability, and flexible architecture. This dissertation explores the computational power of FPGAs for an important scientific application: linear algebra. First of all, optimized linear algebra subroutines are presented based on enhancements to both algorithms and hardware architectures. Compared to microprocessors, these routines achieve significant speedup. Second, computing with mixed-precision data on FPGAs is proposed for higher performance. Experimental analysis shows that mixed-precision algorithms on FPGAs can achieve the high performance of using lower-precision data while keeping higher-precision accuracy for finding solutions of linear equations. Third, an execution time model is built for reconfigurable computers (RC), which plays an important role in performance analysis and optimal resource utilization of FPGAs. The accuracy and efficiency of parallel computing performance models often depend on mean maximum computations. Despite significant prior work, there have been no sufficient mathematical tools for this important calculation. This work presents an Effective Mean Maximum Approximation method, which is more general, accurate, and efficient than previous methods. Together, these research results help address how to make linear algebra applications perform better on high performance reconfigurable computing architectures.

This dissertation addresses long range automated persistent surveillance with focus on three topics: sensor planning, size preserving tracking, and high magnification imaging. field of view should be reserved so that camera handoff can be executed successfully before the object of interest becomes unidentifiable or untraceable. We design a sensor planning algorithm that not only maximizes coverage but also ensures uniform and sufficient overlapped camera’s field of view for an optimal handoff success rate. This algorithm works for environments with multiple dynamic targets using different types of cameras. Significantly improved handoff success rates are illustrated via experiments using floor plans of various scales. Size preserving tracking automatically adjusts the camera’s zoom for a consistent view of the object of interest. Target scale estimation is carried out based on the paraperspective projection model which compensates for the center offset and considers system latency and tracking errors. A computationally efficient foreground segmentation strategy, 3D affine shapes, is proposed. The 3D affine shapes feature direct and real-time implementation and improved flexibility in accommodating the target’s 3D motion, including off-plane rotations. The effectiveness of the scale estimation and foreground segmentation algorithms is validated via both offline and real-time tracking of pedestrians at various resolution levels. Face image quality assessment and enhancement compensate for the performance degradations in face recognition rates caused by high system magnifications and long observation distances. A class of adaptive sharpness measures is proposed to evaluate and predict this degradation. A wavelet based enhancement algorithm with automated frame selection is developed and proves efficient by a considerably elevated face recognition rate for severely blurred long range face images.

In the design of any multi-port network with more than one antenna, mutual coupling between these different ports must be accounted for. In an effort to investigate and control these mutual coupling effects, we have selected three structures to be thoroughly analyzed. Furthermore, they have been fabricated and tested to develop relevant design guides for these selected structures to have minimal mutual coupling effects. These selected structures included a feed network for a multi-port antenna, a dual feedhorn for a large reflector antenna, as well as a set of Multi- Input Multi-Output (MIMO) laptop antennas. In the first study, we analyzed a 30- port radial splitter that can be used for an in-phase feeding of a 30-high power transmitter. Our objectives here have been geared towards estimating the mutual coupling between the 30 ports and exploring the port and alignment failure analysis, its graceful degradation results, and relevant efficiency performance for such high power multi-port network will be presented. In the second study, we investigated the mutual coupling of a multifeedhorn structure of a large reflector antenna in order to allow multi-beam radiation or reception. This high gain antenna utilizes integrated feeds with precise physical tight spacing and could suffer from strong inter-coupling. Mutual coupling effects here include input match deterioration, beam width broadening, and cross-polarization degradation due to the proximity coupling of these various feeds. Our study derived accurate feed location expressions as well as methods to improve the decoupling between the feeds that have been implemented. These results will be discussed. For the third study, we carried out extensive investigates into the mutual coupling effects amidst wireless laptop antennas for a MIMO system implementation. For a laptop use, it is required to determine the best location, optimum spacing, and orientations of these antennas in order to achieve the maximum benefits of the system’s diversity. First, we studied the coupling between two antennas as a function of their spacing, types, and orientations. Subsequently, we extended the study to a controlled multi-antenna system for a MIMO implementation. Design rules for such implementation have been derived and will be discussed in detail.

The acquisition process of digitizing a large-scale environment produces an enormous amount of raw geometry data. This data is corrupted by system noise, which leads to 3D surfaces that are not smooth and details that are distorted. Any scanning system has noise associate with the scanning hardware, both digital quantization errors and measurement inaccuracies, but a mobile scanning system has additional system noise introduced by the pose estimation of the hardware during data acquisition. The combined system noise generates data that is not handled well by existing noise reduction and smoothing techniques. This research is focused on enhancing the 3D models acquired by mobile scanning systems used to digitize large-scale environments. These digitization systems combine a variety of sensors – including laser range scanners, video cameras, and pose estimation hardware – on a mobile platform for the quick acquisition of 3D models of real world environments. The data acquired by such systems are extremely noisy, often with significant details being on the same order of magnitude as the system noise. By utilizing a unique 3D signal analysis tool, a denoising algorithm was developed that identifies regions of detail and enhances their geometry, while removing the effects of noise on the overall model. The developed algorithm can be useful for a variety of digitized 3D models, not just those involving mobile scanning systems. The challenges faced in this study were the automatic processing needs of the enhancement algorithm, and the need to fill a hole in the area of 3D model analysis in order to reduce the effect of system noise on the 3D models. In this context, our main contributions are the automation and integration of a data enhancement method not well known to the computer vision community, and the development of a novel 3D signal decomposition and analysis tool. The new technologies featured in this document are intuitive extensions of existing methods to new dimensionality and applications. The totality of the research has been applied towards detail enhancing denoising of scanned data from a mobile range scanning system, and results from both synthetic and real models are presented.

Flexibility of any proposed communication links is becoming one of the most challenging features. Direct broadcasting satellite services, for example, will be greatly enhanced by providing service-on-the-move. This market is very demanding as it necessitates the development of a low cost, low profile antenna that can be mounted on top of SUVs and minivans, which is capable of continuously tracking the satellite. Another example is the wireless antennas for laptops and smart-phones, where the antennas should fit within an extremely small volume and should be capable of addressing many services over wide frequency range. In this dissertation, both DBS and the wireless antennas are addressed. In these studies, efforts have been concentrated in developing low profile planar antennas, in particular, slot arrays. Travelling wave slotted waveguide arrays have been utilized to minimize the scanning angle range limits due to their inherent beam tilt angle. CNC machines were utilized first to fabricate the early prototypes for sub-array developments. Subsequently, a low cost fabrication technology is adopted to develop a low cost and light weight full array using substrate integrated waveguides (SIWs). The SIW is fully characterized and an excellent equivalent model has been derived to allow easy "translation" of metallic waveguide components to SIW. Various SIW junctions, transitions, and arrays have been developed for array feed networks including a 64 radiating SIW full array and a 32 radiating SIW array with folded feed. Meanwhile, for the wireless antennas, the utilization of reconfigurable hardware has been introduced to provide the required multi-functionality services and wide frequency coverage. Various reconfigurable antennas were developed and utilized to demonstrate their advantages compared to other design options such as wide-band or multi-band approaches. Both micro-electro-mechanical switches "MEMS" and "PIN" diodes have been successfully utilized to switch between the different configurations. The placement, control, and modeling of the switches are also discussed and novel modeling and biasing topologies are introduced. A novel and practical concept of reconfigurable multiband antenna is introduced here too, where advantages of both the multi-band and the reconfigurable antenna structures can be simultaneously achieved while supporting more services.