Electrical and Electronics Commons^™

Communications Using Deep Learning Techniques, Priti Gopal Pachpande

Legacy Theses & Dissertations (2009 - 2024)

Deep learning (DL) techniques have the potential of making communication systems

Sparse Adaptive Local Machine Learning Algorithms For Sensing And Analytics, Jack Cannon

Undergraduate Research & Mentoring Program

The goal of digital image processing is to capture, transmit, and display images as efficiently as possible. Such tasks are computationally intensive because an image is digitally represented by large amounts of data. It is possible to render an image by reconstructing it with a subset of the most relevant data. One such procedure used to accomplish this task is commonly referred to as sparse coding. For our purpose, we use images of handwritten digits that are presented to an artificial neural network. The network implements Rozell's locally competitive algorithm (LCA) to generate a sparse code. This sparse code is …

Configuring The Radial Basis Function Neural Network, Insoo Sohn

Theses

The most important factor in configuring an optimum radial basis function (RBF) network is the training of neural units in the hidden layer. Many algorithms have been proposed, e.g., competitive learning (CL), to train the hidden units. CL suffers from producing "dead-units." The other major factor Which was ignored in the past is the appropriate selection of the number of neural units in the hidden layer. The frequency sensitive competitive learning (FSCL) algorithm was proposed to alleviate the problem of dead-units, but it does not alleviate the latter problem. The rival penalized competitive learning (RPCL) algorithm is an improved version …

Embedology And Neural Estimation For Time Series Prediction, Robert E. Garza

Theses and Dissertations

Time series prediction has widespread application, ranging from predicting the stock market to trying to predict future locations of scud missiles. Recent work by Sauer and Casdagli has developed into the embedology theorem, which sets forth the procedures for state space manipulation and reconstruction for time series prediction. This includes embedding the time series into a higher dimensional space in order to form an attractor, a structure defined by the embedded vectors. Embedology is combined with neural technologies in an effort to create a more accurate prediction algorithm. These algorithms consist of embedology, neural networks, Euclidean space nearest neighbors, and …

Head Related Transfer Function Approximation Using Neural Networks, John K. Millhouse

Theses and Dissertations

This thesis determines whether an artificial neural network (ANN) can approximate the Armstrong Aerospace Medical Research Laboratories (AAMRL) head related transfer functions (HRTF) data obtained from research at AAMRL during the fall of 1988. The first test determines whether HRTF lends any support in sound localization when compared to no HRTF (Interaural Time Delay only). There is a statistically significant interaction between the location of the sound and whether the HRTF or no HRTF is used. When this interaction is removed using the alternate F.Value, the statistics give the result of equal means for the filters and azimuth. This means …

Automatic Tuning Of Integrated Filters Using Neural Networks, Lutz Henning Lenz

Dissertations and Theses

Component values of integrated filters vary considerably due to· manufacturing tolerances and environmental changes. Thus it is of major importance that the components of an integrated filter be electronically tunable. The method explored in this thesis is the transconductance-C-method.

A method of realizing higher-order filters is to use a cascade structure of second-order filters. In this context, a method of tuning second-order filters becomes important.

The research objective of this thesis is to determine if the Neural Network methodology can be used to facilitate the filter tuning process for a second-order filter (realized via the transconductance-C-method). Since this thesis is, …

Design Of An Artificial Neural Network Based Tactile Sensor For The Utah/Mit Dexterous Hand, Jeffery D. Nering

Theses and Dissertations

The Neural Tactile Sensor (NTS) is a high resolution, easily manufactured tactile sensor consisting of electrodes, a thin resistive 'skin', and pattern recognition circuitry that is capable of resolving dynamic and static contact location, force, and slip throughout the continuum of the sensor's active region. The sensor operates by means of a resistive 'skin' harboring the electric field generated when a current is injected into it, and a plurality of electrodes for taking measurements of said electric field. When current flows through the resistive medium from the location of tactile contact, an electric field within the resistive medium is established, …

