Real-time patient monitoring is an essential task in the critical care unit. Care providers need to process a large amount of data obtained from patient monitoring devices and the hospital information system. Information overload can lead to sub-optimal decisions and therapeutic actions. SIMON is a system being developed to address these issues by acquiring and processing data from the bedside monitoring devices and the hospital information system. The initial SIMON prototype was deployed in the Coronary Care Unit of the Vanderbilt University Medical Center. Experience acquired with this system revealed the need for a change in architecture and a complete reimplementation. The revised SIMON has been designed with distribution in mind to achieve reliability, expandability, scalability, and flexibility. It is divided into three layers. The Data Layer provides the functionality to collect the information. The Task Layer implements signal evaluation functions to detect required event. The Knowledge Layer provides high-level reasoning capabilities. Each layer is subdivided into separate but communicating components. This thesis begins with an introduction to patient monitoring systems, the previous SIMON architecture and the revised SIMON architecture are then described. This is followed by a description of the Data Layer, the Task Layer, and the revised Task Layer. We conclude with a discussion on results we have obtained, the current status of the system, and future research recommendations.

In this thesis, firstly a real-time gait intent recognition approach for use in controlling a fully powered transfemoral prosthesis is described. Rather than utilize an echo control as proposed by others, which requires instrumentation of the sound-side leg, the proposed approach infers user intent based on the characteristic shape of the force and moment vector of interaction between the user and prosthesis. The real-time intent recognition approach utilizes a K-nearest neighbor algorithm with majority voting and threshold biasing schemes to increase its robustness. The ability of the approach to recognize in real time a persons intent to stand or walk at one of three different speeds is demonstrated on measured biomechanics data. Secondly, an active passive torque decomposition procedure for use in controlling a fully powered transfemoral prosthesis is described. The active and passive parts of the joint torques are extracted by solving a constrained least squares optimization problem. The proposed approach generates the torque reference of joints by combining the active part, which is a function of the force and moment vector of the interaction between user and prosthesis and the passive part, which has a nonlinear spring-dashpot behavior. The ability of the approach to reconstruct the required joint torques is again demonstrated in simulation on measured biomechanics data. Finally, the calibration procedure of the three axis socket load cell of the prosthesis' mechanical sensory interface is also presented in the thesis.

In robotics, one important objective is the ability to teach the robot new skills and have it reason about the current tasks at hand without explicit programming. Indeed, this idea is central to open-ended development, developmental robotics and autonomous mental development. One approach to this issue is to have the robot learn from its own past experience, which would help the robot adapt to changing environments. However, in a learning process, a critical issue to both robots and biological creatures is efficient use of the limited resources available for survival. A robot, operating in a complex, unstructured environment, will encounter many percepts, and typically most of them are not relevant to the current task. This suggests the need for a capability to focus attention on the smaller number of items that are relevant to the task. Thus, prefrontal cortex working memory models may be a good fit for learning to associate perception and action, and perhaps other concepts as well, in order to perform a task. Many of the systems in the literature have only crude perceptual capabilities and as a result, the environments are usually very simplified by modifications, such as by using artificial percepts. Such systems may fail in complex, uncontrolled environments, especially under changing lighting conditions. Thus, successful task execution strongly depends on a reliable perceptual system in these types of environments. In this work, a novel implementation of a perceptual system, which operates on an extremely high dimensional feature space, is combined with a biologically inspired working memory model. The perceptual system does not rely on any parametric techniques (i.e., computing eigenvectors, covariance matrices, etc.) and the computational cost does not depend strongly on the number of dimensions. Only vision is used, as the main sensory input for the system. The resulting system initially learns basic behaviors and skills, which in turn, are used to learn more complex behaviors. The success of the system is demonstrated with a vision guided navigation task in a complex, noisy, and unmodified environment.

