Erdol, Nurgun

This thesis presents a comprehensive analysis of a relatively new transform for discrete time signals, called the Discrete Wavelet Transform (DWT). We find how this transform is connected with the already existing theory of perfect reconstruction filter banks and the recently introduced theory of multiresolution analysis. We use the conditions obtained from these two theories in order to understand the construction of wavelet filters, which also generate continuous functions that prove to constitute an orthonormal basis for the L$\sp2$ space. We also investigate the connection of this transform to the sampled wavelet series of nonorthogonal functions with good time-frequency localization properties. Finally, we see the way that the DWT maps a discrete signal in the phase plane and the applications that such representations incorporate.

A waveform substitution technique using interpolation based on such slow varying parameters of speech as short-time energy and average zero-crossing rate is developed for a packetized speech communication system. The system uses 64 Kbps conventional PCM for encoding and takes advantage of active talkpurts and silence intervals to increase the utilization efficiency of a digital link. The short-time energy and average zero-crossing rates calculated for the purpose of determining talkpurts are transmitted in a preceeding packet. Hence, when a packet is pronounced "lost", its envelope and frequency characteristics are obtained from the previous packet and used to synthetize a substitution waveform which is free of annoying sounds that are due to abrupt changes in amplitude. Informal listening tests show that tolerable packet loss rate up to 40% are achievable with these procedures.

This thesis describes a software model of a Linear Predictive Coding (L.P.C.) that is written in the Ada language. The novel feature of this program is that it attempts to execute the maximum possible number of concurrent arithmetic operations in the L.P.C. algorithm. Each arithmetic operation is implemented by an active process which is the "task" construct in the Ada language. The computational part of the algorithm is implemented as a wavefront array of computing tasks. These computational arrays are driven by a driver task which coordinates the flow of data into and out of the computing surfaces. If the inter process communications time between tasks is small, then this model shows a potential for speed-up. If this be the case, one may conclude that this model is an appropriate implementation for a linear predictive coding.

The purpose of this thesis is to show the use of digital techniques for Electronic Countermeasures (ECM) signal processing. The main objective is the use of only digital circuitry for the processing of the ECM signals. A recent design of an ECM controller called the Oscillator Waveform Controller (OWC) follows this philosophy. The OWC digitally controls the generation of its nine jamming modes plus the modes generated by the other ECM modules within the ECM system. The use of advance microcircuitry technology allows the OWC the capability of controlling all the parameters within an ECM system. The most desirable feature of the OWC is the use of high level communications for programming ECM mode parameters from an external computer or terminal and digitally storing this parameters upon removal of power.

This thesis deals with the application of Line Spectrum Pairs to
tone detection. Linear Predictive Coding (LPC) is described as a
background to deriving the Line Spectrum Pairs. Two sources of LPC
prediction coefficients are used to calcul?te Line Spectrum Pairs.
One source is the polynomial roots of an LPC inverse filter; various
locations of up to 3 pairs of complex conjugate roots are used to
provide filter coefficients. The radii of the conjugate roots are
varied to see the effect on the calculated Line Spectrum Pairs. A
second source of the filter coefficients is single and multiple
sinusoidal tones that are LPC analyzed by the autocorrelation method
to generate filter prediction coefficients. The frequencies and
amplitudes of the summed sinusoids, and the length of the LPC analysis
window are varied to determine the ability to detect the sinusoids by
calculating the related Line Spectrum Pairs.

The human speech production system is reviewed through
general acoustic theory. Based upon that, the
characteristics of helium speech is compared to normal
speech. The Linear Prediction algorithm is derived
for computer implementation by recursive formulas.
The correction factors for the vocal tract area
functions are found from simulated helium speech and
normal speech data for four vowels. By the correction
factors, new corrected area functions are applied to
the Linear Prediction algorithm so that new synthesis
filters can be built. The output of the algorithm is
enhanced helium speech.

Surveys are made of both character recognition and image
processing. The need to apply image processing techniques
to character recognition is pointed out. The
fields are then combined and tested in sample programs.
Simulations are made of recognition systems with and
without image preprocessing. Processing techniques
applied utilize Walsh-Hadamard transforms and l ocal window
operators. Results indicate that image prepro c ess i ng
improves recognition rates when noise degrades input
images. A system architecture is proposed for a hardware
based video speed image processor operating on local
image windows. The possible implementation of this
processor is outlined.

We characterize the Multitaper Spectral Estimation method as a tool for stationary signal analysis. We compare its performance to the conventional periodogram, the parametric autoregressive and multitaper autoregressive spectral estimates. We analyze single and two frequency sinusoids with additive Gaussian white noise and autoregressive processes of orders 2, 4 and 24. We extend its application to non-stationary signals and develop the multitaper spectrogram. We test the spectrograms with simulated non-stationary autoregressive process of order 2 as the magnitude of its poles vary between 0 and 1 and the angle of the poles vary between 0 and pi. Our results show that the multitaper spectral estimate can be parameterized and is more accurate than others tested for non-sinusoidal signals. We also show applications to aero-acoustic data analysis.

This thesis is concerned with the development of a new face recognition method that has a high recognition performance and is computationally efficient, so that it can be applied to real time processes. A background research is presented, summarizing the most dominant face recognition methods, with an emphasis to the most popular statistical method, the 'Eigenfaces'. Initially, a new algorithm is developed based only on the computational efficiency criterion. It is simulated, and criterions for achieving higher recognition rates are experimentally and theoretically determined. A new space transform is introduced, which enhances the algorithm's recognition capabilities. Its optimum classification measure is mathematically proven to be one that is inherently provided by the new face recognition algorithm. Finally, the developed method is evaluated, and experimentally compared against the 'Eigenfaces' method, using face data.

The need for reliable underwater communication at Florida Atlantic University is critical in transmitting data to and from Autonomous Underwater Vehicles (AUV) and remote sensors. Since a received signal is corrupted with ambient ocean noise, the nature of such noise is investigated. Furthermore, we establish connection between ambient ocean noise and fractal noise. Since the matched filter is designed under the assumption that noise is white, performance degradation of the matched filter due non-white noise is investigated. We show empirical results that the wavelet transform provides an approximate Karhunen-Loeve expansion for 1/f-type noise. Since whitening can improve only broadband signals, a new method for synchronization signal design in wavelet subspaces with increased energy-to-peak amplitude ratio is presented. The wavelet detector with whitening of fractal noise and detection in wavelet subspace is shown. Results show that the wavelet detector improves detectability, however this is below expectation due to differences between fractal noise and ambient ocean noise.