A New Method To Optimize The Satellite Broadcasting Schedules Using The Mean Field Annealing Of A Neural Network, Youyi Yu

Theses

This thesis reports a new method for optimizing satellite broadcasting schedules based on the Hopfield neural model in combination with the mean field annealing theory. A clamping technique is used with an associative matrix, thus reducing the dimensions of the solution space. A formula for estimating the critical temperature for the mean field annealing procedure is derived, hence enabling the updating of the mean field theory equations to be more economical. Several factors on the numerical implementation of the mean field equations using a straightforward iteration method that may cause divergence are discussed; methods to avoid this kind of divergence …

Binaural Sound Localization Using Neural Networks, Rushby C. Craig

Theses and Dissertations

The purpose of this study was to investigate the use of Artificial Neural Networks to localize sound sources from simulated, human binaural signals. Only sound sources originating from a circle on the horizontal plane were considered. Experiments were performed to examine the ability of the networks to localize using three-different feature sets. The feature sets used were: time-samples of the signals, mena Fast Fourier Transform magnitude and cross correlation data, and auto-correlation and cross correlation data. The two different types of sound source signals considered were tones and gaussian noise. The feature set which yielded the best results in terms …

Function Prediction Using Recurrent Neural Networks, Randall L. Lindsey

Theses and Dissertations

A fully recurrent neural network was applied to the function prediction problem. The real-time recurrent learning (RTRL) algorithm was modified and tested for use as a viable function predictor. The modification gave the algorithm a variable learning rate and a linear/sigmoidal output selection. Verifying the networks ability to temporally learn both the classic exclusive-OR (XOR) problem and the internal state problem, the network was then used to simulate the frequency response of a second order IIR lowpass Butterworth filter. The recurrent network was then applied to two problems: head position tracking, and voice date reconstruction. The accuracy at which the …

An Investigation Of The Application Of Artificial Neural Networks To Adaptive Optics Imaging Systems, Andrew H. Suzuki

Theses and Dissertations

Recurrent and feedforward artificial neural networks are developed as wavefront reconstructors. The recurrent neural network studied is the Hopfield neural network and the feedforward neural network studied is the single layer perceptron artificial neural network. The recurrent artificial neural network input features are the wavefront sensor slope outputs and neighboring actuator feedback commands. The feedforward artificial neural network input features are just the wavefront sensor slope outputs. Both artificial neural networks use their inputs to calculate deformable mirror actuator commands. The effects of training are examined.

Broadband Isdn Control Using Neural Networks, Stavros Zervoudakis

Theses

Broadband Integrated Services Digital Networks (B-ISDN) are considered to become the standard communications networks in the near future. Major research efford is being applied in determining the appropriate mode of operation for B-ISDN systems. Two proposals, the Synchronous Transfer Mode (STM) and the Asynchronous Transfer Mode (ATM) claim to best fit the B-ISDN transmission requirements individually. A new method for B-ISDN traffic control is proposed in this paper. This method assigns priority levels to incoming traffic while integrating STM with ATM. A neural network is used to decide how the multiplexing of the incoming packets will be performed. Several traffic …

The Bandwidth Allocation Of Atm Control By Neural Networks, Jin-Syan Chou

Theses

In this thesis, we develop a neural network method based on Adaptive Resonance Theory to train and control the optimum bandwidth allocation of ATM network. In Broadband Integrated Service Digital Network (BISDN), the Asynchronous Transfer Mode (ATM) is already adopted as the transfer facility by CCITT. ATM is a high-bandwidth, low- delay, fast-packet switching and multiplexing technique. Using ATM technique, we can flexibly rearrange the network and reassign the bandwidth to meet the requirement of all types of services. As an effective optimization method, Genetic Algorithm (GA) is applied to implement the bandwidth allocation of ATM. Then, we use Adaptive …