The human voice is a source of important information regarding the physical, psychological, and mental health conditions of a speaker. Acoustic properties of speech have previously been reported as possible cues to risk of committing suicide in persons suffering from severe depression. Certain vocal parameters may be capable of objectively distinguishing depressive speech from near-term suicidal speech. Studies were performed to analyze and statistically compare the speech acoustics of separate female and male samples comprised of subjects attempting suicide and subjects carrying diagnoses of depression and remission (recovery from depression). In this study, two types of speech recordings, spontaneous and reading speech, were collected from each subject of diagnostic groups participating in interview and text-reading sessions. Acoustic analyses of energy distribution within a 0-2,000 Hz frequency range and energy concentration characterizing the vocal tract spectral response based on the Gaussian mixture model (GMM) were performed on speech samples. Discriminant analyses demonstrated the significance of energy distribution and GMM-based vocal features as being effective indicators of perceptual changes in speech production and articulation caused by the severity of psychological state, and as powerful discriminators of diagnostic groups in both female and male studies. Based on the most important pairwise study of depressed and suicidal speech, the 12-fold cross validations yielded the correct classification scores of 86% and 90.33% in classifying spontaneous and reading speech of females, and 86% and 88.50% in classifying male spontaneous and reading speech, respectively. Results suggest the investigated features derived from the reading speech capable of identifying the degree of psychological state as effective as those derived from the spontaneous speech among diagnostic groups.

Single Event transients caused by incident charged particles on a semiconductor device, affect response of circuits in space applications. The shape of the transient pulse, especially the Full Width Half Max (FWHM) pulse width has important effects on the Single Event response of the circuit. Aggressive scaling in deep sub micron CMOS regimes has caused major changes in sequence and location of drift-diffusion events most important to the overall charge collection process, which results in a Single Event pulse shape that differs considerably from the shapes predicted by existing charge collection models. In this thesis, we use extensive TCAD (Technology Computer Aided Design) analysis to identify an electrostatic model explaining the observed pulse shapes. While the first half of this work deals with analysis and modeling of Single Event transient pulses based on circuit and contact boundary conditions, the second half looks at a pulse width mitigation technique based on the impact of the substrate doping profile on pulse shape. Some of the widest pulses in highly scaled CMOS are due to parasitic bipolar conduction. We demonstrate through TCAD modeling, the application of a novel buried layer scheme to reduce charge collection due to parasitic bipolar conduction - leading to truncation of some of the widest pulses.

Internal rehearsal is a relatively new research area allowing a robot to contemplate the consequences of its actions before they are attempted by the actuation system. This thesis focuses on integrating internal rehearsal into the area of collision detection. This thesis was inspired by Murray Shanahans work involving brain-based robotic internal simulation. This research explores how internal rehearsal, by using a collision detection method via collision spheres, may enhance task execution. When the cognitive robot ISAC performs a task involving percepts and arm behaviors, the Internal Rehearsal System (IRS), with a help from the Central Executive Agent, performs this task internally and determines if a collision occurs. The CEA Internal rehearsal covered in this thesis involves a pair of one percept (stored in the short-term memory) and one arm behavior (stored in the long-term memory and recalled into the working memory) with two objects perceived on the robots Sensory Egosphere. If an undesirable collision occurs within the IRS, the robot will try a different behavior to avoid this predicted collision.

Echo Planar Imaging (EPI) is a magnetic resonance imaging (MRI) technique permitting very rapid data acquisition. This makes EPI a widely used fast imaging technique in many applications, such as functional magnetic resonance imaging (fMRI), diffusion weighted imaging (DWI), and dynamic imaging with contrast agent enhancement. However, a well-known problem with EPI is that it is more sensitive to image distortions than conventional MRI due to the encoding scheme that is used. <p> In this wok we developed methods based on nonrigid registration to correct in EPI images the geometric and intensity distortions caused by the inhomogeneity in the main magnetic field. Our methods are founded on the physics of EPI. First, we incorporate along with a standard Jacobian correction factor a new rephasing factor into our nonrigid registration algorithm to account for signal loss due to dephasing in Gradient Echo EPI images such that not only geometric distortion but also intensity distortion and attenuation in Gradient Echo (GE) EPI images can be corrected after the registration process. Second, we introduce a spatially varying scale mechanism into our registration algorithm to adapt the local scale properties of the deformation field to match these with the characteristics of the actual geometric distortions in EPI images. Third, we proposed in addition a hybrid method that combines both a field map and nonrigid registration for correction of distortions in EPI images. Finally, we also compared three distortion correction methods to study their advantages and disadvantages, which provides insights into the selection of field-map or registration methods for distortion correction of EPI images under certain circumstances. Experiments are performed on both simulated and real images.

In order to provide preventative security to a homogeneous device network, techniques in addition to static encryption must be implemented to assure network integrity by identifying possible deviant nodes within the collective. This thesis proposes a set of algorithms and techniques for an intrusion detection system, which when combined, provide a two-stage approach that seeks to reduce or eliminate training period requirements, while providing multiple anomaly detection and a degree of self tuning. By utilizing a high level of behavioral abstraction, these intrusion detection techniques can be applied to a broad range of devices, network implementations, and scenarios. Each device node is supplied with an embedded intrusion detection system which allows it to monitor inter-device requests, enabling machine learning techniques for purposes of deviant node analysis. The two principal methods, a maxima detection scheme, and a cross-correlative detection scheme, are combined to create a two-phase detection scheme that can successfully determine deviant node pervasion percentages of up to 22% within the homogeneous device network.

Single-event transients (SETs) due to terrestrial or space radiation exposure have become a growing concern in modern high-speed analog and mixed-signal electronics. Recent work with computer circuit-level simulation techniques has enabled the understanding of SET effects in mixed-signal and radio-frequency (RF) applications such as the phase-locked loop (PLL). Furthermore, a PLL radiation sensitivity weak point has been identified as the conventional current-based charge pump (CP), with ion strikes in the CP resulting in at least two orders of magnitude higher output phase displacement than any other module within the PLL. <br> This thesis presents a CP topology as a novel method to solving this critical SET problem with the potential of significantly improving overall PLL resistance to SET effects. A method of PLL design employing a tri-state, voltage-based charge pump (V-CP) circuit has been implemented that significantly hardens the PLL to SET effects. Simulations and experimental testing have been performed on PLL circuits designed and fabricated in the IBM 130nm CMRF8RF CMOS technology available through the MOSIS foundry system. Analysis of the measured PLL output error signatures is used to quantify the relative hardness of PLL circuits implementing a V-CP stage over a conventional CP module, demonstrating a maximum of 2.3 orders of magnitude improvement in the SET response. The design effectively eliminates the charge pump as the most susceptible element in the PLL; as a result, this hardened design technique, which can be applied to other PLL topologies, provides SE performance that is orders of magnitude better than typical PLL designs.

Separating a complex software system into individual components with well-defined interfaces is a common practice in software engineering intended to simplify reasoning about the system. Establishing a precise set of rules that define components and how they interact over the interfaces is necessary in order to formulate expectations about the possible behaviors that can arise from their composition. These rules are commonly called a Model of Computation (MoC); they establish the legal syntactical, structural, behavioral, and temporal patterns over which components execute. Various approaches have been developed to define MoC-s and their influence over system behavior. The work presented here describes a new approach for specifying MoC-s operationally within a formal modeling framework that captures both event-triggered and time-triggered behaviors. Equipped with this new reasoning framework, an illustrative example is provided to show how a common MoC for real-time systems can be implemented on physical hardware and off-the-shelf software. By first modeling the MoC, the execution logic that orchestrates components is established without introducing limitations of the physical system. Following the realization of the implementation, the model can be updated to reflect these limitations in order to give developers a more realistic view of how a MoC influences system behavior in a realistic deployment setting.