The Neural Network Based Control System For Dynamic Channel Allocation In Pcn Communications, Ching-Yao Huang

Theses

A neural network based control system (NNCS) is adopted in this paper to conduct the channel allocation task in the personal communication networks (PCN). Binary adaptive resonance theory (ART-1) that was modified for pattern matching and provided the corresponding responses is selected as the basic algorithm for the system. Digital portable radio technology can provide reliable access to high quality wireless service for many users. Low power transmission requirement and small cell size configuration that can support more users and diverse services than cellular radio system are the attractive attributes of PCN. Assigning multiple time slots to a user in …

Comparative Study Of Prediction Gain Based On Neural Network Architecture, Prashant M. Shah

Theses

This thesis describes the Neural Network approach to design predictor using Delta and Generalized Delta Rule. The predictor is designed by supervised training based on the typical sequence of pixel values. Neural Network is used to find the coefficients of the predictor. Both 1-D and 2-D scheme of the pixels as well as linear and non-linear correlations are used to find the coefficients by training. Different combinations of pixels are used to find the "best" combination among the order of the predictor.

Fast Packet Switching For B-Isdn With Neural Net Control, Ajit K. Chaudhuri

Theses

The concept of integrated network where both the voice packet and data packet are delt with, started taking shape since 1970. Later, Users? demand for communication of data, voice, file, facsimile, image, videotext, videophone, videomovie has also been felt. All these services need bit rate of transmission in the order of 10, even 100 Mbits. An economical way to transmit this higher bit rate, with flexibility of bandwidth, is to treat voice/video in the form of packets and transmit those packets in real time. Studies of this kind of communication is carried out in Broadband ISDN ( BISDN). At this …

Network Reduction Using Error Prediction, Michael V. Gilsdorf

Theses and Dissertations

This thesis investigates gradient descent learning algorithms for multi-layer feed forward neural networks. A technique is developed which uses error prediction to reduce the number of weights/nodes in a network. The research begins by studying the first and second order back-prop training algorithms along with their convergence properties. A network is reduced by making an estimate of the amount of error which would occur when a weight(s) is removed. This error estimate is then used to determine if a particular weight is essential to the operation of the network. If not, it is removed and the network retrained. The process …

A Method Of Moments Solution For The Electric Currents On An Aperture-Fed, Stacked Patch Microstrip Antenna, William L. Irvin

Theses and Dissertations

This study presents a method of moments solution for the surface currents and charge distributions on an aperture-fed, stacked patch antenna. To make the solution independent of the antenna excitation technique, an aperture magnetic current distribution is assumed, and the spatial Green's functions of the antenna are used to calculate the tangential fields on the antenna patches and coefficients for a finite series expression of the surface currents and charge densities. The Green's functions of the antenna are modeled with polynomials that are a function of the radial distance separating observer and source locations. The method of moments solution is …

Gabor Filters And Neural Networks For Segmentation Of Synthetic Aperture Radar Imagery, Albert P. L'Homme

Theses and Dissertations

This research investigates Gabor filters and artificial networks for autonomous segmentation of 1 foot by 1 foot) high resolution polarimetric synthetic aperture radar (SAR). Processing involved frequency correlation between the SAR imagery and biologically motivated Gabor functions. Methods for selecting the Gabor tuning parameters from the endless choices of frequency, rotation, standard deviation and bandwidth are discussed. Using these parameters, resulting Gabor correlation images were reduced in speckle, and more detailed. This research used cosine Gabor functions and operated on single polarization HH magnitude data. Following selection of the appropriate Gabor features, multiple Gabor representations were generated and converted for …

Characterization Of Radar Signals Using Neural Networks, Daniel R. Zahirniak

Theses and Dissertations

Recent work concerning artificial neural networks has focused on decreasing network training times. Kernel Classifier networks, using radial basis functions (RBFs) as the kernel function, can be trained quickly with little performance degradation. Short training times are critical for systems which must adapt to changing environments. The function of Kernel Classifier networks is based on the principle that multivariate functions can be approximated via linear combinations of RBFs. RBFs can also perform probability density estimations, making classifications approximating a Baye's optimal descriminant. Methods used to set the RBF centers included matching the training data, Kohonen Training, K-Means Clustering and placement …

Classification Of Correlation Signatures Of Spread Spectrum Signals Using Neural Networks, Richard A. Chapman

Theses and Dissertations

The major goals of this thesis were to determine if Artificial Neural Networks (ANNs) could be trained to classify the correlation signatures of two classes of spread spectrum signals and four classes of spread spectrum signals. Also, the possibility of training an ANN to classify features of the signatures other than signal class was investigated. Radial Basis Function Networks and Back-Propagation Networks were used for the classification problems. Correlation signatures of four types or classes were obtained from United States Army Harry Diamond Laboratories. The four types are as follows: direct sequence (DS), linearly-stepped frequency hopped (LSFH), randomly-driven frequency hopped …

Landmark-Based Partial Shape Recognition By A Two-Stage Bam Neural Network, Xianjun Liu

Theses

In this thesis, we develop a Bitfirectional Associative Memory (BAM) based neural network to achieve partial shape recognition. To recognize objects which are partially occluded, we represent each object by a set of landmarks. The landmarks of an object are points of interest relative to the object that have important shape attributes. To achieve recognition, feature values (landmark values) of each model object are trained and stored in the network. Each memory cell is trained to store landmark values of a model object for all possible positions. Given a scene which may consist of several objects, landmarks in the scene …

Neural Net Classification Of States In Finite State Vector Quantization, Zhigang Zhang

Theses

In this thesis, a new image coding technique based on Neural Network Finite State Vector Quantization(NNFSVQ) is presented. The new design takes advantage of the high interblock spatial correlation of pixels in typical grey-level images, in addition to intrablock correlation already exploited by vector quantization (VQ). Although many FSVQ require large memory space for the storage of the numerous state codebooks, the memory space requirement of NNFSVQ can be reduced by a very large factor (about 10²-10³) to manageable size without impairing image quality. Simulation experiments have shown that by using NNFSVQ the bit rate can …

Landmark-Based Shape Recognition By A Modified Hopfield Neural Network, Kuowei Li

Theses

In this thesis, we introduce a new thethod to achieve partial shape recognition by means of a modified Hopfield neural network. To recognize partially visible object, we represent each object by a set of "landmarks." The landmarks of an object are points of interest relative to the object that have important shape attributes. Given a scene consisting of partially occluded objects, a model object in the scene is hypothesized by matching the landmarks of the model with those in the scene. A measure of similarity between two landmarks, one from a model and the other from the scene, is needed …

An Artificial Neural Network For Redundant Robot Inverse Kinematics Computation, Wibawa Utama

Theses

A redundant manipulator can be defined as a manipulator that has more degrees of freedom than necessary to determine the position and orientation of the end effector. Such a manipulator has dexterity, flexibility, and the ability to maneuver in presence of obstacles. One important and necessary step in utilizing a redundant robot is to relate the joint coordinates of the manipulator with the position and orientation of the end-effector. This specification is termed as the direct kinematics problem and can be written as x = f(q)
where x is a vector representing the position and orientation of the end-effector, q …

Image Data Compression With Neural Networks, Jianjun Li

Theses

Three efficient image coding schemes, based orb neural networks, have been developed in this thesis: (1) Neural network vector quantization (NNVQ). The main advantage of this new technology is that it can accomplish the complex encoding process much faster than the previous algorithms. Its properties are studied and demonstrated by simulations. (2) A adaptive NNVQ for image sequence coding. Simulation experiments have been carried out with 4 x 4 blocks of pixels from an image sequence consisting of 40 frames. At 0.67 bits/pixel, this scheme achieves good image quality suitable for videoconferencing systems; (3) Neural network prediction. This new prediction …